Transverse federal learning electricity larceny detection method, system and equipment and medium

Abstract

The invention discloses a transverse federal learning electricity stealing detection method, system and device, and a medium. The method comprises the steps of obtaining an electricity stealing detection request; acquiring power utilization data information of a user in a preset time period according to the electricity larceny detection request; the electricity consumption data information of the user in the preset time period is input into a trained electricity stealing detection model to judge whether electricity stealing exists or not, and the electricity stealing detection model is obtained through training of an asynchronous federal learning architecture based on the differential privacy technology. And the security of data privacy is relatively high.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric power detection and relates to a horizontal federated learning electricity theft detection method, system, equipment and medium. Background Art

[0002] Electricity marketing data contains a large amount of data value and a large amount of user sensitive data. Traditional data-driven methods may face privacy threats in user data collection, making it difficult to share data, thus forming data islands. Due to the lack of data, the trained models often have low generalization and cannot adapt to the needs of actual scenarios. At present, for the problem of electricity theft detection in the power system, the electricity theft detection model can better detect electricity theft, but when training the electricity theft detection model, the data needs to be collected uniformly and then trained uniformly, which leads to the problem of data leakage, especially in the scenario of electricity consumption inspection, the security of data privacy is poor. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a horizontal federated learning electricity theft detection method, system, device and medium. The method, system, device and medium can realize electricity theft detection and have high security of data privacy.

[0004] In order to achieve the above object, the present invention adopts the following technical scheme:

[0005] In one aspect, the present invention provides a horizontal federated learning electricity theft detection method, comprising:

[0006] Get electricity theft detection request;

[0007] Acquiring the user's electricity usage data information within a preset time period according to the electricity theft detection request;

[0008] The electricity usage data information of the user within a preset time period is input into a trained electricity theft detection model to determine whether there is electricity theft. The electricity theft detection model is trained using an asynchronous federated learning architecture based on differential privacy technology.

[0009] The horizontal federated learning electricity theft detection method described in the present invention is further improved in that:

[0010] Furthermore, before inputting the electricity usage data information of the user within a preset time period into the trained electricity theft detection model, the method further includes:

[0011] Construct electricity theft detection dataset;

[0012] Build a power theft detection model;

[0013] Based on the electricity theft detection data set, the electricity theft detection model is trained using a federated learning architecture of differential privacy technology to obtain a trained electricity theft detection model.

[0014] Furthermore, the data in the electricity theft detection data set is electricity consumption data of users in low-voltage substations, and the electricity consumption data includes current, voltage, power, user profile data and abnormal electricity consumption event data.

[0015] Furthermore, the federated learning architecture of differential privacy technology is used to train the electricity theft detection model. A for:

[0016]

[0017] in, is the benchmark global model time of the kth client at time t, a, b, c are all constants, β k is the privacy budget weight.

[0018] Furthermore, an electricity theft detection model is constructed based on the multi-head self-attention mechanism and convolutional neural network.

[0019] In a second aspect of the present invention, the present invention provides a horizontal federated learning electricity theft detection system, comprising:

[0020] A first acquisition module, used to acquire an electricity theft detection request;

[0021] A second acquisition module, configured to acquire the user's electricity usage data information within a preset time period according to the electricity theft detection request;

[0022] The detection module is used to input the electricity usage data information of the user within a preset time period into a trained electricity theft detection model to determine whether there is electricity theft. The electricity theft detection model is trained using an asynchronous federated learning architecture based on differential privacy technology.

[0023] The horizontal federated learning electricity theft detection system of the present invention is further improved in that:

[0024] Furthermore, before inputting the electricity usage data information of the user within a preset time period into the trained electricity theft detection model, the method further includes:

[0025] The first building module is used to build a data set for electricity theft detection;

[0026] The second building block is used to build an electricity theft detection model;

[0027] The training module is used to train the electricity theft detection model based on the electricity theft detection data set and utilize the federated learning architecture of differential privacy technology to obtain a trained electricity theft detection model.

[0028] Furthermore, the federated learning architecture of differential privacy technology is used to train the electricity theft detection model. A for:

[0029]

[0030] in, is the benchmark global model time of the kth client at time t, a, b, c are all constants, β k is the privacy budget weight.

[0031] In a third aspect of the present invention, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the horizontal federated learning electricity theft detection method when executing the computer program.

[0032] In a fourth aspect of the present invention, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the horizontal federated learning electricity theft detection method are implemented.

[0033] The present invention has the following beneficial effects:

[0034] During specific operation, the horizontal federated learning electricity theft detection method, system, device and medium described in the present invention input the electricity usage data information of the user within a preset time period into the trained electricity theft detection model to determine whether there is electricity theft. The electricity theft detection model is trained using an asynchronous federated learning architecture based on differential privacy technology. It should be noted that the asynchronous federated learning architecture based on differential privacy technology is used to train the electricity theft detection model in the present invention to avoid problems such as slow convergence speed and poor model performance caused by client heterogeneity. In addition, during training, the data of each client is only trained locally and not uploaded to the aggregation server, so that privacy can be protected to a certain extent and the security of data privacy can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0036] Figure 1 This is the structural diagram of the electricity theft detection model;

[0037] Figure 2 This is the structural diagram of the global model of federated learning;

[0038] Figure 3 This is a comparison diagram of the experimental results of the present invention and the baseline method FedAvg;

[0039] Figure 4 It is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only an embodiment of a part of the present invention, not all embodiments, and is not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concepts disclosed in the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0041] The accompanying drawings show schematic diagrams of structures according to embodiments disclosed in the present invention. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0042] Embodiment 1

[0043] refer to Figure 1 The horizontal federated learning electricity theft detection method of the present invention comprises the following steps:

[0044] 1) Construct an electricity theft detection model;

[0045] Specifically, a power theft detection model is constructed based on the multi-head self-attention mechanism and convolutional neural network, refer to Figure 1 ;

[0046] 2) Constructing a data set for electricity theft detection;

[0047] The data in the electricity theft detection data set comes from the electricity consumption data of users in the low-voltage area. The electricity consumption data includes current, voltage, power, user profile data and abnormal electricity consumption event data, forming a time series of daily electricity consumption of each user. The user types are industrial and commercial users.

[0048] 3) Using the electricity theft detection data set, an asynchronous federated learning architecture based on differential privacy technology is used to train the electricity theft detection model to obtain a trained electricity theft detection model.

[0049] 31) Asynchronous federated learning;

[0050] Specifically, the present invention adopts an asynchronous federated learning framework to adapt to the heterogeneity between clients, where the heterogeneity is mainly reflected in: a) data are not independent and identically distributed; b) computing speeds are different, and computing power has obvious differences; c) communication bandwidths are different, and there may be different and unstable network delays.

[0051] Let C = {1,2,L,C N} represents the client list in the system, |C|=C N , each client k(k∈C) holds local data D k ,|D k |=n k ,In the asynchronous scenario, considering the imbalance of the number of samples on each client, that is make represents the set of local samples of all clients. In the asynchronous federated learning process, the global goal is to use the local data of all clients to jointly train a power theft detection model. Final Solution Problems, among which, Each heterogeneous client needs to solve s k is the data D from client k k A sample randomly selected from .

[0052] During the asynchronous federation training process, the aggregation server sends the initial global model to all clients. The selected clients train local models based on local data and then send them to the aggregation server. In this way, problems such as slow convergence and poor model performance caused by client heterogeneity can be effectively avoided.

[0053] 32) Federated learning architecture based on differential privacy technology;

[0054] The present invention adopts a federated learning architecture based on differential privacy technology to prevent privacy backtracking, specifically:

[0055] The differential privacy mechanism can conceal the slight differences between adjacent data sets. When the output of algorithm M on any adjacent data sets D and D′ satisfies the following conditions, algorithm M is said to provide (ε, δ)-differential privacy:

[0056] Pr[M(D)∈Ω]≤exp(ε)·Pr[M[D']∈Ω]+δ

[0057] Among them, Ω is any subset of the output of algorithm M. The privacy protection mechanism based on differential privacy can confuse the submitted data while ensuring the model training effect, thereby better preventing sensitive data from being stolen by the adversary. Among them, the non-negative parameter ε is the privacy budget, which indicates the degree of privacy protection. The smaller ε is, the higher the degree of privacy protection is. Therefore, the client can set different privacy budgets according to its own privacy preferences.

[0058] 33) Training the electricity theft detection model;

[0059] A method for constructing a power theft detection model in a power system is proposed by combining differential privacy technology with asynchronous federated learning. Specifically, considering the asynchronous federated learning process, the time required to aggregate a local model is defined as a time unit. It is assumed that the model convergence requires a total of T time units. The global model generated at t (t∈[0,T], t∈N) is represented by w t It means that after the aggregation server sends the initial global model w0, it receives the local model w of client k at time t. k , perform aggregation and get:

[0060] w t+1 =(1-α t ) t +α t w k

[0061] Among them, α t It is the weight of the model aggregation at time t, indicating the proportion of the client in the aggregation process. That is, the new global model is calculated by the weighted sum of the old global model and the local model of the new client. Due to the different asynchronous delays of the clients, the privacy budgets set are also different. Therefore, the calculation of this parameter takes two factors into consideration at the same time.

[0062] 331) Differential privacy budget: Different clients set different privacy budgets. The higher the privacy protection level, the higher the degree of data obfuscation, and the priority of its aggregation can be appropriately lowered.

[0063] 332) Asynchronous delay: Different client model updates may experience different degrees of time delay during aggregation. The shorter the delay, the newer the baseline global model used for training, and the higher the priority of aggregation can be appropriately increased.

[0064] The detailed process of federated training is shown in Table 1.

[0065] Table 1

[0066]

[0067] First, the system is initialized. The aggregation server generates the initial global model w0 and sends it down. Each client sets a privacy budget ε according to the sensitivity of the local data. k , and sent to the aggregation server, which calculates the privacy budget weights of all clients as At t∈[0,T], after client k receives the current global model, it first records the current time t and assigns t to Indicates the time of the benchmark global model currently trained locally. This variable is used to calculate the delay of client k's current calculation. Each client must perform E rounds of local iterations, and the global final solution Problems that need to be solved by heterogeneous clients s k is the data D from client k k The present invention does not limit the local optimizer algorithm. Taking stochastic gradient descent as an example, let the learning rate be η, then in the e-th round of local iteration (1≤e≤E):

[0068]

[0069] in, After round E of local iterations, the client uses the privacy budget ε k For the final local model Add Laplace noise:

[0070]

[0071] Among them, s represents local sensitivity. At this time, the local training is completed, and the client k adds the noise to the local model Together with the baseline global model time and privacy budget ε k Sent to the aggregation server.

[0072] After the aggregation server broadcasts the initial global model, it enters the listening state. Every time a client model is received, the model aggregation process is executed and sent down. The aggregation server receives the model at time t. First, we use the baseline global model time and the privacy budget ε k Calculate the aggregation weight α t ←α×f A , α t ∈[0,1], where α is the mixing constant and the aggregation weight function f A for:

[0073]

[0074] Among them, a, b, c are all constants, β k is the privacy budget weight, fA The value of is only related to the privacy budget when the asynchronous delay is less than the threshold b. When it exceeds the threshold, both the asynchronous delay and the privacy budget are considered. The larger the asynchronous delay, that is, The larger the f A The smaller it is, the smaller the weight of the client in aggregation; similarly, the higher the sensitivity of a client's local data, the smaller the privacy budget, and the privacy budget weight β k The smaller the f A The smaller it is, the smaller the weight of the client during aggregation.

[0075] Then, according to the calculated aggregation weights, the aggregation algorithm is executed to generate a new global model: The training process continues to iterate until T , the global training is completed, and the trained electricity theft detection model is obtained.

[0076] 4) obtaining an electricity theft detection request, and obtaining the user's electricity usage data information within a preset time period according to the electricity theft detection request;

[0077] The electricity usage data information of the user within a preset time period is input into the trained electricity theft detection model to determine whether there is electricity theft.

[0078] Confirmatory trials

[0079] The experiment demonstrates the training process of the entire electricity theft detection model, including the process in which each peer node encrypts the local model after local training and sends it to the proxy node, and the proxy node aggregates the model according to the two-stage aggregation algorithm and records it on the blockchain. Assume that the alliance chain system contains 500 peer nodes and 10 proxy nodes. The minimum number of honest and stable proxy nodes that the system can tolerate is 5, and the security parameter is set to 128.

[0080] Assume that the global model of the system initialization is a simple convolutional neural network CNN, and the network structure is as follows Figure 2 As shown, the proportion of clients selected in each round is 20%, the client local iterations are 3 times in each round, and the batch size is 32. In order to simulate the learning effect of the proposed model, a public dataset is used for testing.

[0081] Each participant uses local samples to train the global model issued in each round and submits the model ciphertext to the aggregation server. The aggregation server uses a two-stage aggregation algorithm to aggregate the local models of each client. To demonstrate the advancedness of the proposed method, the baseline algorithm (federal average algorithm) and the present invention are compared to test the accuracy of the global model in 100 rounds. Figure 3 As shown, it can be seen that the present invention can avoid the problem of uneven distribution of client data to a certain extent, thereby improving the accuracy of the global model.

[0082] It should be noted that the present invention combines differential privacy technology with asynchronous federated learning, studies the output confusion technology of federated learning models based on differential privacy, and designs a privacy protection model to prevent malicious adversaries from backtracking user privacy information. Compared with traditional federated learning, it is conducive to improving the confidentiality of privacy information. By adopting a horizontal federated learning framework, the data of each client is only trained locally and not uploaded to the aggregation server, so privacy can be protected to a certain extent. Overcoming the drawbacks of the "data collection" method used in traditional electricity theft detection model training, achieving data privacy security in electricity consumption audit scenarios, and greatly improving the accuracy of user data mining and privacy protection capabilities.

[0083] Embodiment 2

[0084] The horizontal federated learning electricity theft detection system of the present invention includes:

[0085] A first acquisition module, used to acquire an electricity theft detection request;

[0086] A second acquisition module, configured to acquire the user's electricity usage data information within a preset time period according to the electricity theft detection request;

[0087] The detection module is used to input the electricity usage data information of the user within a preset time period into a trained electricity theft detection model to determine whether there is electricity theft. The electricity theft detection model is trained using an asynchronous federated learning architecture based on differential privacy technology.

[0088] As an implementation manner of the present invention, before inputting the electricity usage data information of the user within a preset time period into the trained electricity theft detection model, the method further includes:

[0089] The first building module is used to build a data set for electricity theft detection;

[0090] The second building block is used to build an electricity theft detection model;

[0091] The training module is used to train the electricity theft detection model based on the electricity theft detection data set and utilize the federated learning architecture of differential privacy technology to obtain a trained electricity theft detection model.

[0092] As an implementation mode of the present invention, the federated learning architecture of differential privacy technology is used to train the electricity theft detection model. A for:

[0093]

[0094] in, is the benchmark global model time of the kth client at time t, a, b, c are all constants, β k is the privacy budget weight.

[0095] As an implementation mode of the present invention, the data in the electricity theft detection data set is electricity consumption data of users in low-voltage substations, and the electricity consumption data includes current, voltage, power, user profile data and abnormal electricity consumption event data.

[0096] As an implementation mode of the present invention, an electricity theft detection model is constructed based on a multi-head self-attention mechanism and a convolutional neural network.

[0097] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present application may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0098] Embodiment 3

[0099] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the horizontal federated learning electricity theft detection method when executing the computer program, for example, comprising: constructing an electricity theft detection data set; constructing an electricity theft detection model; based on the electricity theft detection data set, training the electricity theft detection model using a federated learning architecture of differential privacy technology to obtain a trained electricity theft detection model; obtaining an electricity theft detection request; obtaining electricity usage data information of a user within a preset time period according to the electricity theft detection request; inputting the electricity usage data information of the user within the preset time period into the trained electricity theft detection model to determine whether there is electricity theft, wherein the electricity theft detection model is trained using an asynchronous federated learning architecture based on differential privacy technology, wherein the memory may include a memory, such as a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk memory, etc.; the processor, the network interface and the memory are interconnected via an internal bus, and the internal bus may be an industrial standard architecture bus, a peripheral component interconnection standard bus, an extended industrial standard architecture bus, etc., and the bus may be divided into an address bus, a data bus, a control bus, etc. The memory is used to store programs, specifically, the programs may include program codes, and the program codes include computer operation instructions. The memory may include internal memory and non-volatile memory, and provide instructions and data to the processor.

[0100] Embodiment 4

[0101] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the horizontal federated learning electricity theft detection method are implemented, for example, including: constructing an electricity theft detection data set; constructing an electricity theft detection model; based on the electricity theft detection data set, using a federated learning architecture of differential privacy technology, training the electricity theft detection model to obtain a trained electricity theft detection model; obtaining an electricity theft detection request; obtaining the electricity usage data information of the user within a preset time period according to the electricity theft detection request; inputting the electricity usage data information of the user within the preset time period into the trained electricity theft detection model to determine whether there is electricity theft, and the electricity theft detection model is trained using an asynchronous federated learning architecture based on differential privacy technology. Specifically, the computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may include a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include a read-only memory (ROM), a hard disk, a flash memory, an optical disk, a magnetic disk, etc.

[0102] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0103] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0104] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A horizontal federated learning electricity theft detection method, characterized in that: include: Get the electricity theft detection request; Acquiring the user's electricity usage data information within a preset time period according to the electricity theft detection request; The electricity usage data information of the user within a preset time period is input into a trained electricity theft detection model to determine whether there is electricity theft. The electricity theft detection model is trained using an asynchronous federated learning architecture based on differential privacy technology.

2. The horizontal federated learning electricity theft detection method according to claim 1 is characterized in that: Before inputting the electricity usage data information of the user within the preset time period into the trained electricity theft detection model, the method further includes: Construct electricity theft detection dataset; Build a model for electricity theft detection; Based on the electricity theft detection data set, the electricity theft detection model is trained using a federated learning architecture of differential privacy technology to obtain a trained electricity theft detection model.

3. The horizontal federated learning electricity theft detection method according to claim 2 is characterized in that: The data in the electricity theft detection data set are electricity consumption data of users in low-voltage areas, and the electricity consumption data include current, voltage, power, user profile data and abnormal electricity consumption event data.

4. The horizontal federated learning electricity theft detection method according to claim 2 is characterized in that: The aggregate weight function f in the training process of the electricity theft detection model is obtained by using the federated learning architecture of differential privacy technology. A for: in, is the benchmark global model time of the kth client at time t, a, b, c are all constants, β k is the privacy budget weight.

5. The horizontal federated learning electricity theft detection method according to claim 2 is characterized in that: A power theft detection model is constructed based on multi-head self-attention mechanism and convolutional neural network.

6. A horizontal federated learning electricity theft detection system, characterized in that: include: A first acquisition module, used to acquire an electricity theft detection request; A second acquisition module, configured to acquire the user's electricity usage data information within a preset time period according to the electricity theft detection request; The detection module is used to input the electricity usage data information of the user within a preset time period into a trained electricity theft detection model to determine whether there is electricity theft. The electricity theft detection model is trained using an asynchronous federated learning architecture based on differential privacy technology.

7. The horizontal federated learning electricity theft detection system according to claim 6 is characterized in that: Before inputting the electricity usage data information of the user within the preset time period into the trained electricity theft detection model, the method further includes: The first building module is used to build a data set for electricity theft detection; The second building block is used to build an electricity theft detection model; A training module is used to train the electricity theft detection model based on the electricity theft detection data set and utilize a federated learning architecture of differential privacy technology to obtain a trained electricity theft detection model.

8. The horizontal federated learning electricity theft detection system according to claim 7 is characterized in that: The aggregate weight function f in the training process of the electricity theft detection model is obtained by using the federated learning architecture of differential privacy technology. A for: in, is the benchmark global model time of the kth client at time t, a, b, c are all constants, β k is the privacy budget weight.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the horizontal federated learning electricity theft detection method as described in any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the horizontal federated learning electricity theft detection method as described in any one of claims 1 to 5 are implemented.

Images