CN120233659B - Self-adaptive high-precision time service method and system based on deep learning - Google Patents
Self-adaptive high-precision time service method and system based on deep learningInfo
- Publication number
- CN120233659B CN120233659B CN202510707197.0A CN202510707197A CN120233659B CN 120233659 B CN120233659 B CN 120233659B CN 202510707197 A CN202510707197 A CN 202510707197A CN 120233659 B CN120233659 B CN 120233659B
- Authority
- CN
- China
- Prior art keywords
- time service
- low
- signal
- result
- code signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G04—HOROLOGY
- G04R—RADIO-CONTROLLED TIME-PIECES
- G04R20/00—Setting the time according to the time information carried or implied by the radio signal
- G04R20/02—Setting the time according to the time information carried or implied by the radio signal the radio signal being sent by a satellite, e.g. GPS
-
- G—PHYSICS
- G04—HOROLOGY
- G04R—RADIO-CONTROLLED TIME-PIECES
- G04R20/00—Setting the time according to the time information carried or implied by the radio signal
- G04R20/08—Setting the time according to the time information carried or implied by the radio signal the radio signal being broadcast from a long-wave call sign, e.g. DCF77, JJY40, JJY60, MSF60 or WWVB
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Telephonic Communication Services (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a self-adaptive high-precision time service method and system based on deep learning, which are used for monitoring a low-frequency time code signal and a Beidou signal acquired in real time through user equipment and storing key parameters of the low-frequency time code signal, wherein when the low-frequency time code signal and the Beidou signal are normal or the low-frequency time code signal is normal and the Beidou signal is abnormal, the quality of the low-frequency time code signal is scored through a signal quality evaluation model and the key parameters, when the scoring result is not smaller than a preset value, the time service result corresponding to the low-frequency time code signal is compensated based on a time service error predicted by a short-term memory model with attention and the key parameters, the time service is completed through a low-frequency time code time service mode by utilizing the compensated time service result, and when the low-frequency time code signal is normal and the Beidou signal is abnormal and the scoring result is smaller than the preset value, the time service result corresponding to the low-frequency time code signal is compensated through the time service error and a window sliding average value, and the time service is completed through the low-frequency time service code time service mode after compensation.
Description
Technical Field
The invention belongs to the technical field of time frequency, and relates to a self-adaptive high-precision time service method and system based on deep learning.
Background
With the deep development of the modern information society to the intelligent and ubiquitous directions, the high-precision time service technology has become a core technology for supporting key infrastructures such as mobile communication, electric power, finance, industrial Internet, navigation and the like. The Beidou satellite navigation system (BDS) and the low-frequency time code time service system (BPC) are taken as two representative time service methods, and have unique advantages. The Beidou satellite navigation system can stably provide high-precision time service in open space and good weather environment by virtue of the advantages of high precision, global coverage and the like. However, when the indoor environment, bad weather or a severe shielding scene is faced, the signal is weakened or even interrupted, so that the time service precision is greatly reduced or even the time service task cannot be completed. The low-frequency time code timing system has certain stability in a complex environment by utilizing the characteristics of small propagation loss of long-wave signals and the like, and has relatively small influence on weather and strong anti-interference capability compared with a Beidou satellite navigation system. However, the time service precision of the system is still a certain gap compared with that of a Beidou satellite navigation system.
In the related art, high-precision time service is performed through a low-frequency time code time service system or a Beidou satellite navigation system, but the method is difficult to meet the current situation of high-precision time service requirements in complex and changeable environments.
Therefore, how to meet the requirement of the modern society for high-precision time service is a problem to be solved urgently.
Disclosure of Invention
In view of this, the embodiment of the invention provides a self-adaptive high-precision time service method and system based on deep learning, which at least solves the problem that the related technology cannot meet the requirement of modern society on high-precision time service.
According to a first aspect of an embodiment of the present invention, there is provided a deep learning-based adaptive high-precision time service method, including:
monitoring a low-frequency time code signal and a Beidou signal which are acquired in real time through user equipment, and storing key parameters of the low-frequency time code signal in the monitoring process, wherein the key parameters comprise signal field intensity, carrier phase, modulation amplitude and signal-to-noise ratio;
When the monitored low-frequency time code signal and the monitored Beidou signal are normal, or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal, scoring the quality of the low-frequency time code signal based on a signal quality evaluation model and the key parameters to obtain a first scoring result;
When the first grading result is not smaller than a preset value, compensating a time service result corresponding to the low-frequency time code signal based on a long-period memory model with attention and a time service error predicted by the key parameter to obtain a first target time service result, and completing time service to the user equipment through a low-frequency time code time service mode by utilizing the first target time service result;
When the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is smaller than a preset value, compensating the time service result corresponding to the low-frequency time code signal through the time service error and a window sliding average value to obtain a second target time service result, and completing time service to the user equipment through a low-frequency time code time service mode by utilizing the second target time service result.
According to a second aspect of the embodiment of the present invention, there is provided an adaptive high-precision time service system based on deep learning, including:
The system comprises a signal acquisition and monitoring module, a signal processing module and a signal processing module, wherein the signal acquisition and monitoring module is used for monitoring a low-frequency time code signal and a Beidou signal which are acquired in real time through user equipment and storing key parameters of the low-frequency time code signal in the monitoring process, wherein the key parameters comprise signal field intensity, carrier phase, amplitude modulation and signal to noise ratio;
the low-frequency time code signal quality evaluation module is used for scoring the quality of the low-frequency time code signal based on a signal quality evaluation model and the key parameters to obtain a first scoring result when the monitored low-frequency time code signal and the monitored Beidou signal are normal or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal;
The time service error prediction and compensation module is used for compensating the time service result corresponding to the low-frequency time code signal based on the long-period memory model with attention and the time service error predicted by the key parameter when the first scoring result is not smaller than a preset value, so as to obtain a first target time service result;
The time service error prediction and compensation module is further configured to compensate a time service result corresponding to the low-frequency time code signal through the time service error and a window sliding average value when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is smaller than a preset value, so as to obtain a second target time service result;
And the dynamic time service decision and execution module is used for completing time service to the user equipment by using the first target time service result through a low-frequency time code time service mode or completing time service to the user equipment by using the second target time service result through a low-frequency time code time service mode.
According to a third aspect of an embodiment of the present invention, there is provided an electronic device, including a processor, a memory, a communication interface, and a communication bus, where the processor, the memory, and the communication interface complete communication with each other through the communication bus, and the memory is configured to store at least one executable instruction, where the executable instruction causes the processor to perform an operation corresponding to the method according to the first aspect or the second aspect.
According to a fourth aspect of embodiments of the present invention, there is provided a computer storage medium having stored thereon a computer program which, when executed by a processor, implements a method according to the first aspect or the second aspect.
According to the scheme provided by the embodiment of the invention, a low-frequency time code signal and a Beidou signal which are acquired in real time through user equipment are monitored, key parameters of the low-frequency time code signal in the monitoring process are stored, the key parameters comprise signal field intensity, carrier phase, amplitude adjustment and signal to noise ratio, when the monitored low-frequency time code signal and the Beidou signal are normal or the monitored low-frequency time code signal is normal and the Beidou signal is abnormal, the quality of the low-frequency time code signal is scored based on a signal quality evaluation model and the key parameters to obtain a first scoring result, when the first scoring result is not smaller than a preset value, the time service result corresponding to the low-frequency time code signal is compensated based on a time service error which is predicted by the key parameter and is provided with attention, a first target time service result is obtained, when the monitored low-frequency time code signal and the Beidou signal are normal through a low-frequency time code time service mode, when the monitored low-frequency time code signal is normal, the quality of the Beidou signal is scored based on the signal quality evaluation model and the key parameter is not smaller than the preset value, the second time service result is obtained through sliding the low-frequency time service mode, and the low-frequency time service result is obtained through a low-frequency time service window, and the error is obtained through the low-frequency time service window. In the process, by introducing a long-period memory model with attention, a signal quality scoring model and a hierarchical compensation strategy, the method can be used for intelligently switching time service modes by combining the real-time states of BDS signals and BPC signals, and realizing accurate regulation and control of error compensation according to the signal quality of BPC when the BDS signals are unavailable, thereby remarkably improving time service precision and stability in a complex environment.
Drawings
For a clearer description of the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly introduced below, it being obvious that the drawings in the description below are only some embodiments of the present invention, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art, wherein:
FIG. 1 is a schematic flow chart of a deep learning-based adaptive high-precision time service system and method according to an embodiment of the present invention;
FIG. 2 is a frame diagram of a long-short term memory model with attention according to an embodiment of the present invention;
FIG. 3 is a block diagram of a deep learning-based adaptive high-precision time service system according to an embodiment of the present invention;
Fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The following examples are illustrative of the invention and are not intended to limit the scope of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with one another without conflict.
It should be noted that the term "first\second\third" related to the embodiments of the present invention is merely to distinguish similar objects, and does not represent a specific order for the objects, it being understood that the "first\second\third" may interchange a specific order or sequencing, where allowed, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.
It will be understood by those skilled in the art that all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which embodiments of the invention belong unless defined otherwise. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
Fig. 1 is a schematic flow chart of a self-adaptive high-precision time service system and method based on deep learning according to an embodiment of the present invention, where the self-adaptive high-precision time service system and method based on deep learning according to the embodiment of the present invention may be executed by an electronic device, and the electronic device may be a computer, a server, etc.
As shown in fig. 1, the adaptive high-precision time service system and method based on deep learning include:
S101, monitoring a low-frequency time code signal and a Beidou signal which are acquired in real time through user equipment, and storing key parameters of the low-frequency time code signal in the monitoring process, wherein the key parameters comprise signal field intensity, carrier phase, modulation amplitude and signal-to-noise ratio.
In the embodiment of the invention, the user equipment is used for collecting the Beidou signal (BDS signal) and the low-frequency time code signal (BPC signal) in real time, monitoring the collected low-frequency time code signal and Beidou signal, and storing the low-frequency time code signal including the signal field intensity E and the carrier phase in the monitoring processKey parameters of modulation amplitude m and signal-to-noise ratio SNR, and the key parameters are used for measuring time service errors of low-frequency time code signals. The change trend of the signal field intensity can effectively reflect the receiving condition of user equipment in a signal coverage range, the change rate of a carrier phase in a certain period of time can reflect the stability of a BPC transmitting system, the modulation amplitude is embodied as the submergence depth of a pulse width amplitude modulation signal negative pulse and can reflect the correctness of a transmitting signal, the signal-to-noise ratio represents the signal-to-noise power ratio, the definition degree of the signal in a noise environment is embodied, the reliability of a time service signal can be reflected to a certain extent, for example, a high signal-to-noise ratio is beneficial to accurate time service, and a low signal-to-noise ratio can increase the demodulation difficulty.
S102, when the monitored low-frequency time code signal and the Beidou signal are normal, or when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal, scoring the quality of the low-frequency time code signal based on a signal quality evaluation model and key parameters to obtain a first scoring result.
In the embodiment of the invention, when the monitored low-frequency time code signal and the Beidou signal are normal, or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal, a signal quality evaluation model of the BPC signal based on deep learning can be designed, and the quality of the low-frequency time code signal is scored through the signal quality evaluation model and key parameters to obtain a scoring result. The score is used for a compensation strategy after linkage adjustment, the robustness of the time service system is enhanced, the higher the score is, the more reliable the BPC signal is, and when the score is lower than a threshold value, an alarm is triggered or a time service mode is switched.
Further, when the monitored low-frequency time code signal and the monitored Beidou signal are normal, the real error is calculated through the low-frequency time code signal and the Beidou signal and is stored, and the real error is subsequently used for training of a Long Short-Term Memory (LSTM) model with Attention, namely an LSTM-Attention mixed model.
And S103, when the first grading result is not smaller than a preset value, compensating the time service result corresponding to the low-frequency time code signal based on the time service error predicted by the short-term memory model with attention and the key parameter to obtain a first target time service result, and completing time service to the user equipment through the low-frequency time code time service mode by utilizing the first target time service result.
In the embodiment of the invention, the timing error refers to the timing error between the BDS timing result and the BPC timing result. Because BDS time service precision is higher than that of BPC, time service errors between the BPC time service result and the BDS time service result are predicted, and the time service result of the BPC is compensated, so that the BPC can reach time service precision close to that of the BDS. Comparing the first grading result with a preset value, inputting key parameters into a trained LSTM model with attention when the first grading result is not smaller than the preset value, obtaining time service errors of low-frequency time code signals in the transmission process, compensating time service results obtained from the low-frequency time code signals through the time service errors, obtaining a compensated accurate first target time service result, and finally finishing time service to user equipment through the low-frequency time code time service mode by utilizing the first target time service result, so as to ensure that the time service is consistent with the time of a low-frequency time code system.
The LSTM (long short term memory-Attention neural network) is a special circulation time network, has a memory unit and a gating mechanism, can effectively process long-term dependency in time sequence data, and meanwhile, attention mechanism allows the model to be dynamically focused on different input parts, so that the Attention degree of the model to key information is improved. For example, key parameters (such as carrier phase mutation and field strength dip) in certain time periods may have larger influence on time service errors, but the conventional LSTM cannot effectively capture information of the key time points, and the attentive LSTM model can automatically learn weights of different time steps by introducing an Attention mechanism and weight and fuse hidden states of all time steps, so that prediction accuracy is improved. Thus, the focused LSTM model is more suitable for dealing with complex relationships between time-varying characteristics of key parameters of BPC and timing errors.
In the embodiment of the invention, the LSTM model with attention is divided into two stages of training and compensation:
In the training stage, a large amount of data is continuously collected under the condition that the Beidou satellite and the BPC signals are good, and monitored key parameters of the BPC signal quality are obtained through sorting according to time sequence Error of BDS time result T BD (T) and BPC time result T BPC (T)Taking the model as an LSTM model training sample with attention, wherein,As input to the LSTM model with attention,As an ideal output of the LSTM model with attention, the training aims at letting the model grasp the mapping relation between the input and the output, i.e. to construct a functionTo make it approach to satisfy the following relationWhere a refers to a specific time, and is a variable value, such as a first second, a second, or a third second.
;
Wherein, the Is a trained attentive LSTM model.
Wherein, when training the LSTM model with attention, the Mean Square Error (MSE) and the attention weight sparse constraint are adoptedThe sum of the two is taken as a loss function and is marked as L, and the specific formula is as follows:
;
;
;
in the above-mentioned method, the step of, For BPC signal quality data sequences at time tAttention weights corresponding to different features of the model, y i is the true timing error of the ith key parameter,Is the prediction time service error of the ith key parameter, n is the sum of data points,In the form of a mean square error,In order to pay attention to the weight-sparsity constraint,As a loss function.
The compensation stage takes the BPC signal quality characteristics (60 s) in the current time window as the input of the LSTM model with attention, and usesCalculating to obtain the output of the LSTM model with attention as the compensation value of the BPC time service result at the t moment:
;
Wherein the preset value can be 0.5, and the dynamic quality score of the BPC signal is defined in actual useWhen it is a high quality signalAnd (3) for low-quality signals, dynamically adjusting a strategy for compensating BPC time service errors by using the LSTM model with attention.
And S104, when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first grading result is smaller than a preset value, compensating the time service result corresponding to the low-frequency time code signal through the time service error and the window sliding average value to obtain a second target time service result, and completing time service to the user equipment through the low-frequency time code time service mode by utilizing the second target time service result.
In the embodiment of the invention, the sliding time window mechanism can collect input data in preset time and input the collected input data into the LSTM model with attention to obtain time service errors. Window moving average refers to the average of all timing error predictors over a given time window (e.g., 60 seconds). After the time service error and the window sliding average value are obtained, the time service result obtained from the low-frequency time code signal is compensated through the time service error and the window sliding average value, an accurate second target time service result after compensation is obtained, and finally the time service of the user equipment is completed through the low-frequency time code time service mode by utilizing the second target time service result, so that the time of the user equipment is ensured to be consistent with the time of the low-frequency time code system.
It can be understood that in the embodiment of the invention, the low-frequency time code signal and the Beidou signal acquired in real time by the user equipment are monitored, and key parameters of the low-frequency time code signal in the monitoring process are stored; key parameters include signal field strength, carrier phase, modulation amplitude and signal-to-noise ratio; when the first grading result is not smaller than a preset value, compensating a time service result corresponding to the low-frequency time code signal based on a time service error predicted by an LSTM model with attention and a key parameter to obtain a first target time service result, and finishing time service to the user equipment through a low-frequency time service mode by utilizing the first target time service result;
When the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is smaller than a preset value, compensating the time service result corresponding to the low-frequency time code signal through the time service error and the window sliding average value to obtain a second target time service result, and completing time service to the user equipment through a low-frequency time code time service mode by utilizing the second target time service result. In the process, through introducing an LSTM model with attention, a signal quality scoring model and a grading compensation strategy, the method can be used for intelligently switching time service modes by combining the real-time states of BDS signals and BPC signals, and realizing accurate regulation and control of error compensation according to the signal quality of BPC when the BDS signals are unavailable, so that time service precision and stability under a complex environment are remarkably improved.
In the embodiment of the present invention, S101 further includes S105 to S107, and the following steps are described.
S105, when the monitored low-frequency time code signal is abnormal and the Beidou signal is normal, or when the monitored low-frequency time code signal and the Beidou signal are both normal, scoring the quality of the low-frequency time code signal based on a signal quality evaluation model and key parameters, and obtaining a second scoring result.
And S106, finishing time service of the user equipment through the Beidou signal time service mode when the second scoring result is smaller than a preset value.
In some embodiments of the present invention, when the monitored low-frequency time code signal is abnormal and the beidou signal is normal, or when the monitored low-frequency time code signal and the beidou signal are both normal, key parameters are input into the signal quality evaluation model to score the quality of the low-frequency time code signal, so as to obtain a second scoring result. And when the second scoring result is smaller than the preset value, completing time service of the user equipment through the Beidou signal time service mode. And the second scoring result obtained when the monitored low-frequency time code signal and the Beidou signal are normal is equal to the first scoring result.
And S107, when the monitored low-frequency time code signal is abnormal and the Beidou signal is abnormal, continuously sending out alarm information.
In some embodiments of the invention, alarm information is continuously sent when the monitored low-frequency time code signal is abnormal and the Beidou signal is abnormal. The alarm can remind the user that the current data may not be trusted, so that the safety risk caused by incorrect positioning information is avoided, and timely finding and reporting of signal abnormality is helpful for maintenance personnel to take measures rapidly to conduct fault investigation and repair.
In some embodiments of the present invention, the scoring the quality of the low frequency time code signal based on the signal quality evaluation model and the key parameters in S102, and the obtaining of the first scoring result may be implemented in S1021, which is described by the following steps.
S1021, inputting key parameters into a bidirectional gating circulating layer by using an input layer to obtain feature vectors after combination processing, and inputting the feature vectors after combination processing into a full-connection layer to obtain a first scoring result.
In some embodiments of the present invention, the signal quality evaluation model may be a bi-directional gating loop model, where the bi-directional gating loop model includes an input layer, a bi-directional gating loop layer, and a full connection layer. Using input layers to sequence key parametersThe two-way gating loop layer generates a hidden state for each time step in the sequence from the beginning to the end of the sequence, and simultaneously, the two-way gating loop layer also generates a hidden state for each time step from the end to the beginning of the sequence, and the two-way gating loop layer generates two hidden states, one from forward processing and the other from backward processing, for each time step in the sequence. These hidden states are typically combined in one or more ways to form the final combined feature vector. And the full connection layer maps the feature vectors after the combination treatment into single-value scores to obtain a first scoring result.
Specifically, an input layer receives a time series data sequence of a BPC signalThe initial design time window is 30s, the step length is 1s, the Bi-gating cycle layer (Bi-GRU layer) is 1 layer of Bi-directional GRU, 64 neurons in each direction, and the full connection layer is used for mapping the output of the Bi-gating cycle layer into a single value scoreSigmoid is selected as an activation function, and the output range is [0,1]. When training the bidirectional gating cycle model, errors of BDS time service result T BD (T) and BPC time service result T BPC (T) are overcomeClassification is performed. Defining sample dataAt maximum valueIn the time-course of which the first and second contact surfaces,And (b) andIs at minimum valueIn the time-course of which the first and second contact surfaces,The intermediate value then generates a score by linear interpolation.
In some embodiments of the present invention, the compensating the time service result corresponding to the low-frequency time code signal based on the attention-carrying long-period memory model and the time service error predicted by the key parameter in S103 further includes S10 to S13 before the first target time service result is obtained, which is described by the following steps.
And S10, carrying out normalization processing on the key parameters to obtain normalized data, and inputting the normalized data into the three long-short-period memory layers through the input layer to obtain a first hidden feature vector. In some embodiments of the invention, the key parameters are recorded as a sequence of time series dataBecause the value ranges and dimension differences of different parameters are large, the data of each parameter needs to be normalized to the [0,1] interval, and the following formula is used for normalizing key parameters:
;
in the above-mentioned method, the step of, For the data to be normalized,As an original key parameter, the key parameter is,AndThe minimum value and the maximum value of the key parameters are respectively used for improving the training effect and the convergence rate of the model.
The LSTM layer is mainly formed by capturing time sequence dependency relations among input features, and is formed by three layers of LSTM hierarchical linkages according to the invention, namely, 3 layers of stacked LSTM are arranged, the output of the front layer is used as the input of the rear layer, the number of neurons of each layer is increased, the number of neurons of each layer is initially set to be 64, 128 and 256 neurons respectively, deep time sequence features are gradually extracted, residual connection is added after each layer of LSTM, and the gradient vanishing problem is relieved, wherein the specific formula is as follows:
;
Wherein, the Representing the number of layers of the film,Representing the hidden state output of the first layer,Representing the result of the hidden state of the first layer after being processed by the LSTM unit, thereby enhancing the training stability and the feature extraction capability of the deep network, and finally outputting the hidden state H of all time steps to obtain the first hidden feature vector.
S12, inputting the first hidden feature vector into the attention layer to obtain a second hidden feature vector.
And S13, carrying out feature extraction and combination on the second hidden feature vector by using the full connection layer to obtain a third hidden feature vector, and inputting the third hidden feature vector to the output layer to obtain a time service error.
In some embodiments of the present invention, after the first hidden feature vector is input to the attention layer, operations such as linear transformation, attention score calculation, attention distribution calculation, weighted summation and the like are performed, so as to obtain a second hidden feature vector. And inputting the second hidden feature vector into the full-connection layer for feature extraction and combination to obtain a third hidden feature vector, and inputting the third hidden feature vector into the output layer to obtain a time service error.
Wherein the LSTM layer and the attention layer each further comprise a discard layer (Dropout layer), the Dropout layer randomly sets the output of a portion of neurons to zero (i.e. "discards" those neurons), thereby forcing the attention-carrying LSTM model to learn a more robust feature, mainly activated during training of the attention-carrying LSTM model. When the LSTM model with attention is trained, the input of the attention layer is the output of the LSTM layer, the importance score of each feature is obtained by calculating the weighted sum of the input features and the weight matrix, and the relative contribution among the features is adjusted accordingly, then a Dropout layer is added to prevent overfitting, and finally a full connection layer is added to make the model more stableMapping the error predicted value to realize conversion from the characteristic to the predicted value, and outputting the predicted value of the BPC time service error, wherein the specific formula is as follows:
;
;
Wherein, the Representing BPC signal quality data sequence at time tAttention weights corresponding to different features of the model; the output of the LSTM layer at the time t is represented; 、 B is a bias term; For the output of the attention layer at time t, the vector integrates the information of all time steps in the time window, but focuses on the time period with higher weight.
In some embodiments of the present invention, S103 may be implemented by S1031, which is illustrated by the following steps.
S1031, acquiring a first target time service result through the time service error, the time service result and a first target time service result formula.
In some embodiments of the present invention, the timing error and the timing result are substituted into a first target timing result formula to obtain a first target timing result.
The first target timing result formula is as follows:
;
in the above-mentioned method, the step of, For the first target time service result,Is the time service result corresponding to the low-frequency time code signal at the time t,The time error at time t.
In some embodiments of the present invention, S104 may be implemented by S1041, which is illustrated by the following steps.
S1041, obtaining a second target time service result through the time service error, the time service result, the window sliding average value and a second target time service result formula.
In some embodiments of the present invention, the timing error and the window sliding average are substituted into a second target timing result formula to obtain a second target timing result.
The second target timing result formula is as follows:
;
in the above-mentioned method, the step of, For the second target time service result,The average value of the error prediction values in the sliding window corresponding to the time t is obtained.
In the embodiment of the present invention, as shown in fig. 2, fig. 2 is a frame diagram of the LSTM model with attention provided in the embodiment of the present invention. In fig. 2, the attentive LSTM model includes an input layer, an LSTM layer, an attentive layer, and an output layer, the LSTM layer (long-short-term memory) includes an LSTM layer and a first discard layer, and the attentive layer includes an attentive layer, a second discard layer, and a full connection layer. Input layer input signal field intensity E, carrier phaseThe first discarding layer and the second discarding layer are discarding layers, the discarding layers are activated when the LSTM model with attention is trained, namely, when the LSTM model with attention is trained, input data is input to the input layer, the LSTM layer, the first discarding layer, the attention layer, the second discarding layer and the full connection layer are processed, and finally, time service errors are obtained through the output layer. When the LSTM model with attention is applied, input data is input into an input layer, the LSTM layer, the attention layer and the full connection layer are used for processing, and finally time service errors are obtained through an output layer. The LSTM layers are three stacked LSTM layers, namely three LSTM layers which are connected in sequence exist.
In the embodiment of the invention, as shown in fig. 3, fig. 3 is a structural diagram of the adaptive high-precision time service system based on deep learning provided in the embodiment of the invention. In fig. 3, the adaptive high-precision time service system based on deep learning includes:
The system comprises a signal acquisition and monitoring module, a signal processing module and a signal processing module, wherein the signal acquisition and monitoring module is used for monitoring a low-frequency time code signal and a Beidou signal which are acquired in real time through user equipment and storing key parameters of the low-frequency time code signal in the monitoring process, wherein the key parameters comprise signal field intensity, carrier phase, amplitude modulation and signal to noise ratio;
the low-frequency time code signal quality evaluation module is used for scoring the quality of the low-frequency time code signal based on a signal quality evaluation model and the key parameters to obtain a first scoring result when the monitored low-frequency time code signal and the monitored Beidou signal are normal or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal;
The time service error prediction and compensation module is used for compensating the time service result corresponding to the low-frequency time code signal based on the long-period memory model with attention and the time service error predicted by the key parameter when the first scoring result is not smaller than a preset value, so as to obtain a first target time service result;
The time service error prediction and compensation module is further configured to compensate a time service result corresponding to the low-frequency time code signal through the time service error and a window sliding average value when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is smaller than a preset value, so as to obtain a second target time service result;
And the dynamic time service decision and execution module is used for completing time service to the user equipment by using the first target time service result through a low-frequency time code time service mode or completing time service to the user equipment by using the second target time service result through a low-frequency time code time service mode.
The system also comprises a preprocessing module, which is used for carrying out normalization processing on the acquired signals.
Referring to fig. 4, a schematic structural diagram of an electronic device according to an embodiment of the present invention is shown, and the specific embodiment of the present invention is not limited to the specific implementation of the electronic device.
As shown in FIG. 4, the electronic device may include a processor 502, a communication interface (Communications Interface) 504, a memory 506, and a communication bus 508.
Wherein:
Processor 502, communication interface 504, and memory 506 communicate with each other via communication bus 508.
A communication interface 504 for communicating with other electronic devices or servers.
The processor 502 is configured to execute the program 510, and may specifically perform relevant steps in the method embodiments described above.
In particular, program 510 may include program code including computer-operating instructions.
The processor 502 may be a central processing unit CPU, or an Application-specific integrated Circuit ASIC (Application SPECIFIC INTEGRATED Circuit), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in the smart device may be the same type of processor, such as one or more CPUs, or different types of processors, such as one or more CPUs and one or more ASICs.
A memory 506 for storing a program 510. Memory 506 may comprise high-speed RAM memory or may also include non-volatile memory (non-volatile memory), such as at least one disk memory.
The program 510 may be specifically configured to cause the processor 502 to perform operations corresponding to the methods described in the above method embodiments.
The specific implementation of each step in the program 510 may refer to the corresponding steps and corresponding descriptions in the units in the above method embodiments, which are not repeated herein. It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the apparatus and modules described above may refer to corresponding procedure descriptions in the foregoing method embodiments, which are not repeated herein.
It should be noted that, according to implementation requirements, each component/step described in the embodiments of the present invention may be split into more components/steps, or two or more components/steps or part of operations of the components/steps may be combined into new components/steps, so as to achieve the objects of the embodiments of the present invention.
The above-described methods according to embodiments of the present invention may be implemented in hardware, firmware, or as software or computer code storable in a recording medium such as a CD ROM, RAM, floppy disk, hard disk, or magneto-optical disk, or as computer code originally stored in a remote recording medium or a non-transitory machine-readable medium and to be stored in a local recording medium downloaded through a network, so that the methods described herein may be stored on such software processes on a recording medium using a general purpose computer, special purpose processor, or programmable or special purpose hardware such as an ASIC or FPGA. It is understood that a computer, processor, microprocessor controller, or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code that, when accessed and executed by a computer, processor, or hardware, performs the methods described herein. Furthermore, when a general purpose computer accesses code for implementing the methods illustrated herein, execution of the code converts the general purpose computer into a special purpose computer for performing the methods illustrated herein.
Those of ordinary skill in the art will appreciate that the elements and method steps of the examples described in connection with the embodiments disclosed herein can be implemented as electronic hardware, or as a combination of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the embodiments of the present invention.
The above embodiments are only for illustrating the embodiments of the present invention, but not for limiting the embodiments of the present invention, and various changes and modifications may be made by one skilled in the relevant art without departing from the spirit and scope of the embodiments of the present invention, so that all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the scope of the embodiments of the present invention should be defined by the claims.
Claims (6)
1. The self-adaptive high-precision time service method based on deep learning is characterized by comprising the following steps of:
monitoring a low-frequency time code signal and a Beidou signal which are acquired in real time through user equipment, and storing key parameters of the low-frequency time code signal in the monitoring process, wherein the key parameters comprise signal field intensity, carrier phase, modulation amplitude and signal-to-noise ratio;
When the monitored low-frequency time code signal and the monitored Beidou signal are normal, or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal, scoring the quality of the low-frequency time code signal based on a signal quality evaluation model and the key parameters to obtain a first scoring result;
When the first grading result is not smaller than a preset value, compensating a time service result corresponding to the low-frequency time code signal based on a long-period memory model with attention and a time service error predicted by the key parameter to obtain a first target time service result, and completing time service to the user equipment through a low-frequency time code time service mode by utilizing the first target time service result;
When the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is smaller than a preset value, compensating a time service result corresponding to the low-frequency time code signal through the time service error and a window sliding average value to obtain a second target time service result, and completing time service to the user equipment through a low-frequency time code time service mode by utilizing the second target time service result;
When the monitored low-frequency time code signal and the monitored Beidou signal are normal and the first scoring result is smaller than the preset value, finishing time service to the user equipment through a Beidou signal time service mode;
When the monitored low-frequency time code signal is abnormal and the Beidou signal is normal, scoring the quality of the low-frequency time code signal based on the signal quality evaluation model and key parameters to obtain a second scoring result;
When the second scoring result is smaller than the preset value, finishing time service to the user equipment through a Beidou signal time service mode;
and when the monitored low-frequency time code signal and the monitored Beidou signal are abnormal, continuously sending out alarm information.
2. The method of claim 1, wherein the signal quality assessment model is a bi-directional gating loop model, the bi-directional gating loop model comprising an input layer, a bi-directional gating loop layer, a full connection layer;
the scoring the quality of the low-frequency time code signal based on the signal quality evaluation model and the key parameter to obtain a first scoring result, including:
Inputting the key parameters into the bidirectional gating circulating layer by utilizing the input layer to obtain the feature vectors after the merging processing; and inputting the feature vectors after the combination processing into the full connection layer to obtain the first scoring result.
3. The method of claim 1, wherein the attention-based long-term memory model comprises an input layer, three long-term memory layers, an attention layer, a full connection layer and an output layer, wherein the three long-term memory layers are in hierarchical association;
the time service error predicted based on the long-period memory model with attention and the key parameters compensates the time service result corresponding to the low-frequency time code signal, and before the first target time service result is obtained, the method further comprises the steps of:
Normalizing the key parameters to obtain normalized data, and inputting the normalized data into the three long-short-term memory layers through the input layer to obtain a first hidden feature vector;
inputting the first hidden feature vector into the attention layer to obtain a second hidden feature vector;
And carrying out feature extraction and combination on the second hidden feature vector by utilizing the full connection layer to obtain a third hidden feature vector, and inputting the third hidden feature vector to the output layer to obtain the time service error.
4. The method according to claim 1, wherein compensating the timing result corresponding to the low frequency time code signal based on the attention-carrying long-term memory model and the timing error predicted by the key parameter to obtain a first target timing result comprises:
Acquiring a first target time service result through the time service error, the time service result and a first target time service result formula, wherein the first target time service result formula is as follows:
;
in the above-mentioned method, the step of, For the first target time service result,The time service result corresponding to the low-frequency time code signal at the time t,And BPC is the low-frequency time code signal for the time service error at the time t.
5. The method of claim 4, wherein compensating the timing result corresponding to the low frequency time code signal by the timing error and the window sliding average value to obtain a second target timing result comprises:
The second target time service result is obtained through the time service error, the time service result, a window sliding average value and a second target time service result formula, wherein the second target time service result formula is as follows:
;
in the above-mentioned method, the step of, For the second target time service result,And the window sliding average value is the window sliding average value corresponding to the time t, and the window sliding average value is the average value of time service error predicted values in a sliding window.
6. An adaptive high-precision time service system based on deep learning, which is characterized by comprising:
The system comprises a signal acquisition and monitoring module, a signal processing module and a signal processing module, wherein the signal acquisition and monitoring module is used for monitoring a low-frequency time code signal and a Beidou signal which are acquired in real time through user equipment and storing key parameters of the low-frequency time code signal in the monitoring process, wherein the key parameters comprise signal field intensity, carrier phase, amplitude modulation and signal to noise ratio;
the low-frequency time code signal quality evaluation module is used for scoring the quality of the low-frequency time code signal based on a signal quality evaluation model and the key parameters to obtain a first scoring result when the monitored low-frequency time code signal and the monitored Beidou signal are normal or when the monitored low-frequency time code signal is normal and the monitored Beidou signal is abnormal;
The time service error prediction and compensation module is used for compensating the time service result corresponding to the low-frequency time code signal based on the long-period memory model with attention and the time service error predicted by the key parameter when the first scoring result is not smaller than a preset value, so as to obtain a first target time service result;
The time service error prediction and compensation module is further configured to compensate a time service result corresponding to the low-frequency time code signal through the time service error and a window sliding average value when the monitored low-frequency time code signal is normal and the Beidou signal is abnormal and the first scoring result is smaller than a preset value, so as to obtain a second target time service result;
And the dynamic time service decision and execution module is used for completing time service to the user equipment by using the first target time service result through a low-frequency time code time service mode or completing time service to the user equipment by using the second target time service result through a low-frequency time code time service mode.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510707197.0A CN120233659B (en) | 2025-05-29 | 2025-05-29 | Self-adaptive high-precision time service method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510707197.0A CN120233659B (en) | 2025-05-29 | 2025-05-29 | Self-adaptive high-precision time service method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN120233659A CN120233659A (en) | 2025-07-01 |
| CN120233659B true CN120233659B (en) | 2025-08-15 |
Family
ID=96157309
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202510707197.0A Active CN120233659B (en) | 2025-05-29 | 2025-05-29 | Self-adaptive high-precision time service method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN120233659B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109085615A (en) * | 2018-08-01 | 2018-12-25 | 桂林电子科技大学 | A kind of BPC big dipper multi-mode time service method and system |
| CN109709584A (en) * | 2018-12-29 | 2019-05-03 | 南京六九零二科技有限公司 | A kind of GNSS Timing Receiver and time service method |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101398474B (en) * | 2007-09-30 | 2011-12-07 | 郑州威科姆技术开发有限公司 | COMPASS and GPS double system second difference switching method |
| JP2014077695A (en) * | 2012-10-10 | 2014-05-01 | Seiko Epson Corp | Radio controlled timepiece and time correction method for radio controlled timepiece |
| US9897634B2 (en) * | 2014-10-04 | 2018-02-20 | Intelligent Fusion Technology, Inc. | Sliding window energy detection for spectrum sensing under low SNR conditions |
| CN105549380B (en) * | 2016-01-28 | 2017-10-27 | 安徽四创电子股份有限公司 | A kind of high-precision time dissemination system of multimode and its time service method |
| CN112164207B (en) * | 2020-09-29 | 2022-04-12 | 封开低频时码授时台 | A timing signal warning system and method for a low frequency time code monitoring system |
| CN112787705B (en) * | 2020-12-28 | 2021-11-02 | 中国科学院国家授时中心 | Time service system and method |
| CN115474268B (en) * | 2022-08-25 | 2024-02-13 | 武汉大学 | Multi-carrier signal tracking and high-precision positioning method and system based on deep learning |
| CN117824598A (en) * | 2023-12-20 | 2024-04-05 | 中国飞行试验研究院 | Intelligent photoelectric theodolite target guiding, positioning and tracking system and method |
| CN118655378A (en) * | 2024-05-08 | 2024-09-17 | 南京纳升微电子科技有限公司 | A second pulse batch testing algorithm |
| CN118795510B (en) * | 2024-09-13 | 2025-02-14 | 长沙金维集成电路股份有限公司 | A method for cluster timing of Beidou navigation equipment |
-
2025
- 2025-05-29 CN CN202510707197.0A patent/CN120233659B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109085615A (en) * | 2018-08-01 | 2018-12-25 | 桂林电子科技大学 | A kind of BPC big dipper multi-mode time service method and system |
| CN109709584A (en) * | 2018-12-29 | 2019-05-03 | 南京六九零二科技有限公司 | A kind of GNSS Timing Receiver and time service method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN120233659A (en) | 2025-07-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3832553B1 (en) | Method for identifying a micro-energy device on basis of bp neural network | |
| CN114297036B (en) | Data processing method, device, electronic equipment and readable storage medium | |
| US8548036B2 (en) | Noise detection method, noise detection apparatus, simulation method, simulation apparatus, and communication system | |
| CN111340238A (en) | A fault diagnosis method, device, equipment and storage medium for an industrial system | |
| CN116804964A (en) | Data processing method, device, electronic equipment and readable storage medium | |
| WO2023137631A1 (en) | Method and apparatus for data driven control optimization for production quality improvement | |
| CN119310516A (en) | A current transformer fault early warning method | |
| CN112910812A (en) | Modulation mode identification method for deep learning based on space-time feature extraction | |
| WO2021088465A1 (en) | Fast modulation recognition method using multilayer perceptron, and employing data fusion of multiple distribution tests | |
| CN117743969A (en) | Fault prediction method, model training method, device, electronic equipment and medium | |
| CN112651519A (en) | Secondary equipment fault positioning method and system based on deep learning theory | |
| CN118487684A (en) | Electromagnetic spectrum anomaly detection and freeze method based on unsupervised learning | |
| CN112666483A (en) | Improved ARMA lithium battery residual life prediction method | |
| CN120233659B (en) | Self-adaptive high-precision time service method and system based on deep learning | |
| CN116935308A (en) | Car washer safety monitoring system and method based on intelligent identification of car scene AI | |
| CN119064712A (en) | A hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection | |
| CN114372495A (en) | Power Quality Disturbance Classification Method and System Based on Deep Spatial Residual Learning | |
| Zhang et al. | Machine learning based protocol classification in unlicensed 5 GHz bands | |
| CN115685083B (en) | Rao-based distance extended target detection method under interference plus noise background | |
| CN118052825A (en) | Glass insulator surface flaw detection method | |
| CN116938654A (en) | An adaptive modulation recognition network structure and method based on reinforcement learning and ConvLSTM network | |
| CN116304553A (en) | An Interference Type Identification Method Based on Deep Learning in Multi-modulation System | |
| CN118797539B (en) | Intelligent detection early warning method based on multi-level dynamic analysis | |
| CN118890568B (en) | A data correction method and system based on centralized meter reading | |
| CN120011790B (en) | LSTM-BiGRU network-based track disease detection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |