Power load data fabrication detection method and apparatus, and storage medium and related device

Abstract

A power load data fabrication detection method and apparatus, and a storage medium and a related device. When power load data is subjected to detection, the power load data can firstly be converted into an image to be subjected to detection, and a target data detection model can be determined, wherein the model comprises a multi-target prediction sub-model and a multi-target optimization sub-model, the multi-target prediction sub-model being used for predicting a fabrication probability corresponding to each fabrication type among a plurality of fabrication types, and the multi-target optimization sub-model being used for performing probability average maximization on the fabrication probabilities. Thus, after said image is input into the target data detection model, a predicted fabrication result output by the target data detection model can be obtained. The target data detection model constructed in the present application can use features on a temporal level for superposition during detection, so as to improve the adaptability to power scenarios, and can use a predicted fabrication result obtained by means of prediction to perform probability analysis, so as to make the probability analysis more scientific, thereby improving the practical application value.

Description

Methods, devices, storage media and related equipment for detecting power load data forgery

[0001] This application claims priority to Chinese Patent Application No. 202410699182.X, filed on May 31, 2024, entitled “Method, Apparatus, Storage Medium and Related Equipment for Detecting Forged Electricity Load Data”, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of power system technology, and in particular to a method, apparatus, storage medium and related equipment for detecting power load data forgery. Background Technology

[0003] With the deepening of digital grid development, power load data has become a core resource for tasks such as grid situational awareness, status assessment, and control decision-making. The power grid has an increasing need for datasets that can reflect the overall picture of the grid. However, since the power grid has been in a stable operating state for a long time, most power data samples reflect stable scenarios and are insufficient to support forward-looking research such as event prediction, fault early warning, and contingency plans for low-probability events. Therefore, forged data generated by power load data forgery technology is indispensable in the power system.

[0004] However, the massive generation of forged data is uncontrollable, introducing data with unexpected characteristics. Furthermore, data labels often lack manual annotation and traceability, negatively impacting the power system. Therefore, the power system needs to develop forgery detection technology to verify the authenticity of power load data. Currently, forgery detection technology requires building forgery detection models that are customized for training based on the type of forgery. This results in low adaptability of the method in power scenarios characterized by high randomness, high complexity, and multiple influencing factors, affecting its practical application value.

[0005] Summary of the Invention

[0006] The purpose of this application is to at least address one of the aforementioned technical deficiencies, particularly the low adaptability of existing forgery detection technologies to power scenarios characterized by high randomness, high complexity, and multiple influencing factors, which increases the cost of building forgery detection models.

[0007] This application provides a method for detecting forged power load data, the method comprising:

[0008] Acquire the power load data to be detected, encode the power load data, and generate the image to be detected;

[0009] A target data detection model is determined, which consists of a multi-objective prediction sub-model and a multi-objective optimization sub-model. The multi-objective prediction sub-model is used to predict the forgery probability corresponding to each forgery type among multiple forgery types, and the multi-objective optimization sub-model is used to maximize the average probability of each forgery probability.

[0010] The image to be detected is input into the target data detection model to obtain the predicted forgery result corresponding to the power load data output by the target data detection model.

[0011] Optionally, the step of encoding the power load data to generate the image to be detected includes:

[0012] Determine multiple timestamps in the power load data according to time sequence, and obtain the load value corresponding to each timestamp;

[0013] Each load value is encoded using a preset encoding method to obtain the encoding result, and the encoding result is visualized using a data visualization tool to generate the image to be detected.

[0014] Optionally, the determination of the target data detection model includes:

[0015] Acquire sample load data, which includes multiple images to be detected with different forgery types and the real forgery result corresponding to each image to be detected;

[0016] The sample load data is input into a preset initial data detection model to obtain the predicted forgery result output by the initial data detection model;

[0017] The initial data detection model is trained with the goal of making the predicted forgery result approximate the actual forgery result of the sample load data.

[0018] When the initial data detection model meets the preset training conditions, the trained initial data detection model is used as the target data detection model.

[0019] Optionally, obtaining sample load data includes:

[0020] Acquire pre-collected real load data and convert the real load data into a real load image;

[0021] Multiple load data forgery models of various forgery types are identified, and the real load image is input into each load data forgery model to obtain the forged load image output by each load data forgery model.

[0022] The real load image and each fake load image are used as images to be detected, and the real fake result of each image to be detected is marked to form sample load data.

[0023] Optionally, the target data detection model further includes a authenticity discrimination network;

[0024] The step of inputting the image to be detected into the target data detection model to obtain the predicted forgery result corresponding to the power load data output by the target data detection model includes:

[0025] The image to be detected is input into the multi-objective prediction sub-model to predict the probability value of the image to be detected belonging to each forgery type, and the prediction probability result output by the multi-objective prediction sub-model is obtained.

[0026] The predicted probability result is obtained by maximizing the average probability value using the multi-objective optimization sub-model.

[0027] The authenticity of the final probability result is determined by the authenticity discrimination network, and a prediction forgery result corresponding to the power load data is generated.

[0028] Optionally, the multi-objective prediction sub-model includes a data format transformation network, a feature decomposition network, and a feature-to-temporal-series network;

[0029] The step of inputting the image to be detected into the multi-object prediction sub-model to predict the probability value of the image to be detected belonging to each forgery type, and obtaining the prediction probability result output by the multi-object prediction sub-model, includes:

[0030] The data format conversion network is used to forge data on the image to be detected, and forged data corresponding to each forgery type of the image to be detected is obtained.

[0031] The feature decomposition network is used to perform feature decomposition on the image to be detected and each forged data to obtain image feature data corresponding to the image to be detected and each forged data. The image feature data is then converted into time-series feature data through the data format conversion network.

[0032] The feature-to-temporal network is used to cross-stack various temporal feature data to obtain multiple temporal stacked data. Based on each temporal stacked data, the probability value of the image to be detected belonging to each forgery type is determined, forming the prediction probability result output by the multi-target prediction sub-model.

[0033] Optionally, the time-series feature data includes power load features, basic model features, superimposed model features, and custom model features;

[0034] The feature-to-temporal network is used to cross-stack various temporal feature data to obtain multiple time-stacked data, including:

[0035] Based on the power load characteristics of the image to be detected, the basic model features, superimposed model features and custom model features in each time series feature data are cross-superimposed to obtain the first time series superimposed data;

[0036] Based on the basic model features of the image to be detected, the power load features, superimposed model features and custom model features in each time series feature data are cross-superimposed to obtain the second time series superimposed data.

[0037] Based on the superposition model features of the image to be detected, the power load features, basic model features and custom model features in each time series feature data are cross-superimposed to obtain the third time series superimposed data;

[0038] Based on the custom model features of the image to be detected, the power load features, basic model features and superimposed model features in each time series feature data are cross-superimposed to obtain the fourth time series superimposed data.

[0039] This application also provides a device for detecting power load data forgery, comprising:

[0040] The information encoding module is used to acquire the power load data to be detected, and to encode the power load data to generate the image to be detected;

[0041] The model determination module is used to determine the target data detection model, which is composed of a multi-objective prediction sub-model and a multi-objective optimization sub-model. The multi-objective prediction sub-model is used to predict the forgery probability corresponding to each forgery type among multiple forgery types, and the multi-objective optimization sub-model is used to maximize the probability average of each forgery probability.

[0042] The model prediction module is used to input the image to be detected into the target data detection model to obtain the predicted forgery result corresponding to the power load data output by the target data detection model.

[0043] This application also provides a storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the power load data forgery detection method as described in any of the above embodiments.

[0044] This application also provides a computer device, including: one or more processors, and memory;

[0045] The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the power load data forgery detection method as described in any of the above embodiments.

[0046] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0047] The power load data forgery detection method, apparatus, storage medium, and related equipment provided in this application allow users to first acquire the power load data to be detected when they need to verify its authenticity. This data is then encoded to generate an image to be detected, enabling better capture of the temporal characteristics of the power load data and improving detection accuracy. Next, a target data detection model is determined, consisting of a multi-objective prediction sub-model and a multi-objective optimization sub-model. The multi-objective prediction sub-model predicts the forgery probability for each of the multiple forgery types, while the multi-objective optimization sub-model maximizes the average probability of each forgery probability. Therefore, by inputting the image to be detected into the target data detection model, the predicted forgery result corresponding to the power load data can be obtained from the model's output. This application, through the target data detection model, improves the adaptability to power scenarios by overlaying features at the temporal level during the detection process and utilizes the predicted forgery result for probabilistic analysis, making it more scientific and thus enhancing its practical application value. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 is a flowchart illustrating a method for detecting forged power load data provided in an embodiment of this application;

[0050] Figure 2 is a flowchart illustrating the target data detection model determination process provided in an embodiment of this application;

[0051] Figure 3 is a flowchart illustrating the detection process of a target data detection model provided in an embodiment of this application;

[0052] Figure 4 is a flowchart illustrating the prediction process of a multi-objective prediction sub-model provided in an embodiment of this application;

[0053] Figure 5 is a structural schematic diagram of a power load data forgery detection device provided in an embodiment of this application;

[0054] Figure 6 is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] The generation process of massive amounts of forged data is uncontrollable, introducing data with unexpected characteristics. Furthermore, data labels often lack manual annotation and traceability, leading to negative impacts on the power system. Therefore, the power system needs to develop forgery detection technology to verify the authenticity of power load data. Currently, forgery detection technology requires building forgery detection models that are customized for training based on the type of forgery. This results in low adaptability of the method in power scenarios with high randomness, high complexity, and multiple influencing factors, affecting its practical application value.

[0057] Based on this, this application proposes the following technical solution, as detailed below:

[0058] In one embodiment, as shown in FIG1, FIG1 is a flowchart illustrating a method for detecting forged power load data provided in this application embodiment; this application provides a method for detecting forged power load data, specifically including the following:

[0059] S110: Acquire the power load data to be detected, encode the power load data, and generate the image to be detected.

[0060] In this embodiment, when a user needs to verify the authenticity of power load data, they can select the corresponding data on the acquisition interface of the computer device so that the computer device can acquire the power load data to be detected, encode the power load data, generate an image to be detected, and then perform relevant detection operations on the image to be detected.

[0061] It is understood that power load data refers to the power load borne by power supply equipment at all levels in a power system. When computer equipment needs to obtain relevant power load data, it can collect data through various power grid companies or open-source data platforms. During the data collection process, computer equipment can collect data according to different classification methods to improve the efficiency of power load data processing. For example, according to the time of load occurrence in the power system, it can be divided into peak load, minimum load, and average load; according to the degree of loss caused by sudden power outages, it can be divided into primary load, secondary load, and tertiary load. In this application, however, the computer equipment mainly collects power load data according to time classification.

[0062] Specifically, after collecting power load data, computer equipment can perform preprocessing operations on the data. This vectorization processing includes, but is not limited to, denoising, data cleaning, feature extraction and dimensionality reduction, and data normalization. Denoising refers to removing noise and interference from the original data to improve its quality and usability. Data cleaning involves outlier removal, missing value imputation, and data smoothing to improve accuracy and completeness. Feature extraction and dimensionality reduction extract the relevant information from the original data and convert it into a low-dimensional vector form for subsequent mathematical operations and analysis. Data normalization normalizes the original data to ensure equal weights for different features, avoiding data bias and errors. Therefore, the images generated by computer equipment after encoding power load data exhibit high realism and accuracy.

[0063] Furthermore, after generating the image to be detected, the computer equipment can perform preprocessing operations on the image to be detected again to improve its clarity. However, unlike data preprocessing, these preprocessing operations include sharpening and denoising. In detail, sharpening refers to compensating for the image's contours, enhancing the edges and areas of gray-level abrupt changes to make the image clearer. It can be divided into spatial domain processing and frequency domain processing. It improves the contrast between the edges of elements and surrounding pixels by highlighting the edges, contours, or features of certain linear target elements in the image. Denoising refers to the process of reducing noise in digital images. Generally, during the digitization and transmission of images, they are often affected by noise interference from imaging equipment and the external environment. That is, the received image information generally includes noise, which becomes a significant cause of image interference. By performing denoising processing on the image, the noise in the image is removed, further improving the realism and accuracy of the obtained image.

[0064] S120: Determine the target data detection model.

[0065] In this embodiment, after generating the image to be detected in step S110, the computer device can determine the target data detection model used to detect the image to be detected, and thus the target data detection model can be used to detect the authenticity of the image to be detected.

[0066] Specifically, the target data detection model of this application includes a multi-objective prediction sub-model and a multi-objective optimization sub-model. Since there are various types of forgery data, each type exhibits significant feature differences. Therefore, the computer device can use the multi-objective prediction sub-model to predict the probability that the image to be detected belongs to that forgery type, and after obtaining the forgery probabilities corresponding to each forgery type, use the multi-objective optimization sub-model to maximize the average probability of each forgery probability. This allows the predicted forgery results generated by the model to more quickly and accurately approach the direction of the true forgery results.

[0067] Maximizing the probability average refers to maximizing the weighted average of a set of probability values ​​to determine the probability distribution of multiple events, thereby achieving a specific optimal objective. Therefore, this application can employ probability average maximization to improve the accuracy of predicting forged results.

[0068] It is understandable that the multi-objective prediction sub-model in the target data detection model has reproducibility and specificity. Therefore, the multi-objective prediction sub-model can extract forgery features unrelated to power load characteristics from the image to be detected, and then calculate the predicted forgery result. However, considering the incomplete decoupling between model features and power load characteristics caused by the mixing and superposition of customized forgery detection models and other factors, the target data detection model built in this application outputs not only the predicted result of whether the power load data is forged, but also the probability result of the power load data belonging to each forgery type and the probability result of the combination of forgery types, making it more scientific and more valuable for practical application.

[0069] Furthermore, the target data detection model in this application can be pre-stored in a computer device. This allows the computer device to directly call the pre-stored target data detection model to perform the detection operation on the power load data when subsequent power load data authenticity verification is required. Additionally, the target data detection model in this application can select a neural network model as the initial model for improvement and training to obtain a target data detection model capable of performing authenticity verification on the image to be detected.

[0070] S130: Input the image to be detected into the target data detection model to obtain the predicted forgery result corresponding to the power load data output by the target data detection model.

[0071] In this embodiment, after the target data detection model is determined in step S120, the computer device can input the image to be detected into the target data detection model and use the target data detection model to detect the image to be detected, thereby obtaining the predicted forgery result corresponding to the power load data output by the target data detection model.

[0072] Specifically, after the image to be detected is input into the target data detection model, the target data detection model can use a multi-objective prediction sub-model to extract forgery features from the image to be detected, and cross-superimpose the extracted forgery features. Then, the probability value of the image to be detected belonging to each forgery type can be determined by the cross-superposition result. Then, the target data detection model can use a multi-objective optimization sub-model to maximize the probability average of each probability value, and then output the predicted forgery result corresponding to the power load data based on the maximization result.

[0073] In the above embodiments, when a user needs to detect the authenticity of power load data, the power load data to be detected can be acquired first, and the power load data can be encoded to generate an image to be detected. This allows for better capture of the temporal features in the power load data through the image to be detected, thereby improving detection accuracy. Next, a target data detection model can be determined. This target data detection model consists of a multi-objective prediction sub-model and a multi-objective optimization sub-model. The multi-objective prediction sub-model is used to predict the forgery probability corresponding to each of the multiple forgery types, while the multi-objective optimization sub-model is used to maximize the average probability of each forgery probability. Therefore, after inputting the image to be detected into the target data detection model, the predicted forgery result corresponding to the power load data can be obtained from the target data detection model. This application, through the target data detection model, can improve the adaptability of the power scenario during the detection process by superimposing features at the temporal level, and utilize the predicted forgery result for probability analysis, making it more scientific and thus improving its practical application value.

[0074] In one embodiment, the step S110 of encoding the power load data to generate the image to be detected may include:

[0075] S111: Determine multiple timestamps in the power load data according to time sequence, and obtain the load value corresponding to each timestamp.

[0076] S112: Use a preset encoding method to encode the information of each load value, obtain the encoding result, and use a data visualization tool to visualize the encoding result and generate the image to be detected.

[0077] In this embodiment, when the computer device converts power load data into an image to be detected, it can first determine multiple timestamps in the power load data according to the time sequence, and obtain the load value corresponding to each timestamp. Then, it can use a preset encoding method to encode each load value to obtain the encoding result, and use a data visualization tool to visualize the encoding result to generate the image to be detected.

[0078] Understandably, since power load data is collected in a time-series manner, before encoding the power load data, the computer equipment can first determine multiple timestamps based on a preset time window, thereby obtaining the load value corresponding to each timestamp in the power load data. Then, the computer equipment can select an appropriate preset encoding method according to the model's detection requirements and use this preset encoding method to encode each load value, obtaining the encoding result. The preset encoding method here can be a line graph, heatmap, bar chart, etc., and is not limited here.

[0079] Specifically, when the preset encoding method is a line graph, the computer equipment can convert the power load data into a line graph with time as the horizontal axis and load value as the vertical axis. When the preset encoding method is a heat map, the computer equipment can convert the power load data into a two-dimensional matrix heat map with time as the horizontal axis, load level as the vertical axis, and load value as the matrix value. When the preset encoding method is a bar chart, the computer equipment can divide the power load data into time periods, calculate the average load data value for each time period, and then represent the average value as the height of the bar chart, thus obtaining a bar chart. Finally, the computer equipment can use data visualization tools to visualize the graph and generate the image to be detected.

[0080] Furthermore, after the computer equipment generates the image to be detected, it can also adjust and optimize the image, such as adjusting the color scheme, line thickness, image size and resolution, to ensure that the image is clear and easy to read and can accurately convey the characteristics and trends of the power load data.

[0081] In one embodiment, as shown in FIG2, FIG2 is a flowchart illustrating a target data detection model determination process provided in an embodiment of this application; in FIG2, step S120, determining the target data detection model, may include:

[0082] S121: Obtain sample load data, which includes multiple images to be detected with different forgery types and the corresponding real forgery results for each image to be detected.

[0083] S122: Input the sample load data into the preset initial data detection model to obtain the predicted forgery result output by the initial data detection model.

[0084] S123: The initial data detection model is trained with the goal of predicting that the forged results will approximate the real forged results of the sample load data.

[0085] S124: When the initial data detection model meets the preset training conditions, the trained initial data detection model is used as the target data detection model.

[0086] In this embodiment, when determining the target data detection model, the computer device can first acquire sample payload data. This sample payload data includes multiple images to be detected with different forgery types and the corresponding real forgery results for each image. Therefore, after inputting the sample payload data into a preset initial data detection model and obtaining the predicted forgery results output by the initial data detection model, the computer device can train the initial data detection model with the goal of making the predicted forgery results approximate the real forgery results of the sample payload data. When the initial data detection model meets preset training conditions, the trained initial data detection model is used as the target data detection model.

[0087] It should be noted that the images to be detected in the sample load data can include not only images converted from forged data of different forgery types, but also images converted from power load data collected from the power system; there are no restrictions on this.

[0088] It is understood that the target data detection model in this application refers to a model that detects input images to be detected and obtains forged results. During model training, this target data detection model can use images of different forged types as training samples, and label each training sample with a sample label, i.e., the corresponding real forged result. After all training samples are labeled, the labeled training samples can be input into a pre-set initial data detection model for forward propagation to train the model. During the backpropagation process, a pre-set target loss function is used to fine-tune the model's parameters. When the model meets certain training conditions or parameter convergence conditions, such as when the number of iterations reaches a set value, training is considered complete. At this point, the trained model can be used as the final target data detection model.

[0089] Of course, before inputting the sample load data into the preset initial matting model, this application can also preprocess the image to be detected, such as performing normalization, sharpening, and denoising, to effectively improve the clarity of the image to be detected and facilitate model training efficiency.

[0090] In one embodiment, obtaining sample load data in step S121 may include:

[0091] S1211: Acquire the pre-collected real load data and convert the real load data into a real load image.

[0092] S1212: Determine multiple load data forgery models of different forgery types, and input the real load image into each load data forgery model to obtain the forged load image output by each load data forgery model.

[0093] S1213: Take the real load image and each fake load image as the images to be detected, and mark the real fake result of each image to be detected to form sample load data.

[0094] In this embodiment, when acquiring sample load data, the computer device can first acquire pre-collected real load data and convert the real load data into a real load image; then, it can determine multiple load data forgery models of forgery types and input the real load image into each load data forgery model to obtain the forged load image output by each load data forgery model; finally, the computer device can use the real load image and each forged load image as images to be detected and mark the real forgery result of each image to be detected, thereby forming sample load data.

[0095] Understandably, when computer equipment generates fake load data using real load data, it can first determine multiple fake load data forgery models based on actual needs or detection objectives. These models can simulate different forgery methods or attack methods, such as adding random noise, modifying trends, and periodic interference, to generate fake load data.

[0096] Specifically, the load data spoofing model in this application can select a neural network model as a preset model for improvement and training. This neural network model can be a basic model such as a generative adversarial network, a deep convolutional network, or an autoencoder, or a hybrid superposition model of different basic models, or a custom model after custom improvement of the above models; no restrictions are placed here. In more detail, for each spoofing type, the computer device can pre-acquire real load data and spoofed load data of the corresponding spoofing type obtained by expanding the real load image, and then use this data to train the preset initial model, thereby obtaining and storing the load data spoofing model for that spoofing type.

[0097] Therefore, when computer equipment needs to falsify real load data, it can first convert the real load data into a real load image, and then directly call the pre-stored load data falsification models of various falsification types. The real load image is then input into each load data falsification model to obtain the falsified load image output by each load data falsification model.

[0098] In one embodiment, as shown in Figure 3, which is a flowchart illustrating the detection process of a target data detection model according to an embodiment of this application, the target data detection model in step S130 may further include a authenticity discrimination network. The step of inputting the image to be detected into the target data detection model to obtain the predicted forgery result corresponding to the power load data output by the target data detection model may include:

[0099] S131: Input the image to be detected into the multi-objective prediction sub-model to predict the probability value of the image to be detected belonging to each forgery type, and obtain the prediction probability result output by the multi-objective prediction sub-model.

[0100] S132: Maximize the average probability of the predicted probability results using a multi-objective optimization sub-model to obtain the final probability result.

[0101] S133: The final probability result is judged as true or false by a true / false discrimination network, and a prediction falsification result corresponding to the power load data is generated.

[0102] In this embodiment, the target data detection model also includes a authenticity discrimination network. Therefore, the computer device can first input the image to be detected into the multi-objective prediction sub-model to predict the probability value of the image to be detected belonging to each forgery type, and obtain the prediction probability result output by the multi-objective prediction sub-model. Then, the multi-objective optimization sub-model can be used to maximize the probability average of the prediction probability result to obtain the final probability result. Finally, the computer device can use the authenticity discrimination network to distinguish the authenticity of the final probability result and generate the prediction forgery result corresponding to the power load data.

[0103] Specifically, the multi-objective prediction sub-model predicts the probability that an image to be detected belongs to each preset forgery type. It includes multiple output nodes, with one output node corresponding to each forgery type. The output value of each node represents the probability that the image belongs to the corresponding forgery type. Therefore, the output of the multi-objective prediction sub-model is a series of probability values, reflecting the model's judgment on whether the image belongs to a certain forgery type. The multi-objective optimization sub-model further processes the predicted probability results of the multi-objective prediction sub-model, aiming to maximize the average value of the predicted probability results. This integrates the probability values ​​of multiple forgery types, resulting in a more comprehensive and accurate judgment.

[0104] Furthermore, the authenticity discrimination network, acting as a binary classifier, can determine whether the image to be detected is real load data or fake load data based on the final probability result. When the image to be detected is real load data, the authenticity discrimination network can directly output the prediction of fake data; while when the image to be detected is fake load data, the prediction of fake data output by the authenticity discrimination network can include the final probability result, so that users can effectively identify fake load data and thus improve its practical application value.

[0105] In one embodiment, as shown in Figure 4, which is a flowchart illustrating the prediction process of a multi-objective prediction sub-model according to an embodiment of this application, the multi-objective prediction sub-model in step S131 may include a data format conversion network, a feature decomposition network, and a feature-to-temporal network. The process of inputting the image to be detected into the multi-objective prediction sub-model to predict the probability value of the image belonging to each forgery type, and obtaining the prediction probability result output by the multi-objective prediction sub-model, may include:

[0106] S1311: A data form conversion network is used to forge data in the image to be detected, and forged data corresponding to each forgery type of the image to be detected is obtained.

[0107] S1312: The feature decomposition network is used to decompose the image to be detected and each forged data to obtain the image feature data corresponding to the image to be detected and each forged data. The image feature data is then converted into time-series feature data through the data format conversion network.

[0108] S1313: The feature-to-temporal network is used to cross-stack various temporal feature data to obtain multiple temporal superimposed data. Based on each temporal superimposed data, the probability value of the image to be detected belonging to each forgery type is determined, forming the prediction probability result output by the multi-target prediction sub-model.

[0109] In this embodiment, the multi-objective prediction sub-model may include a data format conversion network, a feature decomposition network, and a feature-to-temporal network. Therefore, the computer device can first use the data format conversion network to perform data forgery on the image to be detected, obtaining the forged data corresponding to each forgery type of the image to be detected. Then, it can use the feature decomposition network to perform feature decomposition on the image to be detected and each forged data, obtaining the image feature data corresponding to the image to be detected and each forged data. Then, it can use the data format conversion network to convert each image feature data into temporal feature data. Finally, the computer device can use the feature-to-temporal network to cross-superimpose each temporal feature data to obtain multiple temporal superimposed data, and determine the probability value of the image to be detected belonging to each forgery type based on each temporal superimposed data, forming the prediction probability result output by the multi-objective prediction sub-model.

[0110] It should be noted that the data format conversion network can convert data between time-series and image formats. Therefore, in one application scenario, if the computer directly inputs time-series power load data into the target data detection model, the data format conversion network can first convert the power load data from time-series format to image format, i.e., the image to be detected, and then perform further detection operations.

[0111] Understandably, most existing data detection models can only detect data in image or video format and cannot directly detect time-series data. Therefore, the target data detection model built in this application can use a data format conversion network to convert power load data from time-series format to image format for feature extraction, and then restore the extracted feature images to time-series format before information fusion. In this way, the target data detection model of this application can improve the adaptability to power scenarios.

[0112] Specifically, the feature decomposition network can decompose the image to be detected and each forged data to obtain image feature data. The image feature data can be further subdivided into multiple model features to express it, thereby avoiding large differences in model features of the same forgery type, which would affect the subsequent feature superposition effect.

[0113] Next, the computer device can use a data format conversion network to convert the decomposed image feature data into temporal feature data, so that the feature-to-temporal network can cross-superimpose the temporal feature data at the temporal level to obtain multiple temporal superimposed data. In this way, the probability value corresponding to each forgery type calculated based on each temporal superimposed data can be closer to the truth.

[0114] In one embodiment, the time-series feature data in step S1313 may include power load features, basic model features, superimposed model features, and custom model features; wherein, using a feature-to-time-series network to cross-superimpose the various time-series feature data to obtain multiple time-series superimposed data may include:

[0115] S3131: Based on the power load characteristics of the image to be detected, the basic model features, superimposed model features and custom model features in each time series feature data are cross-superimposed to obtain the first time series superimposed data.

[0116] S3132: Based on the basic model features of the image to be detected, the power load features, superimposed model features and custom model features in each time series feature data are cross-superimposed to obtain the second time series superimposed data.

[0117] S3133: Based on the superimposed model features of the image to be detected, the power load features, basic model features and custom model features in each time series feature data are cross-superimposed to obtain the third time series superimposed data.

[0118] S3134: Based on the custom model features of the image to be detected, the power load features, basic model features and superimposed model features in each time series feature data are cross-superimposed to obtain the fourth time series superimposed data.

[0119] In this embodiment, the time-series feature data obtained from the target data detection model decomposition may include power load features, basic model features, superimposed model features, and custom model features. Therefore, when cross-superimposing the various time-series feature data, the basic model features, superimposed model features, and custom model features in each time-series feature data can be cross-superimposed based on the power load features of the image to be detected to obtain first time-series superimposed data; the power load features, superimposed model features, and custom model features in each time-series feature data can be cross-superimposed based on the basic model features of the image to be detected to obtain second time-series superimposed data; the power load features, basic model features, and custom model features in each time-series feature data can be cross-superimposed based on the superimposed model features of the image to be detected to obtain third time-series superimposed data; and the power load features, basic model features, and superimposed model features in each time-series feature data can be cross-superimposed based on the custom model features of the image to be detected to obtain fourth time-series superimposed data.

[0120] In one specific implementation, the computer device can obtain n sets of time-series feature data through target data detection model decomposition, specifically including power load features de_n, basic model features db_n, superimposed model features da_n, and custom model features dc_n. The computer can first cross-superimpose the n sets of time-series feature data to obtain an n*n*n*n superimposed time-series form: de_w + db_x + da_y + dc_z, where w = 1, 2, ..., n; x = 1, 2, ..., n; y = 1, 2, ..., n; z = 1, 2, ..., n. Then, the computer device can constrain this superimposed time-series form by minimizing the connection time-series classification loss, thereby forming four different sets of superimposed time-series data, specifically including the following:

[0121] (1) Based on the power load characteristics of the image to be detected, the basic model features, superimposed model features and custom model features in each time series feature data are cross-superimposed to obtain the first time series superimposed data: d1=de_w+db_x+da_y+dc_z, where x=1,2,...,n; y=1,2,...,n; z=1,2,...,n; so that the de_w obtained by the target data detection model is as close as possible to the original de_w, and thus the probability η1 of complete proximity is obtained.

[0122] (2) Based on the basic model features of the image to be detected, the power load features, superimposed model features and custom model features in each time series feature data are cross-superimposed to obtain the second time series superimposed data: d2=de_w+db_x+da_y+dc_z, where w=1,2,...,n; y=1,2,...,n; z=1,2,...,n; and then the probability η2 of d2 being classified as the basic model corresponding to db_x is obtained.

[0123] (3) Based on the superimposed model features of the image to be detected, the power load features, basic model features and custom model features in each time series feature data are cross-superimposed to obtain the third time series superimposed data: d3=de_w+db_x+da_y+dc_z, where w=1,2,...,n; x=1,2,...,n; z=1,2,...,n; and then the probability η3 of d3 being classified as the superimposed model corresponding to db_y is obtained.

[0124] (4) Based on the custom model features of the image to be detected, the power load features, basic model features and superimposed model features in each time series feature data are cross-superimposed to obtain the fourth time series superimposed data: d4=de_w+db_x+da_y+dc_z, where w=1,2,...,n; x=1,2,...,n; y=1,2,...,n; and then the probability η4 of d4 being classified as the custom model corresponding to db_z is obtained.

[0125] The power load data forgery detection device provided in the embodiments of this application is described below. The power load data forgery detection device described below can be referred to in correspondence with the power load data forgery detection method described above.

[0126] In one embodiment, as shown in FIG5, FIG5 is a structural schematic diagram of an electricity load data forgery detection device provided in an embodiment of this application; this application also improves an electricity load data forgery detection device, including an information encoding module 210, a model determination module 220, and a model prediction module 230, specifically including the following:

[0127] The information encoding module 210 is used to acquire the power load data to be detected, encode the power load data, and generate the image to be detected.

[0128] The model determination module 220 is used to determine the target data detection model. The target data detection model consists of a multi-objective prediction sub-model and a multi-objective optimization sub-model. The multi-objective prediction sub-model is used to predict the forgery probability corresponding to each forgery type among multiple forgery types, and the multi-objective optimization sub-model is used to maximize the average probability of each forgery probability.

[0129] The model prediction module 230 is used to input the image to be detected into the target data detection model and obtain the predicted forgery result corresponding to the power load data output by the target data detection model.

[0130] In the above embodiments, when a user needs to detect the authenticity of power load data, the power load data to be detected can be acquired first, and the power load data can be encoded to generate an image to be detected. This allows for better capture of the temporal features in the power load data through the image to be detected, thereby improving detection accuracy. Next, a target data detection model can be determined. This target data detection model consists of a multi-objective prediction sub-model and a multi-objective optimization sub-model. The multi-objective prediction sub-model is used to predict the forgery probability corresponding to each of the multiple forgery types, while the multi-objective optimization sub-model is used to maximize the average probability of each forgery probability. Therefore, after inputting the image to be detected into the target data detection model, the predicted forgery result corresponding to the power load data can be obtained from the target data detection model. This application, through the target data detection model, can improve the adaptability of the power scenario during the detection process by superimposing features at the temporal level, and utilize the predicted forgery result for probability analysis, making it more scientific and thus improving its practical application value.

[0131] In one embodiment, the information encoding module 210 may include:

[0132] The numerical determination submodule is used to determine multiple timestamps in the power load data in chronological order and obtain the load value corresponding to each timestamp.

[0133] The numerical coding submodule is used to encode information for each load value using a preset coding method, obtain the coding result, and visualize the coding result using a data visualization tool to generate the image to be detected.

[0134] In one embodiment, the model determination module 220 may include:

[0135] The data acquisition submodule is used to acquire sample payload data, which includes multiple images to be detected with different forgery types and the corresponding real forgery results for each image to be detected.

[0136] The model prediction submodule is used to input sample load data into a preset initial data detection model to obtain the predicted forgery results output by the initial data detection model.

[0137] The model training submodule is used to train the initial data detection model with the goal of predicting forged results that approximate the real forged results of the sample load data.

[0138] The model generation submodule is used to use the trained initial data detection model as the target data detection model when the initial data detection model meets the preset training conditions.

[0139] In one embodiment, the data acquisition submodule may include:

[0140] The data conversion unit is used to acquire pre-collected real load data and convert the real load data into a real load image.

[0141] The image forgery unit is used to determine multiple forgery types of load data forgery models, and input the real load image into each load data forgery model to obtain the forged load image output by each load data forgery model.

[0142] The data forming unit is used to take the real load image and each fake load image as the image to be detected, and to mark the real fake result of each image to be detected, so as to form sample load data.

[0143] In one embodiment, the model prediction module 230 may include:

[0144] The probability prediction submodule is used to input the image to be detected into the multi-object prediction sub-model to predict the probability value of the image to be detected belonging to each forgery type, and obtain the prediction probability result output by the multi-object prediction sub-model.

[0145] The probability maximization submodule is used to maximize the average probability of the predicted probability results using a multi-objective optimization sub-model, thus obtaining the final probability result.

[0146] The authenticity discrimination submodule is used to distinguish the authenticity of the final probability results through the authenticity discrimination network and generate prediction forgery results corresponding to the power load data.

[0147] In one embodiment, the probability prediction submodule may include:

[0148] The data forgery unit is used to forge data in the image to be detected using a data form conversion network, thereby obtaining forged data corresponding to each forgery type of the image to be detected.

[0149] The image conversion unit is used to perform feature decomposition on the image to be detected and each forged data using a feature decomposition network to obtain the image feature data corresponding to the image to be detected and each forged data, and to convert each image feature data into time-series feature data through a data format conversion network.

[0150] The result output unit is used to cross-superimpose various temporal feature data using a feature-to-temporal network to obtain multiple temporal superimposed data, and to determine the probability value of the image to be detected belonging to each forgery type based on each temporal superimposed data, forming the prediction probability result output by the multi-target prediction sub-model.

[0151] In one embodiment, the result output unit may include:

[0152] The first cross-overlay subunit is used to cross-overlay the basic model features, overlay model features and custom model features in each time series feature data based on the power load features of the image to be detected, so as to obtain the first time series overlay data.

[0153] The second cross-overlay subunit is used to cross-overlay the power load features, overlay model features and custom model features in each time series feature data based on the basic model features of the image to be detected, so as to obtain the second time series overlay data.

[0154] The third cross-overlay subunit is used to cross-overlay the power load features, basic model features and custom model features in each time series feature data based on the overlay model features of the image to be detected, so as to obtain the third time series overlay data.

[0155] The fourth cross-overlay subunit is used to cross-overlay the power load features, basic model features, and overlay model features in each time series feature data based on the custom model features of the image to be detected, so as to obtain the fourth time series overlay data.

[0156] In one embodiment, this application also provides a storage medium storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the power load data forgery detection method as described in any of the above embodiments.

[0157] In one embodiment, this application also provides a computer device storing computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the power load data forgery detection method as described in any of the above embodiments.

[0158] Schematably, as shown in FIG6, FIG6 is a schematic diagram of the internal structure of a computer device 300 provided in an embodiment of this application. The computer device 300 can be provided as a server. Referring to FIG6, the computer device 300 includes a processing component 302, which further includes one or more processors, and memory resources represented by memory 301 for storing instructions executable by the processing component 302, such as application programs. The application programs stored in memory 301 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 302 is configured to execute instructions to perform the power load data forgery detection method of any of the above embodiments.

[0159] The computer device 300 may also include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate on an operating system stored in memory 301, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.

[0160] Those skilled in the art will understand that the structure shown in Figure 6 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0161] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0162] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0163] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A power load data falsification detection method characterized by, The method comprises: acquiring power load data to be detected, and performing information coding on the power load data to generate a to-be-detected image; determining a target data detection model, the target data detection model being composed of a multi-target prediction sub-model and a multi-target optimization sub-model, the multi-target prediction sub-model being used for predicting a forgery probability corresponding to each of multiple forgery types, and the multi-target optimization sub-model being used for maximizing a probability average value of each forgery probability; inputting the to-be-detected image into the target data detection model to obtain a predicted forgery result corresponding to the power load data and output by the target data detection model.

2. The power load data falsification detection method according to claim 1, characterized by, The information coding on the power load data to generate a to-be-detected image comprises: determining multiple time stamps in the power load data in chronological order, and acquiring a load value corresponding to each time stamp; performing information coding on each load value by using a preset coding mode to obtain a coding result, and performing data visualization on the coding result by using a data visualization tool to generate a to-be-detected image.

3. The power load data falsification detection method according to claim 1, characterized by, The determination of the target data detection model comprises: acquiring sample load data, the sample load data comprising multiple to-be-detected images of different forgery types and a true forgery result corresponding to each to-be-detected image; inputting the sample load data into a preset initial data detection model to obtain a predicted forgery result output by the initial data detection model; training the initial data detection model with a target of the predicted forgery result approaching the true forgery result of the sample load data; when the initial data detection model meets a preset training condition, taking the trained initial data detection model as the target data detection model.

4. The power load data falsification detection method according to claim 3, characterized by, The acquisition of the sample load data comprises: acquiring pre-collected true load data, and converting the true load data into a true load image; determining multiple load data forgery models of different forgery types, and inputting the true load image into each load data forgery model to obtain a forged load image output by each load data forgery model; taking the true load image and each forged load image as a to-be-detected image, marking a true forgery result of each to-be-detected image, and forming sample load data.

5. The power load data falsification detection method of claim 1, wherein, The target data detection model further comprises a true-forgery discrimination network. The inputting of the to-be-detected image into the target data detection model to obtain a predicted forgery result corresponding to the power load data and output by the target data detection model comprises: inputting the to-be-detected image into the multi-target prediction sub-model to predict a probability value of the to-be-detected image belonging to each forgery type, and obtaining a prediction probability result output by the multi-target prediction sub-model; maximizing a probability average value of the prediction probability result by using the multi-target optimization sub-model to obtain a final probability result; performing true-forgery discrimination on the final probability result by using the true-forgery discrimination network to generate a predicted forgery result corresponding to the power load data.

6. The power load data falsification detection method according to claim 5, characterized by, The multi-target prediction sub-model comprises a data form conversion network, a feature decomposition network, and a feature-to-time sequence network. The inputting the to-be-detected image into the multi-target prediction sub-model to predict a probability value of the to-be-detected image belonging to each forgery type, to obtain a prediction probability result output by the multi-target prediction sub-model, includes: The data form conversion network is used to perform data forgery on the to-be-detected image, to obtain corresponding forgery data of the to-be-detected image on each forgery type; The feature decomposition network is used to perform feature decomposition on the to-be-detected image and each forgery data, to obtain image feature data corresponding to the to-be-detected image and each forgery data, and each image feature data is converted into time sequence feature data through the data form conversion network; The feature-to-time sequence network is used to cross and superimpose each time sequence feature data, to obtain a plurality of time sequence superimposed data, and to determine the probability value of the to-be-detected image belonging to each forgery type based on each time sequence superimposed data, to form the prediction probability result output by the multi-target prediction sub-model.

7. The power load data falsification detection method according to claim 6, characterized by, The time sequence feature data includes power load features, basic model features, superimposed model features, and custom model features; The feature-to-time sequence network is used to cross and superimpose each time sequence feature data, to obtain a plurality of time sequence superimposed data, and to determine the probability value of the to-be-detected image belonging to each forgery type based on each time sequence superimposed data, to form the prediction probability result output by the multi-target prediction sub-model. The first time sequence superimposed data is obtained by cross superimposing the basic model features, the superimposed model features, and the custom model features in each time sequence feature data based on the power load features of the to-be-detected image; The second time sequence superimposed data is obtained by cross superimposing the power load features, the superimposed model features, and the custom model features in each time sequence feature data based on the basic model features of the to-be-detected image; The third time sequence superimposed data is obtained by cross superimposing the power load features, the basic model features, and the custom model features in each time sequence feature data based on the superimposed model features of the to-be-detected image; The fourth time sequence superimposed data is obtained by cross superimposing the power load features, the basic model features, and the superimposed model features in each time sequence feature data based on the custom model features of the to-be-detected image.

8. An electric power load data falsification detection device characterized by comprising: The information encoding module is configured to obtain power load data to be detected, and encode information of the power load data to generate a to-be-detected image; The model determination module is configured to determine a target data detection model, the target data detection model being composed of a multi-target prediction sub-model and a multi-target optimization sub-model, the multi-target prediction sub-model being configured to predict a forgery probability corresponding to each of a plurality of forgery types, and the multi-target optimization sub-model being configured to maximize a probability average value of each forgery probability; The model prediction module is configured to input the to-be-detected image into the target data detection model to obtain a predicted forgery result corresponding to the power load data output by the target data detection model. The storage medium stores computer readable instructions, and the computer readable instructions are executed by one or more processors to perform the steps of the power load data forgery detection method in any one of claims 1 to 7.

9. A storage medium characterized by: The one or more processors and the memory are included.

10. A computer device, characterized in that, ​ ​ The memory stores computer readable instructions which, when executed by the one or more processors, perform the steps of the power load data falsification detection method of any one of claims 1 to 7.

Images