CN108132423B - A Fast Locating Method for Distortion of Power System Monitoring Data Based on State Transition Probability - Google Patents

Abstract

The invention relates to a fast positioning method for monitoring data distortion in a power system based on the state transition probability. The method focuses on the state change probability of the monitoring data, and uses the characteristics of the probability distribution after multiple transfers and the threshold of normal data changes during equipment operation. Compare and export the distortion location matrix to quickly locate the location of the distortion data. The whole process is divided into four steps: 1. Monitoring data attribute entity division, 2. Power equipment monitoring data transfer probability, power equipment multiple monitoring transfer matrix, 4. Distortion data location and subsystem distortion degree measurement. The data acquisition stage of the present invention can unify various data formats or data structures into state transition probabilities, thereby avoiding the influence of different data formats in multi-source heterogeneous data on data analysis, and reducing the complexity of the analysis system.

Description

A Fast Location of Distortion in Power System Monitoring Data Based on State Transition Probability method

technical field

The present invention belongs to the applied research on the fusion of electric power communication data and big data technology. By analyzing the state change probability of the detection data, the format of each data in the electric power communication system is unified into the transition probability distribution of the data state, and then through the Chapman- The Kolmogorov equation performs multiple transfers on the initial transfer distribution to obtain the theoretical distribution of multiple transfers, and calculates the distortion location matrix that can directly locate the distorted data by limiting the change threshold of each data.

Background technique

The power communication system is a broad concept, which generally refers to the subsystems related to the power grid and the data information generated by them. With the continuous development of my country's power grid and the continuous expansion of electricity demand, the data generated in the power communication system is also increasing. At the same time, the speed of data generation is getting faster and faster, and the data structure between different subsystems is also very different. The data generated by the power communication system has become a typical big data.

The power communication system is an important system to ensure the normal operation of the power system. It monitors equipment through various sensors, provides decision-making for equipment failures, and provides a basis for equipment maintenance. Large-scale power communication systems generate massive amounts of monitoring data, and data distortion will inevitably occur during the collection, entry, transmission, exchange and storage of these data. In reality, these distorted data have become an important obstacle to locate and analyze power equipment faults. Improving the data quality of the power communication system is an important part of perfecting the power grid system. Experts at home and abroad have proposed a variety of solutions to the detection of distorted data in power systems. A researcher studies the causes of data distortion in energy management systems, and then starts from the reasons to solve the problem of data distortion. A certain research started from the data platform to try to improve the data quality. A research predicts data quality from the perspective of interpolation fitting. A research based on the common information model (Common Information Model, CIM) high-speed model exchange format CIM/E text as the data verification technology between different systems, using the improved The means of multi-source data screening for better quality data, and the method of feedbacking field data according to the state estimation of the master station improve the overall data quality of the power grid dispatching system.

The above bad data detection based on the state estimation of power equipment has a certain effect in improving the quality of local data in the off-site system, but it still does not have good applicability to the multi-source heterogeneous big data generated by the entire power communication system, and is aimed at each The cost of establishing a corresponding knowledge base for such data distortion is relatively high. The present invention proposes the rapid positioning of power system monitoring data distortion based on the state transition probability, does not pay attention to the multi-source heterogeneous format of the power monitoring data, and instead focuses on the state change of the data, focusing on whether the probability of the monitoring data change is consistent with the actual change of the equipment , the algorithm is verified by the real data set of the State Grid, and the experimental results show that the method is suitable for the rapid location of multi-source heterogeneous data distortion generated by the power communication system in the big data environment.

Contents of the invention

The power communication system is a necessary information transmission system to ensure the normal and stable operation of the power system. It monitors electrical equipment in real time through various sensors, reports abnormal data in time to ensure stable operation of equipment, provides decision support for equipment failures, and provides positioning basis for equipment maintenance. At present, with the increasing scale of the power grid system and the increasing types of monitoring sensors for various power equipment, the power communication system generates a large amount of monitoring data every day. These data are collected, entered, transmitted, exchanged and stored. Data distortion is inevitable. In fact, in the era of big data, such distorted data has become one of the important obstacles to locating and analyzing power equipment faults. Data distortion in power communication systems mainly includes the following two aspects:

1. Violation of monitoring data consistency. Monitoring data consistency refers to whether the data actually recorded by the system satisfies a certain functional dependency or logical relationship, and whether there is data beyond the domain of the attribute definition.

2. Violation of the integrity of the monitoring data. The integrity of the monitoring data refers to whether the data actually entered in the power information system is missing, and whether all the data recorded according to the design requirements are completely recorded.

Aiming at the problem of low data quality in the current power communication system, the present invention aims to establish a random process rapid discrimination method for automatic distortion identification, location and distortion degree measurement of power data. The focus of this method is the status of power data The consistency between the change and the actual equipment running state change. Unify the format of each data in the power communication system into the transition probability distribution of the data state, and then perform multiple transfers on the initial transfer distribution through the Chapman-Kolmogorov equation to obtain the theoretical distribution of multiple transfers. The data change threshold is limited, and the distortion location matrix that can directly locate the distortion data is calculated.

For accomplishing above object, the present invention comprises four steps as a whole, and overall flow chart is shown in accompanying drawing 1

Step 1 Monitor data attribute entity division

Monitoring data collection essentially refers to the process in which various sensors monitor equipment in the power grid system and transmit the monitoring data to a designated location for storage. There are multiple storage mechanisms for monitoring data in different subsystems. The purpose of this step is to divide the collected data into entity data sets according to the entity equipment of the data source. This step is divided into 3 sub-steps:

Step 1.1 Raw monitoring data collection

Definition 1 Raw data set

D＝{d ₁₁ ,d ₁₂ ,...,d ₂₁ ,d ₂₂ ,...,d _n1 ,d _n2 ,...,d _nk }

Among them, D represents the attribute collection of various raw data collected in the system, and d _ij represents the k-th attribute of the i-th device. In actual information collection, because the data collection methods of each subsystem of the power grid are different, so each After the data collected by the subsystem is summarized, it is often a messy data collection.

Step 1.2 Monitoring data is classified by entity source address

In the data collection process, the data is classified according to the source index field. This step is divided into two forms.

Form 1: Classification of collected data

For the collected data in the form of definition 1, it needs to be classified according to the data source index field in the data attribute, and the monitoring data from the same entity should be divided into one group

Form 2: Collect monitoring data by entity

For subsystems that can output monitoring data by entity, directly collect the monitoring data, and indicate the unique attribute of the entity in the data.

Definition 2 device data vector

d _i ＝(d _i1 ,d _i2 ,...,d _ik )

Among them, d _i represents the monitoring data vector from the i-th (1≤i≤n) device, and d _ij represents the j-th (1≤j≤k) attribute of the i-th device. In this way, the data classified according to the data source of any physical device i is recorded by d _i in the form of vector

Step 1.3 Construct attribute expansion matrix

In the power communication system, the power grid contains various subsystems, and all kinds of equipment work together to support the normal operation of a subsystem. The data changes of a set of subsystems can be formalized through the attribute expansion matrix.

Definition 3 attribute expansion matrix

Among them, M(t) represents the attribute expansion matrix at time t, d _ij (t) represents the jth of the i (1≤i≤n) device, and (1≤j≤k _i ≤k) attributes at time t monitoring value. The upper bound k of the number of attributes owned by the equipment is defined in the above formula, where k=max(k ₁ ,k ₂ ,...) indicates the number of attributes of the equipment with the most attributes in the subsystem, and it specifies the columns of the extended matrix number. The number of attributes of different devices in the power subsystem can be different, so most of the row vectors in the extended matrix M(t) have no defined attribute values, and such undefined attributes are called extended attributes, and their role is to maintain the extended The rectangular structure of the matrix is convenient for subsequent mathematical processing. The extended matrix M(t) completely contains the attribute values of the power subsystem at time t.

Step 2. Monitoring data transfer probability of power equipment

This step is divided into 3 sub-steps

Step 2.1 State division of data

The value of each attribute of electrical equipment has a certain range of normal range. When a certain attribute value exceeds its normal range, it is said that the attribute has an abnormal value. For discrete attribute values, its definition domain is countable discrete points. For continuous attribute values, its definition domain is a continuous interval. According to specific electrical equipment, different monitored data values can be divided into different states according to their definition domains. When the equipment is in a normal state, the state of the monitoring data is called steady state. For discrete data values, different points can be classified into one category according to specific data attribute characteristics to form a state. You can also directly treat each point as a state. For continuous data values, the continuous value interval can be divided into segments according to specific characteristics, and each segment is a state.

The function of this step is to describe the data of the device in a discrete state, to monitor the transfer of data in the state to describe the change of the data.

Step 2.2 Status of the device

Electric equipment can operate in different states, and different equipment states represent the stage characteristics of equipment operation. For example, the operating status of a transformer can be divided into normal operation, high temperature operation, and equipment abnormality. Different equipment has different operating states, and the essence of power data inconsistency is the inconsistency between the actual operating state of the equipment and the status of the monitoring data, that is, the monitoring data cannot truly reflect the actual operating status of the equipment, and the operating status of the equipment is varied, while the data There are more types of states. Usually a device state corresponds to a specific combination of a series of data states. In the past, it was complicated to find the inconsistent correspondence, but the technical method adopted by the present invention is not to pay attention to the state of the equipment and data itself, but to locate the inconsistent correspondence by examining the changes of these states.

When the equipment in the power system is running stably, the values of various monitoring data of the equipment should also be stable within a certain range, and its change rules are also stable. When the state of the equipment changes, a part of the monitoring data will deviate from the original state with a greater probability and enter a new state, thus breaking the previous stable rules.

This step establishes a channel to describe the data state transition in a probabilistic manner, that is, the data state transition can be described in the mathematical form of probability distribution.

Step 2.3 Statistical state transition frequency of historical monitoring data

Definition 4 Data State Transition Frequency

Among them, f _ij represents the frequency of data transfer from state i to state j after N _i + N _j times of equipment monitoring, N _i represents the number of times in state i, and N _j represents the number of times in state j.

Law 1 Bernoulli's law of large numbers

Among them, P represents probability, and ε represents a positive number. Bernoulli's law of large numbers shows that the frequency of state transition calculated by collecting a large number of state transition data of power equipment monitoring data will converge according to probability. This law shows that the state transition probability of power equipment can be estimated by monitoring current data multiple times or directly counting historical data.

Step 3 Multi-monitoring transfer matrix of power equipment

This step is divided into 4 sub-steps

Step 3.1 Markov property and time homogeneity of data transition probability

Since each monitoring data is essentially sampled at a discrete time point, and the monitoring data is also divided into discrete states according to the state division method in step 2.2, the state transition of power monitoring data is essentially a time and state transition. Discrete random process. Since each monitoring, the state of the data is only related to the previous state (that is, the previous state is the initial state for probability transfer), so this random process has Markov properties, and its essence is a Markov chain . The result of each monitoring sampling of the data has nothing to do with the state of the first monitoring, which shows that the state transition probability of the data is also time-homogeneous. This step establishes the connection between the state transition probability and the Markov chain, which theoretically shows that the transition probability of equipment monitoring data can construct a transition matrix with Markov properties.

Step 3.2 Multiple transfer matrix of power data

Define 6 single state transition matrix

Among them, E represents a single state transition matrix of electric power data, and its element p _ij represents the probability of data transitioning from state i to state j after a monitoring time interval. Its probability value can be calculated by the method described in step 2.3. The single state transition matrix E fully describes the probability distribution of all state transitions after any power data undergoes one monitoring. According to the Chapman-Kolmogorov equation

The transition matrix of the power data after any number of state transitions can be obtained,

From the Chapman-Kolmogorov equation, we can directly calculate the state transition matrix after any m+n times, E ⁽ⁿ⁺ The element in row i and column j in ^m) represents the probability that the data is monitored in the initial state i, and then after m+n times of monitoring, the data is in state j. The actual meaning of the multiple transition probability obtained by this calculation method refers to the theoretical state transition probability that the data should reach when the power equipment is in a steady state.

Step 3. Theoretical probability distribution after 4 transitions

Definition 7 Initial distribution of power data

Among them, φ(0) represents the initial distribution vector of monitoring data, and the element (0≤j≤n) represents the probability of each state of the data at the initial time 0. According to the Chapman-Kolmogorov equation, the theoretical probability distribution at any m+n time can be obtained.

Theorem 1 Theoretical probability distribution of multiple transition states

φ(m+n)=φ(0)E ^m+n

Among them, φ(m+n) represents the theoretical probability distribution of the calculated power data after m+n transfers

Step 4 Distortion data location and measurement of subsystem distortion degree

The equipment is in a stable state within a certain period of time, which means that the equipment does not change its state within this period of time, and the corresponding state probability distribution of monitoring data reflects the characteristics of the data state probability mode that the equipment should have in this state. Once the power communication system detects that the current state of the power equipment does not conform to the distribution characteristics of the corresponding state transition probability, it can be inferred that the monitoring data may be distorted in the system. During the monitoring period, if the operating state of the equipment does not change, but the corresponding data state changes, this situation is called the first type of distortion. If the operating state of the device changes without a corresponding change in the state of the data, this is called Type 2 distortion. This step is divided into 5 sub-steps

Step 4.1 Deviation degree of power data state transition distribution

Definition 9 Deviation Degree of Power Data State Transition Distribution

δ＞v(m+n)＝||φ′(m+n)-φ(m+n)|| ₂

Among them, v(m+n) represents the deviation degree of the power data state transition distribution, φ′(m+n) is the actual statistical transition probability distribution of the power equipment after m+n times of monitoring, φ(m+n) is the theoretical probability distribution of power equipment after m+n times of monitoring, and the deviation between the two distributions is measured by the 2-norm of the vector. δ>0 is a deviation threshold that can be defined according to the actual situation, and it is determined by the actual inherent state transition characteristics of the equipment in the subsystem.

When δ>v(m+n), it indicates that the deviation degree of the actual state transition distribution of the equipment data is within a reasonable range.

When δ≤v(m+n), it indicates that the deviation degree of the actual transfer distribution of the device data exceeds the threshold.

When the device is running stably, δ≤v(m+n) appears, indicating that the device data has the first type of distortion symptoms, because the data appears in an infrequent state with a high probability at certain moments. When there is a change during the operation of the equipment, δ>v(m+n) appears, indicating that the equipment has the second type of distortion symptoms, because as the operating state of the equipment changes, the monitoring data should undergo corresponding state changes, while the actual state distribution No major changes occurred.

Step 4.2 Localization of distorted data in subsystems

Definition 10 Subsystem monitoring data deviation matrix

The grid subsystem is an organic whole including the corresponding equipment of the system, and the distorted data in the subsystem can be quickly located through the deviation matrix.

Among them, V(t) represents the deviation matrix of the monitoring data of the subsystem after t standard periods, and the element v _ij (t) represents the data status of the jth attribute of the i-th device in the subsystem after t standard periods Skewness of the transfer distribution. According to the properties of the extended matrix in step 2.1, if the attribute does not exist in the device corresponding to the corresponding position of the matrix, the value of the command element is -1.

Step 4.3 performs distortion localization, based on the following formula:

Definition 11 Distortion Localization Matrix

in, Represents an n×k distortion positioning matrix, the elements are only composed of values 0, 1 and -1, F represents the matrix 0-1 transformation function, its function is to perform 0-1 transformation on the elements in the matrix, Δ(t) represents Deviation threshold matrix, its element δ _ij (t) represents the deviation threshold corresponding to the degree of deviation v _ij (t), and the function f(z _ij ∈ V(t)-Δ(t)) represents the 0-1 Transformation, which is defined as follows: define an 11-element 0-1 transformation function

The actual significance of the distortion localization matrix is to mark all elements that exceed the threshold as 1 and those that do not exceed the threshold as 0, and the positions of the elements in the matrix correspond to the positions of the data in the subsystem one by one. Such a distortion localization matrix contains the change state and position information of the distortion data in a subsystem.

Step 4.4 Handling of missing data

Missing data is also a manifestation of data distortion. Regardless of the actual operating status of the equipment, as long as there is a data null value in the monitoring data, it is considered to be data distortion. At this time, the missing data transformation is performed.

Among them, g(d _ij ) represents the missing transformation function, which maps all null elements in M(t) to 1, and enters When the null value data is missing, the transformation function is mapped to When the 1 element in , indicates its distortion property. After a complete distortion location is performed, the original data of this round needs to be used as historical data to update the state transition frequency. In this way, the probability elements of the single state transition matrix constructed in step 3.2 can be guaranteed to have real-time accuracy.

Step 4.5 Data distortion location, and power subsystem distortion degree measurement.

Traverse the distortion location matrix once, and record the location of each element equal to 1 in the matrix corresponding to the location of a distortion data, and its row label represents the device number. The list represents the property number of the device. The ratio of extracted elements to all elements quantitatively reflects the degree of distortion of the power subsystem.

It should be noted that: step 3.1 involves the Markov property proof

Proof: the power data state transition process {X _n ,n=0,1,2,…} is a Markov chain

Since n is finitely listable, and And state i,j,i ₀ ,i ₁ ,…,i _n-1 total existence conditional probability P(X _n+1 =j|X ₀ =i ₀ ,X ₁ =i ₁ ,…,X _n-1 = i _n-1 ,X _n =i) such that P(X _n+1 =j|X ₀ =i ₀ ,X ₁ =i ₁ ,...,X _n-1 =i _n-1 ,X _n =i)= P(X _n+1 ＝j|X _n ＝i)

That is, the sampling process satisfies the Markov property, and {X _n , n=0, 1, 2,...} is a Markov chain.

Therefore, the present invention has the following advantages: 1. The data collection stage of the present invention can unify various data formats or data structures into a state transition probability, thus avoiding the impact of different data formats in multi-source heterogeneous data on data analysis, reducing analyzed the complexity of the system. 2. Data distortion discrimination can be performed dynamically in real time. In theory, as long as the initial distribution of the data state at a certain moment is given, the data distortion at that time can be calculated. At the same time, the distorted data will return to update the historical data. 3. Data distortion discrimination and positioning are simultaneous. When data distortion is determined, the location of the distortion is also determined according to the distortion positioning matrix, which can shorten the repair time of equipment failure.

Description of drawings

Accompanying drawing 1 is the overall flow diagram of the present invention.

Detailed ways

Step 1 Monitor data attribute entity division

Monitoring data collection essentially refers to the process in which various sensors monitor equipment in the power grid system and transmit the monitoring data to a designated location for storage. There are multiple storage mechanisms for monitoring data in different subsystems. The purpose of this step is to divide the collected data into entity data sets according to the entity equipment of the data source. This step is divided into 3 sub-steps

Step 1.1 Raw monitoring data collection

Definition 1 Raw data set

D＝{d ₁₁ ,d ₁₂ ,...,d ₂₁ ,d ₂₂ ,...,d _n1 ,d _n2 ,...,d _nk }

Among them, D represents the attribute collection of various raw data collected in the system, and d _ij represents the k-th attribute of the i-th device. In actual information collection, because the data collection methods of each subsystem of the power grid are different, so each After the data collected by the subsystem is summarized, it is often a messy data collection.

Step 1.2 Monitoring data is classified by entity source address

In the data collection process, the data is classified according to the source index field. This step is divided into two forms. Form 1: Classification of collected data

For the collected data in the form of definition 1, it needs to be classified according to the data source index field in the data attribute, and the monitoring data from the same entity should be divided into one group

Form 2: Collect monitoring data by entity

For subsystems that can output monitoring data by entity, directly collect the monitoring data, and indicate the unique attribute of the entity in the data.

Definition 2 device data vector

d _i ＝(d _i1 ,d _i2 ,...,d _ik )

Among them, d _i represents the monitoring data vector from the i-th (1≤i≤n) device, and d _ij represents the j-th (1≤j≤k) attribute of the i-th device. In this way, the data classified according to the data source of any entity i device is recorded by d _i in the form of vector

Step 1.3 Construct attribute expansion matrix

In the power communication system, the power grid contains various subsystems, and all kinds of equipment work together to support the normal operation of a subsystem. The data changes of a set of subsystems can be formalized through the attribute expansion matrix.

Definition 3 attribute expansion matrix

Among them, M(t) represents the attribute expansion matrix at time t, d _ij (t) represents the jth of the i (1≤i≤n) device, and (1≤j≤k _i ≤k) attributes at time t monitoring value. The upper bound k of the number of attributes owned by the equipment is defined in the above formula, where k=max(k ₁ ,k ₂ ,...) indicates the number of attributes of the equipment with the most attributes in the subsystem, and it specifies the columns of the extended matrix number. The number of attributes of different devices in the power subsystem can be different, so most of the row vectors in the extended matrix M(t) have no defined attribute values, and such undefined attributes are called extended attributes, and their role is to maintain the extended The rectangular structure of the matrix is convenient for subsequent mathematical processing. The extended matrix M(t) completely contains the attribute values of the power subsystem at time t.

Step 2. Monitoring data transfer probability of power equipment

This step is divided into 3 sub-steps

Step 2.1 State division of data

The value of each attribute of electrical equipment has a certain range of normal range. When a certain attribute value exceeds its normal range, it is said that the attribute has an abnormal value. For discrete attribute values, its definition domain is countable discrete points. For continuous attribute values, its definition domain is a continuous interval. According to specific electrical equipment, different monitored data values can be divided into different states according to their definition domains. When the equipment is in a normal state, the state of the monitoring data is called steady state. For discrete data values, different points can be classified into one category according to specific data attribute characteristics to form a state. You can also directly treat each point as a state. For continuous data values, the continuous value interval can be divided into segments according to specific characteristics, and each segment is a state.

The function of this step is to describe the data of the device in a discrete state, to monitor the transfer of data in the state to describe the change of the data.

Step 2.2 Status of the device

Electric equipment can operate in different states, and different equipment states represent the stage characteristics of equipment operation. For example, the operating status of a transformer can be divided into normal operation, high temperature operation, and equipment abnormality. Different equipment has different operating states, and the essence of power data inconsistency is the inconsistency between the actual operating state of the equipment and the status of the monitoring data, that is, the monitoring data cannot truly reflect the actual operating status of the equipment, and the operating status of the equipment is varied, while the data There are more types of states. Usually a device state corresponds to a specific combination of a series of data states. In the past, it was complicated to find the inconsistent correspondence, but the technical method adopted by the present invention is not to pay attention to the state of the equipment and data itself, but to locate the inconsistent correspondence by examining the changes of these states.

When the equipment in the power system is running stably, the values of various monitoring data of the equipment should also be stable within a certain range, and its change rules are also stable. When the state of the equipment changes, a part of the monitoring data will deviate from the original state with a greater probability and enter a new state, thus breaking the previous stable rules.

This step establishes a channel to describe the data state transition in a probabilistic manner, that is, the data state transition can be described in the mathematical form of probability distribution.

Step 2.3 Statistical state transition frequency of historical monitoring data

Definition 4 Data State Transition Frequency

Among them, f _ij represents the frequency of data transfer from state i to state j after N _i + N _j times of equipment monitoring, N _i represents the number of times in state i, and N _j represents the number of times in state j.

Law 1 Bernoulli's law of large numbers

Among them, P represents probability, and ε represents a positive number. Bernoulli's law of large numbers shows that the frequency of state transition calculated by collecting a large number of state transition data of power equipment monitoring data will converge according to probability. This law shows that the state transition probability of power equipment can be estimated by monitoring current data multiple times or directly counting historical data.

Step 3 Multi-monitoring transfer matrix of power equipment

This step is divided into 4 sub-steps

Step 3.1 Markov property and time homogeneity of data transition probability

Since each monitoring data is essentially sampled at a discrete time point, and the monitoring data is also divided into discrete states according to the state division method in step 2.2, the state transition of power monitoring data is essentially a time and state transition. Discrete random process. Since each monitoring, the state of the data is only related to the previous state (that is, the previous state is the initial state for probability transfer), so this random process has Markov properties, and its essence is a Markov chain . The result of each monitoring sampling of the data has nothing to do with the state of the first monitoring, which shows that the state transition probability of the data is also time-homogeneous. This step establishes the connection between the state transition probability and the Markov chain, which theoretically shows that the transition probability of equipment monitoring data can construct a transition matrix with Markov properties.

Step 3.2 Multiple transfer matrix of power data

Define 6 single state transition matrix

Among them, E represents a single state transition matrix of electric power data, and its element p _ij represents the probability of data transitioning from state i to state j after a monitoring time interval. Its probability value can be calculated by the method described in step 2.3. The single state transition matrix E fully describes the probability distribution of all state transitions after any power data undergoes one monitoring. According to the Chapman-Kolmogorov equation

The transition matrix of the power data after any number of state transitions can be obtained,

From the Chapman-Kolmogorov equation, we can directly calculate the state transition matrix after any m+n times, E ⁽ⁿ⁺ The element in row i and column j in ^m) represents the probability that the data is monitored in the initial state i, and then after m+n times of monitoring, the data is in state j. The actual meaning of the multiple transition probability obtained by this calculation method refers to the theoretical state transition probability that the data should reach when the power equipment is in a steady state.

Step 3. Theoretical probability distribution after 4 transitions

Definition 7 Initial distribution of power data

Among them, φ(0) represents the initial distribution vector of monitoring data, and the element (0≤j≤n) represents the probability of each state of the data at the initial time 0. According to the Chapman-Kolmogorov equation, the theoretical probability distribution at any m+n time can be obtained.

Theorem 1 Theoretical probability distribution of multiple transition states

φ(m+n)=φ(0)E ^m+n

Among them, φ(m+n) represents the theoretical probability distribution of the calculated power data after m+n transfers.

Step 4 Distortion data localization

The equipment is in a stable state within a certain period of time, which means that the equipment does not change its state within this period of time, and the corresponding state probability distribution of monitoring data reflects the characteristics of the data state probability mode that the equipment should have in this state. Once the power communication system detects that the current state of the power equipment does not conform to the distribution characteristics of the corresponding state transition probability, it can be inferred that the monitoring data may be distorted in the system. During the monitoring period, if the operating state of the equipment does not change, but the corresponding data state changes, this situation is called the first type of distortion. If the operating state of the device changes without a corresponding change in the state of the data, this is called Type 2 distortion. This step is divided into 5 sub-steps

Step 4.1 Deviation degree of power data state transition distribution

Definition 9 Deviation Degree of Power Data State Transition Distribution

δ＞v(m+n)＝||φ′(m+n)-φ(m+n)|| ₂

Among them, v(m+n) represents the deviation degree of the power data state transition distribution, φ′(m+n) is the actual statistical transition probability distribution of the power equipment after m+n times of monitoring, φ(m+n) is the theoretical probability distribution of power equipment after m+n times of monitoring, and the deviation between the two distributions is measured by the 2-norm of the vector. δ>0 is a deviation threshold that can be defined according to the actual situation. When δ>v(m+n), it means that the deviation degree of the actual state transition distribution of the equipment data is within a reasonable range. When δ≤v(m+n) , indicating that the deviation degree of the actual transfer distribution of the device data exceeds the threshold.

When the device is running stably, δ≤v(m+n) appears, indicating that the device data has the first type of distortion symptoms, because the data appears in an infrequent state with a high probability at certain moments. When there is a change during the operation of the equipment, δ>v(m+n) appears, indicating that the equipment has the second type of distortion symptoms, because as the operating state of the equipment changes, the monitoring data should undergo corresponding state changes, while the actual state distribution No major changes occurred.

Step 4.2 Localization of distorted data in subsystems

Definition 10 Subsystem monitoring data deviation matrix

The grid subsystem is an organic whole including the corresponding equipment of the system, and the distorted data in the subsystem can be quickly located through the deviation matrix.

Among them, V(t) represents the deviation matrix of the monitoring data of the subsystem after t standard periods, and the element v _ij (t) represents the data status of the jth attribute of the i-th device in the subsystem after t standard periods Skewness of the transfer distribution. According to the properties of the extended matrix in step 2.1, if the attribute does not exist in the device corresponding to the corresponding position of the matrix, the value of the command element is -1.

Step 4.3

Definition 11 Distortion Localization Matrix

in, Represents an n×k distortion positioning matrix, the elements are only composed of values 0, 1 and -1, F represents the matrix 0-1 transformation function, its function is to perform 0-1 transformation on the elements in the matrix, Δ(t) represents Deviation threshold matrix, its element δ _ij (t) represents the deviation threshold corresponding to the degree of deviation v _ij (t), and the function f(z _ij ∈ V(t)-Δ(t)) represents the 0-1 Transformation, which is defined as follows: define an 11-element 0-1 transformation function

The actual significance of the distortion localization matrix is to mark all elements that exceed the threshold as 1 and those that do not exceed the threshold as 0, and the positions of the elements in the matrix correspond to the positions of the data in the subsystem one by one. Such a distortion localization matrix contains the change state and position information of the distortion data in a subsystem.

Step 4.4 Handling of missing data

Missing data is also a manifestation of data distortion. Regardless of the actual operating status of the equipment, as long as there is a data null value in the monitoring data, it is considered to be data distortion. At this time, the missing data transformation is performed.

Among them, g(d _ij ) represents the missing transformation function, which maps all null elements in M(t) to 1, and enters When the null value data is missing, the transformation function is mapped to When the 1 element in , indicates its distortion property. After a complete distortion location is performed, the original data of this round needs to be used as historical data to update the state transition frequency. In this way, the probability elements of the single state transition matrix constructed in step 3.2 can be guaranteed to have real-time accuracy.

Step 4.5 Data distortion location, and power subsystem distortion degree measurement.

Traverse the distortion location matrix once, and record the location of each element equal to 1 in the matrix corresponding to the location of a distortion data, and its row label represents the device number. The list represents the property number of the device. The ratio of extracted elements to all elements quantitatively reflects the degree of distortion of the power subsystem.

The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which the present invention belongs can make various modifications or supplements to the described specific embodiments or adopt similar methods to replace them, but they will not deviate from the spirit of the present invention or go beyond the definition of the appended claims range.

Claims (1)

1. A rapid positioning method for monitoring data distortion of an electric power system based on state transition probability is characterized by comprising the following steps:

step 1, attribute entity division of monitoring data, acquisition of original monitoring data of equipment, and division of the acquired original monitoring data into entity data sets according to entity equipment of a data source, specifically comprising:

step 1.1, collecting original monitoring data, and classifying the collected original detection data into a set according to the following definitions:

definition 1, raw data set

D＝{d₁₁,d₁₂,...,d₂₁,d₂₂,...,d_n1,d_n2,...,d_nk}

Wherein D represents the attribute set of various types of raw data collected in the system, D_ijThe j attribute of the ith equipment is represented, and in the actual information acquisition, because the data acquisition modes of all subsystems of the power grid are different, the data acquired by all subsystems are gathered and are often a disordered data set;

step 1.2, classifying the acquired original monitoring data according to the entity source address, classifying the data according to the source index field in the data acquisition process, and selecting the following form for classification according to the type of the data:

classification type one: for classification of collected data, for the collected data in definition 1, it is necessary to classify the collected data according to the data source index field in the data attribute, and classify the monitoring data from the same entity into a group

Classification type two: classifying the collected monitoring data according to the entity, directly collecting the monitoring data of a subsystem which can output the monitoring data according to the entity, and indicating the unique attribute of the entity in the data;

defining 2 device data vectors

d_i＝(d_i1,d_i2,...,d_ik)

Wherein d is_iRepresenting the vector of monitored data from the ith (1. ltoreq. i.ltoreq.n) device, d_ijJ (1 is not less than j not more than k) attribute of the ith device; data sorted by data source of arbitrary entity device i is represented in vector form by d_iRecording

Step 1.3, constructing an attribute expansion matrix: in an electric power communication system, an electric power grid comprises various subsystems, various devices support the normal operation of one subsystem through cooperative work, and data change of one subsystem can be formalized through an attribute expansion matrix, wherein the attribute expansion matrix is defined based on the following definition

Defining 3-Attribute extension matrices

Where M (t) represents the attribute expansion matrix at time t, d_ij(t) represents the j-th (1. ltoreq. i.ltoreq.n) of the i-th plant (1. ltoreq. j.ltoreq.k)_iK) is less than or equal to the monitoring value of the attribute at the moment t; the above equation defines an upper bound k on the number of attributes owned by the device, where k is max (k)₁,k₂,..) represents the number of attributes of the device with the most attributes in the subsystem, which specifies the number of columns of the spreading matrix; the number of attributes of different devices in the power subsystem can be different, so that most row vectors in the expansion matrix M (t) have undefined attribute values, the undefined attributes are called as expansion attributes, and the undefined attributes are used for maintaining the rectangular structure of the expansion matrix so as to facilitate the subsequent mathematical processing; the expansion matrix M (t) completely comprises the attribute value of the power subsystem at the time t;

step 2, obtaining the monitoring data transfer probability of the power equipment, specifically comprising:

step 2.1 State partitioning of data

The values of all the attributes of the power equipment have a normal domain in a certain range, and when a certain attribute value exceeds the normal domain, the attribute is called to have an abnormal value; for a discrete attribute value, its domain is a countable discrete point; for the continuous type attribute value, the definition domain is a continuous interval;

according to specific power equipment, dividing the monitored different data values into different states according to the definition domains of the data values;

when the equipment is in a normal state, the state of monitoring data is called a stable state;

for discrete data values, different points can be classified into one type according to specific data attribute characteristics to form a state, or each point is directly regarded as a state; for continuous data values, or dividing a continuous numerical value interval into segments according to specific characteristics, wherein each segment is in a state;

step 2.2 State of the devices

The power equipment can operate in different states, and the different equipment states represent the stage characteristics of the equipment operation; different devices have different running states, and the essential expression of inconsistent power data is that the running state of the actual device is inconsistent with the state of the monitored data, namely the monitored data cannot truly reflect the actual running condition of the device, the running states of the devices are various, and the types of the data states are more; a device state corresponds to a particular combination of a series of data states;

when equipment in the power system stably runs, various monitoring data values of the equipment should be stabilized in a certain range, and meanwhile, the change rule of the equipment also has stability; when the state of the equipment is changed during operation, a part of monitoring data deviates from the original state with higher probability and enters a new state, so that the previous stable rule is broken;

step 2.3, counting the state transition frequency of the historical monitoring data, and obtaining the state transition frequency based on the following definitions:

defining 4 data State transition frequencies

Wherein f is_ijIndicating monitoring of a device N_i+N_jFrequency, N, of transition of data from state i to state j_iRepresenting the number of times in state i, N_jRepresents the number of times in state j;

the state transition probability of the power equipment is then estimated by monitoring current date data or directly counting historical data a plurality of times based on the following formula

Wherein, P represents probability, epsilon represents a positive number, state transition data of the monitoring data of the power equipment is collected in a large quantity, and the calculated frequency of state transition can be converged according to the probability; the state transition probability of the power equipment can be estimated by monitoring current date data or directly counting historical data for multiple times;

step 3, acquiring a multiple monitoring transfer matrix of the power equipment, specifically comprising:

step 3.1 Markov and chronology of data transition probability

Since each monitoring data is essentially sampling at a discrete time point, and meanwhile, according to the state division method of step 2.2, the monitoring data is also divided into discrete states, the state transition of the power monitoring data is essentially a random process with discrete time and state; because the state of the data is only related to the state of the data in the last time of monitoring, the random process has Markov property and is essentially a Markov chain; the result obtained by sampling each monitoring of the data is irrelevant to the state of the first monitoring, which shows that the state transition probability of the data also has timeliness;

step 3.2, acquiring a multiple transfer matrix of the power data, wherein the multiple transfer matrix is defined based on the following steps:

defining 6 Single State transition matrices

Where E represents a single state transition matrix of power data, whose element p_ijRepresenting the probability of data transitioning from state i to state j after a monitoring interval; the probability value can be obtained by the method in the step 2.3; the single state transition matrix E completely describes the probability distribution condition of all state transitions after one-time monitoring of any power data; according to the Chipman-Kolmogorov equation

Obtaining a transfer matrix of the electric power data after state transfer for any time,

as can be known from the Chipman-Kolmogorov equation, the state transition matrix after any m + n times can be directly calculated by only constructing the single state transition matrix of the power data through the method of the step 3.2, and E^(n+m)The element in the row i and the column j in the middle represents the probability that the data is monitored in an initial state i and then in a state j after m + n times of monitoring; the practical meaning of the multiple transition probability obtained by the calculation method is the theoretical state transition probability which the data should reach when the power equipment is in a stable state;

step 3.4, obtaining the theoretical probability distribution of the secondary transfer, wherein the theoretical probability distribution is defined based on the following:

defining 7 an initial distribution of power data

Wherein the phi (0) table monitors the initial distribution vector, element, of the data(j is more than or equal to 0 and less than or equal to n) represents the probability of each state of the data at the initial time 0; according to the Chepman-Kolmogorov equation, theoretical probability distribution at any m + n moment can be obtained;

theorem 1 theoretical probability distribution of multiple transition states

φ(m+n)＝φ(0)E^m+n

Wherein phi (m + n) represents the theoretical probability distribution of the calculated power data after m + n transfers

Step 4, obtaining the distortion data positioning and subsystem distortion degree measurement, which specifically comprises the following steps:

step 4.1, calculating the deviation degree of the state transition distribution of the power data, wherein the deviation degree is defined on the basis of the following steps:

defining 9 degrees of deviation of power data state transition distributions

δ＞v(m+n)＝||φ′(m+n)-φ(m+n)||₂

Wherein v (m + n) represents the deviation of the state transition distribution of the power data, phi' (m + n) is the actually counted transition probability distribution of the power equipment after m + n times of monitoring, phi (m + n) is the theoretical probability distribution of the power equipment after m + n times of monitoring, and the deviation of the two distributions is measured by the 2-norm of the vector; delta > 0 is a deviation threshold which can be defined according to actual conditions, when delta > v (m + n), the deviation degree of the actual state transition distribution of the equipment data is in a reasonable range, and when delta is less than or equal to v (m + n), the deviation degree of the actual state transition distribution of the equipment data exceeds the threshold;

when the device is operating stably, delta ≦ v (m + n) occurs, indicating that the device data has a first type of distortion symptom because the data is present in an infrequent state with a greater probability at some time; when the equipment changes in the operation process, delta & gt v (m + n) appears, which indicates that the equipment has a second type of distortion symptoms, because the monitoring data should have corresponding state change along with the change of the operation state of the equipment, and the actual state distribution does not have great change;

and 4.2, positioning the distortion data in the subsystem, wherein the positioning is defined on the basis of the following steps:

defining 10 monitor data deviation matrices for subsystems

The power grid subsystem is an organic whole containing corresponding equipment of the system, and distorted data in the subsystem can be quickly positioned through the deviation matrix;

wherein V (t) represents a deviation matrix of the monitoring data of the subsystem after t standard periods, and element v_ij(t) represents the degree of deviation of the data state transition distribution of the jth attribute of the ith device in the subsystem after t standard cycles; according to the property of the extended matrix in the step 2.1, if the attribute does not exist in the equipment corresponding to the corresponding position of the matrix, the value of the command element is-1;

and 4.3, carrying out distortion positioning based on the following formula:

defining 11 a distortion localization matrix

Wherein,representing an nxk distortion localization matrix with elements consisting of only the values 0,1 and-1, F representing a matrix 0-1 transformation function, the function of which is to transform the elements of the matrix 0-1, Δ (t) representing a deviation threshold matrix with elements δ_ij(t) represents a deviation v corresponding to_ij(t) deviation threshold, function f (z)_ije.V (t) - Δ (t)) represents a 0-1 transformation of its independent variables, which is defined as follows

Defining 11 element 0-1 transformation functions

The practical meaning of the distortion positioning matrix is that all elements exceeding the threshold are marked as 1, and elements not exceeding the threshold are marked as 0, and the positions of the elements in the matrix correspond to the positions of the data in the subsystem one by one; the distortion positioning matrix comprises the change state and position information of the distortion data in a subsystem;

step 4.4, processing the missing data, specifically, executing missing data transformation, based on the following formula

Wherein g (d)_ij) Represents a missing transform function that maps all null elements of M (t) to 1 and entersWhen emptyThe value data multiple missing transform function is mapped intoElement 1 indicates its distorted property; after the primary distortion positioning is completely executed, the state transition frequency needs to be updated by taking the original data of the current round as historical data; thus, the probability elements of the single state transition matrix constructed in the step 3.2 can be ensured to have real-time accuracy;

and 4.5, carrying out data distortion positioning, and calculating a power subsystem distortion degree measurement, specifically: traversing the distortion positioning matrix once, recording the positions of the elements which are equal to 1 in the matrix, wherein the positions of the elements correspond to the positions of distortion data one by one, and the row marks represent equipment numbers; the list represents the attribute number of the device; and meanwhile, the proportion of the extracted elements in all the elements quantitatively reflects the distortion degree of the power subsystem.