CN106375344B - A kind of smart grid load integrality attack detection method towards cloud storage - Google Patents

Abstract

A cloud storage-oriented smart grid load integrity attack detection method, the purpose is to ensure the safe and stable operation of the power system. The cloud is dynamically saved to the cloud storage; then the data analyst initiates a verification attack; then the cloud storage generates verification evidence; finally, the data analyst conducts the attack verification, and decides to analyze or re-acquire the data in the cloud storage based on the verification result operate. The invention protects system measurement values through encryption technology, reduces redundancy of measurement values, reduces attack detection time, and prevents false data injection attacks. This method uses cloud storage technology to store the collected massive system measurement values in the cloud, making load integrity attack detection applicable to large-scale data, thus ensuring the safe and stable operation of the power system.

Description

A smart grid load integrity attack detection method for cloud storage

technical field

The invention relates to a method for detecting load integrity attacks based on cryptography, and belongs to the technical field of detection.

Background technique

Under the situation of in-depth advancement of the smart grid, the digitization, informatization and intelligence of the power system are developing rapidly. The application of smart substations, smart meters, real-time monitoring systems, on-site mobile maintenance systems, and integrated measurement and control systems has rapidly increased the scale and types of data. Cloud storage technology seamlessly integrates physical storage resources in the cluster into a unified storage system through functions such as cluster applications, network technology, or distributed file systems, so that it can store big data in smart grids. Load data mainly comes from power consumption information collection system, load control system, power quality inspection and other systems, which is the basis for load forecasting and load characteristic analysis, as well as the basis for load adjustment and control on the demand side. For example, the fluctuation of the load in the power system will change the operating frequency and voltage level of the power grid. Load integrity attack is a new attack method for system state estimation. The attacker collects and analyzes the data of the smart meter, injects false data in advance, thereby effectively bypassing the detection and defense of the system, affecting the result of the system state estimation, and then causing the control center to make a wrong system operating state and make an error. Wrong decisions can also mislead the system into an unsafe operating state, causing the load on some transmission lines to exceed its capacity. Therefore, in the smart grid environment, it is of great theoretical and practical significance to analyze the data loopholes in the system and study the corresponding detection and prevention methods.

Many scholars have studied the prevention methods of load data attacks, and achieved a series of research results, such as preventing attacks by protecting a set of measurement values; Measured values; use a greedy algorithm to select a subset of measured values, and identify injected false data attacks by increasing the number of detected measured values; use support vector machine (SVM) method for anomaly detection, etc. However, these methods all have shortcomings of one kind or another, such as causing a certain degree of redundancy in measurement values; they are only suitable for traditional false data injection attacks, and it is difficult to adapt to large-scale data, etc., and cannot guarantee the safe and stable operation of the power system.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a cloud storage-oriented smart grid load integrity attack detection method to ensure the safe and stable operation of the power system in view of the drawbacks of the prior art.

Problem described in the present invention is solved with following technical scheme:

A cloud storage-oriented smart grid load integrity attack detection method, the method first encrypts the grid load data collected by the smart meter by the data collector, and then dynamically saves it to the cloud storage through the cloud; then the data analyst initiates the verification Then, the verification evidence is generated by the cloud storage; finally, the data analyst conducts the attack verification, and decides to analyze or re-acquire the data in the cloud storage according to the verification result.

The above cloud storage-oriented smart grid load integrity attack detection method, the method is carried out according to the following steps:

a. Data collectors preprocess the data collected by smart meters, process various data structures into CIM/XML format, and perform block operations on files:

After collecting the data in the smart instrument, the data collector first processes the data of various structures into the CIM/XML data file format F, and divides the file F into n sub-blocks to obtain the sub-block m _i (1≤i≤n ), and then divide each sub-block into k basic blocks to obtain basic blocks m _i,j (1≤i≤n.1≤j≤k), and build a state table S-Table to record all sub-blocks m _i ;

b. Data collectors perform initialization operations to generate public and private keys and label information for each basic block:

The data collector calls the key generation algorithm to generate the public key pk=(e,N) and the private key sk=d, the public key is made public, and the private key is kept by itself; and each basic block m _i,j (1≤i≤n .1≤j≤k) label information T _i,j :

Where N=p×q, p and q are two large prime numbers randomly generated, and p≠q, Φ(N)=(p-1)(q-1), random number e∈[0,Φ(N )-1], and is relatively prime to Φ(N), d satisfies ed≡1(modΦ(N)), h(·) is a homomorphic hash function, and f(·) is a pseudorandom function;

The data collector associates the sub-block and sub-block label information with the S-Table, uploads the file F and the associated table S-Table to the cloud storage for storage, and shares the S-Table with the data analyst;

c. The data analyst launches a verification attack:

The data analyst randomly selects a coefficient c (1≤c≤n) and sends it to the cloud storage. After receiving c, the cloud storage finds the corresponding cth sub-block, and according to the label information T _{i of the basic block contained in the sub-block, j} calculates the label information T _i of the sub-block:

Data analyst generates random number r∈[1,2 ^k -1] and attack verification request chal={r,c};

d. Cloud storage generates verification evidence R:

Cloud storage computing:

e _r =e ^r mod N

And calculate the output verification proof:

Then send the tag information T _i and verification evidence R of the sub-block to the data analyst;

e. Data analysts conduct attack verification:

The data analyst calculates from the verification evidence R, the public key pk of the file to be detected, the label information T _i of the sub-block _mi and the verification request chal:

N is the parameter in the public key pk, T _i is the label of the i-th data block, and T _t is the same meaning, representing the label of the t-th data block;

Among them, t,1≤t≤c is the number of randomly selected data blocks, and T _t ^r is the label information generated by any data block using a random number r;

The data analyst generates the correct verification information for the file to be detected:

R'＝T ^r mod N

Among them, T ^r is the total label information generated by the extracted data, and compares whether the equation R'=R is established;

Verify that the verification evidence R generated by the cloud storage is equal to the correct verification information R' generated by the data analyst. If R'=R, it means that the measured values in the cloud storage are stored correctly, and the data analyst can safely download the required amount Then analyze the system status; otherwise, it means that the measured value in the cloud storage has been tampered with or deleted, and the data analyst can ask the data collector to re-collect the data.

For the above cloud storage-oriented smart grid load integrity attack detection method, when a data analyst needs to perform M integrity verification requests at the same time, the method of batch verification is adopted:

First, the data analyst sends M verification requests {chal} to the cloud storage; secondly, after receiving the verification requests, the cloud storage generates verification evidence R _m for each verification request, and Finally, the data analyst generates verification information R' _m , and Then verify the integrity of the data according to whether R' and R are equal.

The invention protects system measurement values through encryption technology, reduces redundancy of measurement values, reduces attack detection time, and prevents false data injection attacks. The method uses cloud storage technology to store the collected massive system measurement values in the cloud, making load integrity attack detection applicable to large-scale data, thus ensuring the safe and stable operation of the power system.

Description of drawings

Fig. 1 is a detection model architecture diagram of the present invention;

Fig. 2 is a flowchart of the present invention;

Fig. 3 is the relationship between sub-block size and required time;

Figure 4 is a comparison of the verification time between the algorithm in this paper and the extreme learning machine.

The labels in the figure are respectively represented as: m _i is a sub-block, m _i,j is a basic block, pk is a public key, sk is a private key, T _i,j is the label information of _mi,j , p and q are random Two large prime numbers are generated, h( ) is a homomorphic hash function, f( ) is a pseudo-random function, c is a randomly selected coefficient, T _i is the label information of the sub-block, and r is the random number generated by the cloud storage , R is the verification evidence, chal is the verification request, The label information R' generated by any data block using random number r is the correct verification information of the file to be detected, and T ^r is the total label information generated by the extracted data.

Detailed ways

The present invention will be further described below in conjunction with accompanying drawing.

1. Load integrity attack detection model design

The entity of Load Integrity Attack (LIA) detection model is mainly divided into three parts: data collector; cloud storage; data analyzer. After the acquisition personnel encrypt the measurement values collected in the smart meter, they are dynamically saved to the cloud storage through the cloud, and the power system can verify the integrity of the data in the cloud storage through the data analyst. After the verification is successful, the state estimation is performed on the correctly stored data and information, thereby effectively preventing false data from being injected into the smart meter. The LIA detection model architecture is shown in Figure 1.

LIA detection is mainly divided into the following three steps:

(1) The data collector stores the measurement values collected in the smart meter to the cloud storage, and the cloud processes and stores the measurement values.

(2) Data analysts can verify the load integrity of the measured values in the cloud storage.

(3) When the verification is passed, the data analyst can download the data and analyze the system status; when the verification fails, it proves that the data in the cloud storage is incorrect, and the wrong data can be found or even collected again.

2. Design idea of LIA detection algorithm

The flow chart of Load Integrity Attack-Attack Detection (LIA-AD) is shown in Figure 2. Firstly, the collected data is preprocessed, various data structures are processed into CIM/XML format, and the files are divided into blocks. Secondly, the data collector performs the initialization operation to generate the public key and private key and the label information of each basic block. Then the data analyst initiates a verification attack to verify the integrity of the measurement values stored in the cloud storage. The cloud storage in turn generates verification evidence. Finally, the data analyst conducts attack verification, and decides to analyze or re-acquire the data in the cloud storage based on the verification results.

3.LIA-AD Algorithm Description

(1) Pretreatment

After collecting the data in the smart meter, the data collector invokes the key generation algorithm to generate key parameters, the public key is made public, and the private key is kept by itself. First, process the data of various structures into CIM/XML data file format F, and divide F into n sub-blocks to obtain sub-blocks m _i (1≤i≤n). Then divide each sub-block into k basic blocks to obtain basic blocks m _i,j (1≤i≤n.1≤j≤k). Construct a state table S-Table to record all sub-blocks m _i , as shown in Table 1.

Table 1 Structure of S-Table

Wherein, FB represents the sub-block, BN represents the physical serial number of the sub-block, and SN represents the insertion order of the sub-block. Cloud storage stores keys, tag information collections and S-Tables. The algorithm mainly includes the following processes:

(2) Initialization

The input and output data in the initialization phase are:

Input: {d, e, N};

Output: {pk, sk, T _i,j , T _i }.

Where N=p×q, p and q are two large prime numbers randomly generated, and p≠q. Φ(N)=(p-1)(q-1), the random number e∈[0,Φ(N)-1], and is relatively prime to Φ(N). d satisfies ed≡1(mod Φ(N)). Public key pk=(e,N), private key sk=d, T _i,j is the label of basic block m _i,j (1≤i≤n.1≤j≤k), T _i is sub-block m _i (1≤i≤n) labels.

Data collectors generate public and private keys at this stage, which are used by data analysts to verify and decrypt data stored in cloud storage. And generate label information T _i,j for each basic block m _i,j (1≤i≤n.1≤j≤k):

Associate the sub-block and the label information of the sub-block with the S-Table. Among them, h(·) is a homomorphic hash function, and f(·) is a pseudo-random function.

Finally, the information collector uploads the file F and the association table S-Table to the cloud storage for storage, and shares the S-Table with the data analyst.

(3) Initiate a verification attack

The input and output data of the verification attack stage are:

input: {c};

Output: {chal}.

Among them, c (1≤c≤n) is a random number, and chal is a verification request.

The main purpose of attack verification is for data analysts to publicly verify the data stored in cloud storage, so as to ensure that the stored measurement values have not been tampered with or deleted. The verification process is that the data collector randomly selects block c in the storage block for verification. The security and efficiency of this method will be proved in the next section.

The data analyst randomly inputs a coefficient c (1≤c≤n) and sends it to the cloud storage. After receiving c, the cloud storage finds the corresponding c-th sub-block, and according to the label information T _i,j of the basic block contained in the sub-block Calculate the label information T _i of the sub-block:

The data analyst generates a random number r ∈ [1,2 ^k -1] and an attack verification request chal.

(4) Generate verification evidence

The input and output data for the production verification evidence phase are:

Input: {pk,chal};

Output: {R}.

Among them, pk is the public key, chal is the attack verification request sent by the data analyst, and R is the verification evidence generated by the cloud storage.

In order to verify whether the measured values in the cloud storage are stored correctly, the data analyst challenges the cloud storage for verification. After receiving the challenge from the data analyst, the cloud storage generates challenge information, which is the proof of attack verification, and then sends it to the data analyst for integrity verification.

The attack verification request includes the public key pk and the verification request chal. After Cloud Storage receives the attack verification request, it calculates:

e _r =e ^r mod N, (3)

And calculate the output verification proof:

(5) Attack verification

The input and output data of the attack verification phase are:

Input: {R,pk,chal};

Output: {0or1}.

Among them, R is the verification evidence generated by the cloud storage, pk is the public key, and chal is the verification request. If the output result is 0 or 1, it means that the verification result is successful, and the file is not attacked; otherwise, 1 means that the verification fails, and the file is attacked.

After the data analyst receives the proof of verification, he starts to verify the integrity of the load. The verification process is as follows.

From the verification evidence R, the public key pk of the file to be detected, the label information T _i of the sub-block m _i and the verification request chal, calculate:

The data analyst generates the correct verification request for the file to be detected:

R'＝T ^r mod N (6)

Verify whether the verification evidence R generated by the cloud storage is equal to the correct verification information R' generated by the data analyst. like:

R'=R (7)

If it returns 0, it means that the measured values in the cloud storage are correctly stored, and data analysts can safely download the required measured values, and then analyze the system status. Otherwise, it returns 1, which means that the measured value in the cloud storage has been tampered with or deleted, and the data analyst can ask the data collector to re-collect.

(6) Batch verification

When there are data analysts who need to perform M integrity verification requests at the same time, we adopt the method of batch verification. First, M verification requests chal are sent to the cloud storage. Secondly, after cloud storage receives the verification request, it generates verification evidence R _m for each verification request, and Finally, the data analyst generates verification information R' _m , and Verify the integrity of the data again.

4. Correctness analysis

The load integrity attack verification method in this paper is mainly to verify the measurement value of the protection system. The system measurement value is dynamically stored in the cloud storage to prevent the attacker from obtaining, tampering or deleting. In order to correctly analyze the state of the system measurement value, the data analyst first performs integrity verification, and then downloads the relevant data after the verification is passed. Perform status analysis. To prove the correctness of this method, the proof results are as follows.

To verify equation (7), the calculation steps are:

As shown above, the equation R'=R is obtained, which proves that the method in this paper is correct. If there is an inequality R'≠R, it proves that the system measurement values stored in the cloud storage have been tampered with or deleted by the attacker, and the data needs to be re-collected for system state analysis.

5. Security Analysis

Smart grid information exchange generally transmits measurement data to the control center, and wrong measurement data will interfere with the control center to make unreasonable state estimates. In conformity integrity attacks, there are two important problems: 1) The attacker tampers or deletes the system measurement values stored in the cloud storage, which affects the system state estimation, so as to achieve the attack purpose. 2) The attacker tries to inject the pre-specified data into the data collected by the smart meter by obtaining and analyzing the measured value of the system, that is, it is safely stored in the cloud storage, and will not be discovered by the data collectors and data analysts. Affects system state estimation. In order to prove the security of this scheme, this paper proves the security of the above two problems.

In order to prove that the system measurement value in the cloud storage has not been tampered or deleted by the attacker, it proves that the integrity verification scheme in this paper is correct and safe.

Theorem 1 Under the assumption that the factorization of large integers is difficult, the LIA-AD algorithm is safe.

Proof: If an attacker conducts an attack, the steps are as follows:

(1) Key generation. The data collector generates a public key and a private key, and makes the public key public.

(2) Label generation. The attacker sends the sub-block m _i (1≤i≤n) to the data collector, and the data collector signs the basic block in the sub-block m _i , obtains the label T _i , and sends the information to the attacker.

(3) Generate verification evidence. The data analyst generates a verification request chal and sends it to the attacker, and the attacker generates a detection information R according to the verification request chal and the sub-block m _i (1≤i≤n) and its corresponding label information T _i , and sends it to the data analyst.

(4) Attack verification. The data analyst calculates the verification information according to the data integrity evidence {m _i ', r', e'} returned by the attacker, and obtains

Among them, m _i 'is the sub-block generated by the attacker, and r' and e' are two random numbers generated by the attacker.

If the attacker wants to attack successfully, he must make m _i '=m _i , r'=r, e'=e, then R'=R is established. When an attacker tampers or deletes a part of sub-block m _i (1≤i≤n), if he wants to pass the integrity verification, the attacker must forge a suitable m _i '=m _i ,r'=r,e'= e makes R'=R established, that is, the attacker has the ability to forge two large random prime numbers p and q, satisfying p≠q, and g ^p =g ^q modN, where g is a generator of the integer set Z _N ^* . so, Established, so that pq can be used to decompose a large integer N.

Therefore, under the assumption of the difficulty of factoring large integers, if an attacker wants to pass integrity verification by tampering or forging data, the attacker must correctly hold all system measurement values and all label information. That is, it is proved that under the assumption of the difficulty of factoring large integers, the system measurement values stored in the cloud storage cannot be tampered or deleted by attackers.

In problem 2, the attacker tries to inject the pre-analyzed data into the smart meter, which is safely stored in the cloud storage, the attacker must have the privileges of the data collector, such as the private key and key. The proof process is as follows.

Theorem 2 Under the assumption of the large integer factorization difficulty problem, it is difficult to save the injected false data to cloud storage.

Proof: LIA is a new type of false data injection attack. Attackers inject pre-formulated false data into smart meters by collecting and analyzing information from a large number of smart meters. But for an attacker to succeed, they must successfully save the false data to cloud storage. Initialize first. The attacker generates a public-private key pair (pk', sk') by the key generation algorithm, randomly generates two large prime numbers p' and q', and p'≠q', calculates N'=p'×q' and Randomly generate random number e'(e'∈Z _N ), make gcd(e',Φ(N'))=1. Then the public key pk'=(N', e'), and the private key sk'=(p', q').

However, the public key is public, that is, pk=pk', N=N', e=e', if an attacker wants to obtain the private key sk, he must calculate N'=N=p×q and decompose it. Therefore, under the assumption of the difficulty of factoring large integers, it is difficult for the attacker to obtain the private key, that is, it is difficult for the attacker to save the injected false data to the cloud storage.

In order to verify the feasibility and effectiveness of the designed LIA-AD algorithm, related experiments were carried out. Using Matlab2010b environment to generate IEEE118 node standard test samples, the hardware configuration is CPU Intel core 2duo 3.4GHz, 1.5G RAM. Simulate components such as physical nodes, sensors, controllers, and control centers in a computer cluster. A model computing center is set up at the bottom layer of the physical node network, and the measurement data of standard nodes is generated with the help of the Matpower toolkit, including 900 pieces of IEEE118-bus node system measurement data and 900 pieces of false measurement data. The optimal power path algorithm is used to simulate the state of the power grid in real time to make it conform to the law of the power grid, and a state estimation mechanism is established in the control center to fully and truly reflect the state of the power system.

6. Experiment and result analysis

The experimental environment of this paper is to build a Hadoop cloud platform with 4 nodes in the laboratory. The machine configuration of each node is Intel(R) Core(TM) i5-24004-core CPU@2.60GHz, 4GB RAM, and the network bandwidth is 100Mbit/s. Hadoop version is 0.20.2.

First, we processed the generated 900 pieces of measurement data into CIM/XML file format, encrypted the parsed document and saved it in HBase, and verified the integrity of the load.

(1) Experiment 1 Safety Experiment

Block and sign the three system measurement values (A, B, C) and store them, delete 10% of the data for A, modify 10% of the data for B, and do nothing for C. Finally, integrity checks are performed on the three system measurements. The test results are shown in Table 2.

Table 2 Test results

Through experiments, it is found that after the file is tampered with or deleted, it cannot pass the integrity test of the LIA-AD algorithm, and only the unmodified file can pass the integrity verification. It is proved that the scheme in this paper is safe.

(2) Experiment 2 time overhead experiment

For the same system measurement value, choose different file sub-block sizes, and find the time overhead in integrity testing. Assume that there are 900 pieces of measurement data in file A. After parsing the CIM/XML data file, it will be about 10MB. When there is a false data attack, find the average time overhead of the LIA-AD algorithm. Select 2, 4, 8, 16, 32, 64, and 128KB in order for the size of the sub-blocks, select 460 blocks each time, and record the time and communication overhead each time. The experimental results are shown in Figure 3.

The experimental results show that in the LIA-AD algorithm, the size of the file sub-block has almost no effect on the overhead of the verification and detection stages, and the calculation time is basically stable. The average time for the verification integrity stage is 1.61s, and the average verification information generation stage is 0.64s. The request stage takes an average of 0.14s, and the calculation time takes 2.39s overall.

(3) Comparison experiment between Experiment 3 and other methods

The verification time of the LIA-AD algorithm is compared with the time required to detect the activation function of neurons in several basic extreme learning machines, and the comparison results are shown in Figure 4. It can be seen from Figure 4 that the average training time, average detection time and total time required for different activation functions are different. However, the total time required by the LIA-AD algorithm in this paper is the least.

The extreme learning machine is the basic method for load integrity attack verification. Through the activation function of the neurons in the extreme learning machine classifier, the data is reduced to 5 dimensions, and then 100 trials are performed to obtain the average detection time. Since the extreme learning machine needs to train the sample data first, and then perform the load integrity attack detection, the LIA-AD algorithm proposed in this paper directly performs the load integrity attack verification after encrypting the data, which greatly saves the attack verification. time.

Claims (2)

1. A cloud-storage-oriented smart grid load integrity attack detection method is characterized in that, the method first encrypts the power grid load data collected by the smart meter by the data collector, and then dynamically saves it to the cloud storage through the cloud; then The verification attack is initiated by the data analyst; then the verification evidence is generated by the cloud storage; finally, the data analyst conducts the attack verification, and decides to analyze or re-acquire the data in the cloud storage according to the verification result; The method is carried out as follows: a. Data collectors preprocess the data collected by smart meters, process various data structures into CIM/XML format, and perform block operations on files: After collecting the data in the smart instrument, the data collector first processes the data of various structures into the CIM/XML data file format F, and divides F into n sub-blocks to obtain the sub-block m _i (1≤i≤n) , and then divide each sub-block into k basic blocks to obtain basic blocks m _i,j (1≤i≤n.1≤j≤k), and build a state table S-Table to record all sub-blocks m _i ; b. Data collectors perform initialization operations to generate public and private keys and label information for each basic block: The data collector calls the key generation algorithm to generate the public key pk=(e,N) and the private key sk=d, the public key is made public, and the private key is kept by itself; and each basic block m _i,j (1≤i≤n .1≤j≤k) label information T _i,j : Where N=p×q, p and q are two large prime numbers randomly generated, and p≠q, Φ(N)=(p-1)(q-1), random number e∈[0,Φ(N )-1], and is relatively prime to Φ(N), d satisfies ed≡1(modΦ(N)), h(·) is a homomorphic hash function, and f(·) is a pseudorandom function; The data collector associates the sub-block and sub-block label information with the S-Table, uploads the file F and the associated table S-Table to the cloud storage for storage, and shares the S-Table with the data analyst; c. The data analyst launches a verification attack: The data analyst randomly selects a coefficient c (1≤c≤n) and sends it to the cloud storage. After receiving c, the cloud storage finds the corresponding cth sub-block, and according to the label information T _{i of the basic block contained in the sub-block, j} calculates the label information T _i of the sub-block: Data analyst generates random number r∈[1,2 ^k -1] and attack verification request chal={r,c}; d. Cloud storage generates verification evidence R: Cloud storage computing: e _r =e ^r mod N And calculate the output verification proof: Then send the tag information T _i and verification evidence R of the sub-block to the data analyst; e. Data analysts conduct attack verification: The data analyst calculates from the verification evidence R, the public key pk of the file to be detected, the label information T _i of the sub-block _mi and the verification request chal: N is the parameter in the public key pk, T _i is the label of the i-th data block, and T _t is the same meaning, representing the label of the t-th data block; Among them, t,1≤t≤c is the number of randomly selected data blocks, and T _t ^r is the label information generated by any data block using a random number r; The data analyst generates the correct verification information for the file to be detected: R'＝T ^r mod N Among them, T ^r is the total label information generated by the extracted data; whether the comparison equation R'=R is established; Verify that the verification evidence R generated by the cloud storage is equal to the correct verification information R' generated by the data analyst. If R'=R, it means that the measured values in the cloud storage are stored correctly, and the data analyst can safely download the required amount Then analyze the system status; otherwise, it means that the measured value in the cloud storage has been tampered with or deleted, and the data analyst can ask the data collector to re-collect the data. 2. a kind of cloud storage-oriented smart grid load integrity attack detection method according to claim 1 is characterized in that, when there are data analysts who need to carry out M integrity verification requests simultaneously, then adopt the method of batch verification : First, the data analyst sends M verification requests {chal} to the cloud storage; secondly, after receiving the verification requests, the cloud storage generates verification evidence R _m for each verification request, and Finally, the data analyst generates verification information R' _m , and Then verify the integrity of the data according to whether R' and R are equal.