New energy power grid false data detection method and system based on deep learning

Abstract

The invention discloses a new energy power grid false data detection method based on deep learning, and belongs to the technical field of smart power grid safety, and the method comprises the steps: collecting a time series data set of an energy monitoring system, the data set including energy consumption, production and system performance parameters; integrating logistic regression and the decision tree into a deep learning model; importing the time sequence data set into the integrated deep learning model, performing FDIA attack simulation, analyzing the influence of false data injection attack on the data set, and performing corresponding adjustment on the deep learning model according to a simulation result to obtain an optimal model; and collecting operation data of the new energy power grid in real time, inputting the operation data into the optimal deep learning model for false data detection, and taking corresponding response measures according to a detection result.

Description

Technical Field

[0001] The present invention belongs to the field of smart grid security technology, and relates to a method for detecting false data in a power grid, and in particular to a method and system for detecting false data in a new energy power grid based on deep learning. Background Art

[0002] With the rapid development of smart grid technology, the security and stability of power systems have become a key issue. Smart grids monitor and manage power systems in real time through sensors and communication networks, achieving efficient allocation and utilization of power resources. However, the high interconnectivity and data dependence of smart grids make them vulnerable to cyber attacks, especially false data injection attacks (FDIA). FDIA can lead to incorrect power dispatch decisions and even cause serious consequences such as large-scale power outages by tampering with sensor data or network communication data.

[0003] In the existing technology, the methods for detecting and defending FDIA mainly include physical model-based methods and data-driven methods. The physical model method uses the physical characteristics of the power system to identify data anomalies through anomaly detection algorithms; while the data-driven method uses machine learning and data mining techniques to learn normal system behavior patterns from historical data to identify anomalies. However, these methods still have certain limitations in practical applications, such as insufficient detection capabilities for complex attack patterns and computational efficiency issues in high-dimensional data. Summary of the invention

[0004] Traditional false data injection attack detection methods have low detection efficiency and lack of intelligence, and the detection algorithm has low accuracy; false data injection attacks are complicated and difficult to identify. In order to overcome the defects of the above-mentioned prior art, the present invention provides a new energy power grid false data detection method and system based on deep learning.

[0005] The first aspect of the present invention provides a new energy grid false data detection method based on deep learning, comprising the following steps:

[0006] Collect time series data sets from energy monitoring systems, including energy consumption, production, and system performance parameters;

[0007] Integrate logistic regression and decision trees into a single deep learning model;

[0008] Import the time series data set into the integrated deep learning model, perform FDIA attack simulation, analyze the impact of false data injection attacks on the data set, and adjust the deep learning model accordingly based on the simulation results to obtain the optimal model;

[0009] The operating data of the new energy power grid is collected in real time and input into the optimal deep learning model for false data detection, and corresponding response measures are taken according to the detection results.

[0010] Optionally, the time series data set includes the timestamp of the recorded data, the energy consumed or generated daily, the energy consumption or production in the past week, the energy consumption or production in the past month, the energy consumption or production in the past year, the voltage measured in the three-phase grid, the current measured in the three-phase grid, and the power measured through the three-phase grid.

[0011] Optionally, the integration of logistic regression and decision tree into a deep learning model is specifically:

[0012] Denote the prediction of logistic regression as Clr and the prediction of decision tree as Edt;

[0013] Logistic regression outputs the probability of a binary classification, between 0 and 1, and these probabilities are thresholded to obtain a binary prediction; decision trees provide binary predictions directly based on the features;

[0014] The predictions of logistic regression and decision trees are combined using a majority voting scheme to integrate logistic regression and decision trees into one deep learning model.

[0015] Optionally, the integrated prediction GEn can be expressed as:

[0016] GEn=MV(Clr,Edt)

[0017] Where Clr is the prediction from the logistic regression, Edt is the prediction from the decision tree, and MV(.) is the function that returns the majority class in its input.

[0018] Optionally, the step of importing the time series data set into the integrated deep learning model, performing FDIA attack simulation, analyzing the impact of the false data injection attack on the data set, and adjusting the deep learning model accordingly according to the simulation results to obtain the optimal model specifically includes:

[0019] Import the dataset;

[0020] Simulate FDIA attacks, generate simulated attack data and inject imported data sets;

[0021] Perform data preprocessing on the dataset injected with simulated attack data;

[0022] Select features from the processed dataset for training the deep learning model and increase the feature value by 2.0 times;

[0023] Use the processed data and features to train deep learning models to detect attack data;

[0024] Evaluate the performance of the trained model and adjust the model accordingly based on the evaluation results to obtain the deep learning model with the best detection effect.

[0025] Optionally, the data preprocessing of the data set injected with the simulated attack data specifically includes:

[0026] The time series dataset is processed by SMOTEENN, which combines oversampling of the minority class with SMOTE and uses ENN to clean the dataset.

[0027] Optionally, taking corresponding response measures according to the detection result specifically includes:

[0028] Determine whether there is a false data attack based on the model output results. If an attack is detected, the early warning mechanism is triggered and the attack event is recorded. Based on the characteristics of the attack, the power grid operation parameters are adjusted to reduce the impact of the attack and ensure safe and stable operation of the system.

[0029] The second aspect of the present invention provides a new energy power grid false data detection system based on deep learning, based on the new energy power grid false data detection method based on deep learning described in the first aspect of the present invention, characterized in that the system comprises:

[0030] Dataset acquisition module, algorithm integration module, model building module and model application module;

[0031] The data set acquisition module is used to acquire the time series data set of the energy monitoring system; the algorithm integration module is used to integrate logistic regression and decision tree into a deep learning model; the model construction module is used to import the time series data set after step processing into the integrated deep learning model, perform FDIA attack simulation, analyze the impact of false data injection attack on the data set, and adjust the deep learning model accordingly according to the simulation results to obtain the optimal model; the model application module is used to collect the operation data of the new energy power grid in real time and input it into the optimal deep learning model for false data detection, and take corresponding response measures according to the detection results.

[0032] The third aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, the method for detecting false data in a new energy power grid based on deep learning according to the first aspect of the present invention is implemented.

[0033] The fourth aspect of the present invention provides a computer-readable storage medium, which stores a computer program, and is characterized in that when the computer program is executed by a processor, it implements a new energy power grid false data detection method based on deep learning according to the first aspect of the present invention.

[0034] Compared with the prior art, the new energy grid false data detection method and system based on deep learning proposed in the present invention has the following advantages:

[0035] Overcoming the defect of low detection efficiency: Traditional detection methods have low detection efficiency for false data injection attacks (FDIA), while the present invention achieves a significant improvement in detection efficiency through the integration of deep learning models of logistic regression and decision trees, as well as efficient preprocessing of time series data.

[0036] Improved detection accuracy: Compared with the limitations of existing methods in dealing with complex attack patterns, the present invention significantly improves the accuracy of false data detection (F1 score reaches 1.0) by integrating multiple algorithms, adopting majority voting strategy and model optimization steps.

[0037] Enhanced ability to cope with complex attack patterns: Traditional methods are difficult to detect complex false data attacks. The present invention applies a variety of advanced technologies in data set feature selection and model training, such as SMOTEENN balanced data and recursive feature elimination method, thereby enhancing adaptability to complex attacks.

[0038] Improvement in high-dimensional data processing efficiency: Existing methods have insufficient computational efficiency when processing high-dimensional data. The present invention significantly improves the processing efficiency of high-dimensional time series data through automatic encoder and deep learning model optimization to meet practical application needs.

[0039] Intelligent real-time response capability: The present invention integrates an early warning mechanism and parameter adaptive adjustment function, which can respond and adjust the grid operation parameters in real time when an attack is detected, ensuring the safety and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is the voltage amplitude before the attack in the embodiment of the present invention;

[0041] Figure 2 is a voltage amplitude curve after attack in an embodiment of the present invention;

[0042] Figure 3 The figure is a schematic diagram of the overall process of detecting FDIA attacks in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0044] In Example 1, the present invention provides a new energy grid false data detection method based on deep learning, comprising the following steps:

[0045] Step 1: Collect the time series data set of the energy monitoring system through the IoT server, which includes energy consumption, production and system performance parameters;

[0046] In a further preferred but non-limiting embodiment, in step 1, the IoT server stores data of a 10kV solar power generation system, which is collected for laboratory research purposes. The following briefly describes each column in the data set:

[0047] Date and time: timestamp of recorded data;

[0048] DlyEn (Daily Energy): Energy consumed or generated daily, measured in appropriate units (including kilowatt-hours or similar);

[0049] TotEn (Total Energy): Cumulative total energy consumption or production up to that point;

[0050] ParEn (parameter energy): energy value of a specific measurement, which may be an additional or derived energy parameter;

[0051] WkE (Weekly Energy): Energy consumption or production over the past week;

[0052] MnE (Monthly Energy): Energy consumption or production in the past month;

[0053] YrE (Annual Energy): Energy consumption or production in the past year;

[0054] Gv3Ph (grid voltage, 3-phase): voltage measured in a three-phase grid;

[0055] Gc3Ph (grid current, 3 phase): current measured in the three-phase grid;

[0056] Gp3Ph (grid power, 3 phase): power measured via the three-phase grid;

[0057] Freq3Ph (frequency, 3 phases): frequency of the three-phase power grid;

[0058] InPw1 (input power 1): power input at point 1;

[0059] InPv1 (input voltage 1): voltage input of point 1;

[0060] InPc1 (input current 1): current input of point 1;

[0061] InPw2 (input power 2): power input at point 2;

[0062] InPv2 (input voltage 2): voltage input at point 2;

[0063] InPc2 (input current 2): current input at point 2;

[0064] InvTemp (inverter temperature): inverter temperature;

[0065] BstTemp (boost temperature): the temperature of the boost component;

[0066] IslRes (isolation resistance): The resistance value of system isolation is critical to safety;

[0067] WGFreq (wind turbine frequency): operating frequency of the wind turbine (if applicable);

[0068] Phi3Ph (Phase Angle, 3-Phase): The phase angle of a three-phase system indicates the power factor and efficiency.

[0069] Step 2, integrate logistic regression and decision tree into a deep learning model;

[0070] In a further preferred but non-limiting embodiment, in step 2, an integrated technology is used, specifically comprising:

[0071] Logistic regression and decision tree models are two standard machine learning techniques for classification applications. While logistic regression provides probabilistic predictions and is interpretable, decision trees are more flexible and can capture nonlinear relationships in the data.

[0072] Combining these two algorithms into an ensemble technique uses the power of each model, thereby improving overall predictive performance and model robustness. The ensemble approach combines logistic regression and decision tree prediction with a majority voting scheme:

[0073] If the two models predict the same class, the ensemble prediction will match; if the two models disagree, the majority vote determines the ensemble prediction;

[0074] By combining logistic regression with decision tree prediction, the ensemble can improve generalization performance and capture a more comprehensive representation of the underlying data patterns. Logistic regression and decision tree ensemble techniques provide a powerful and general solution for classification tasks, leveraging the advantages of both methods to improve prediction accuracy and model interpretability.

[0075] In a further preferred but non-limiting embodiment, the integration of logistic regression and decision tree into a deep learning model specifically includes:

[0076] Denote the prediction of logistic regression as Clr and the prediction of decision tree as Edt;

[0077] In this ensemble approach, logistic regression outputs probabilities of binary classification, usually between 0 and 1, which can be thresholded to obtain binary predictions (0 or 1);

[0078] Decision trees provide binary predictions directly based on features;

[0079] Ensembles combine the predictions of logistic regression and decision trees using a majority voting scheme.

[0080] Specifically, if the logistic regression and decision tree predict the same class, then the ensemble prediction is that class, and if the logistic regression and decision tree predict different classes, then the ensemble prediction is determined by a majority vote of the two predictions.

[0081] In a further preferred but non-limiting embodiment, the integrated prediction GEn can be expressed as:

[0082] GEn=MV(Clr,Edt)

[0083] Where Clr is the prediction from the logistic regression, Edt is the prediction from the decision tree, and MV(.) is the function that returns the majority class in its input.

[0084] This ensemble technique combines the advantages of decision trees (non-linear, flexible models) and logistic regression (interpretable, probabilistic predictions) to improve predictive performance and model robustness. In addition, this ensemble can provide a wider range of predictions, reduce the possibility of overfitting and capture all aspects of the data.

[0085] Step 3: Import the time series data set processed in step 1 into the deep learning model integrated in step 2, perform FDIA attack simulation, analyze the impact of false data injection attack on the data set, and adjust the deep learning model accordingly according to the simulation results to obtain the optimal model;

[0086] In a further preferred but non-limiting embodiment, step 4 specifically comprises:

[0087] Step 3.1, import the dataset;

[0088] Step 3.2, simulate the FDIA attack, generate simulated attack data and inject the dataset imported in step 4.1;

[0089] Step 3.3, preprocessing the data set injected with simulated attack data;

[0090] Step 3.4, select features for training the deep learning model from the processed data set and increase the feature value by 2.0 times;

[0091] Step 3.5, use the processed data and features to train a deep learning model to detect attack data;

[0092] Step 3.6, evaluate the performance of the trained model, adjust the model accordingly based on the evaluation results, and obtain the deep learning model with the best detection effect.

[0093] In a further preferred but non-limiting embodiment, the step of selecting features for training the deep learning model from the processed data set and increasing the feature value by 2.0 times specifically includes:

[0094] Select Gp3Ph and Gc3Ph in the time series dataset, increase the values ​​of Gp3Ph and Gc3Ph features by 2.0 times, obtain Gp3Ph_a and Gc3Ph_a, and generate labels;

[0095] The generated labels are subjected to feature selection using recursive feature elimination and cross validation (RFECV) to identify the most critical features for training the model.

[0096] In a further preferred but non-limiting embodiment, the data preprocessing of the data set injected with the simulated attack data specifically includes:

[0097] Process the time series dataset obtained in step 1 through SMTEENN to solve the class imbalance problem in the dataset;

[0098] Specifically, SMOTEENN combines oversampling of the minority class with SMOTE and uses ENN to clean the dataset, thus effectively balancing the class distribution.

[0099] Specifically, in the code, after RFECV, SMOTEENN is used on the feature selection data (X_rfecv and y) to generate a balanced dataset (X_resampled and y_resampled), and then train_test_split is used to split this balanced dataset into training and test sets; using SMOTEENN ensures that the decision tree classifier is trained on a dataset with roughly equal representation of the two classes, which helps prevent the model from being biased towards the majority class and improves its ability to effectively detect anomalies.

[0100] In a further preferred but non-limiting embodiment, the using the processed data and features to train the deep learning model specifically includes:

[0101] To demonstrate the impact of False Data Injection Attack (FDIA) on power data, the IEEE14 bus model was used, FDIA was injected on Bus 2 (the attack increased the active power output of the generator on Bus 2), and power flow analysis was performed before and after FDIA. To conduct this experiment, we used Python tools; the following steps were performed to perform the simulation:

[0102] The predefined power network is loaded with the help of the original power dataset.

[0103] Then, a power flow analysis is performed to determine whether it has converged successfully, and the power network data is presented as a result, which contains the following parameters: vm_pu (unit voltage amplitude), va_degree (voltage phase angle, usually in degrees, indicating the angular displacement of voltage relative to a reference point or reference phase angle), p_mw (megawatt active power) and q_mvar (MVAR reactive power). The results are shown in Tables 1 and 2 below.

[0104] Table 1. Current flow results before FDIA

[0105] node Virtual Machines va_degree power q_mvar 0 1.060000 0.000000 -1845.104233 -6684.031922 1 1.045000 0.028608 -85.000000 2046.038644 2 1.010000 0.193523 200.000000 10,699.913750 3 1.035265 -0.179392 200.000000 0.000000 4 1.040079 -0.098594 200.000000 0.000000 5 1.056000 -0.561160 0.000000 -6610.052732 6 1.050408 -0.471364 200.000000 0.000000 7 1.052931 -0.578354 0.000000 0.000000 8 1.052931 -0.578354 200.000000 0.000000 9 1.052885 -0.588686 0.000000 0.000000 10 1.052668 -0.638291 200.000000 0.000000 11 1.052423 -0.694122 200.000000 0.000000 12 1.052282 -0.726391 200.000000 0.000000 13 1.052235 -0.737150 200.000000 0.000000

[0106] Table 2. Current flow results after FDIA

[0107]

[0108]

[0109] Figure 1 , Figure 2 These are the voltage curves before and after the FDIA attack, which clearly show the impact of the attack on bus 2.

[0110] In the simulation of the present invention, Gc3Ph and Gp3Ph were used as two features. The present invention applied the following machine learning model to identify FDIA on the data set:

[0111] Isolation Forest, Random Forest, Support Vector Machine, Decision Tree, Autoencoder, and Logistic Regression.

[0112] The purpose of the simulation is to show the impact of false data injection attacks on the dataset and employ various machine learning algorithms to identify the attacks. Figure 3 The overall approach used in the simulation is explained. For other models, the training and model definition need to be modified; the rest of the steps remain the same.

[0113] (1) Isolation Forest

[0114] The Isolation Forest algorithm uses machine learning to detect anomalies. It uses a binary tree to randomly partition the data into subsets. Fewer splits in these trees isolate anomalies, which are data points that are significantly different from the majority. A threshold can be defined based on the algorithm's anomaly score to find anomalies. Isolation Forest is efficient, works well with high-dimensional data, and does not use complex distance metrics, although data distribution and parameter settings can affect its performance. By partitioning the dataset into separate segments and arbitrarily selecting features, the Isolation Forest method can separate anomalies until a single data point is isolated. Anomalies are expected to be easier to isolate and require fewer splits. Based on the average path length in the isolation tree, the anomaly score is calculated as follows:

[0115] p(y,m)=2 -F(g(y)) / e(m)

[0116] where F(g(y)) is the average path length of data point y in m isolation trees and e(m) is the normalization factor.

[0117] (2) Random Forest

[0118] The random forest ensemble machine learning technique uses multiple decision trees to enhance predictive power. Each decision tree is trained with randomly selected data and features to minimize overfitting and maximize generalization. Since random forest is a machine learning method that can handle a wide variety of data and tasks, it is often used for classification and regression problems. Aggregating the predictions from individual trees can help obtain the predictions of the random forest, as shown below:

[0119] t = trees (prediction from a single tree)

[0120] (3) SVM

[0121] Support Vector Machine (SVM) is a machine learning method for anomaly identification that is trained using only the majority class of data. It finds a hyperplane that encapsulates the majority class and identifies anomalies as data points outside this boundary during testing. It is useful for fraud detection and quality control when the goal is to find rare events in a regular dataset. The decision function of a one-class SVM is used to determine whether a data point is an anomaly:

[0122] g(a) = sign(distance(a)-th)

[0123] Where distance(a) represents the distance between data point a and the decision boundary, which is a threshold or predefined value.

[0124] (4) Decision Tree

[0125] A decision tree is a machine learning technique that models predictions and choices as a tree. It is used for classification and regression. The root node reflects the initial decision or feature in the tree. The leaf nodes represent the outcome or prediction, while the internal nodes reflect the intermediate judgments. Decision trees partition data by reducing impurity or increasing knowledge. They can be pruned to avoid overfitting. In classification, leaf nodes are class labels; in regression, they are numbers [27,28]. The prediction of a decision tree is made by traversing the tree according to the feature values ​​of the input sample until a leaf node is reached: y = prediction at the leaf node.

[0126] (5) Autoencoder

[0127] An autoencoder is a variant of an artificial neural network that is commonly used in unsupervised machine learning. The main applications of this technique include dimensionality reduction, feature learning, and data compression. An autoencoder consists of two main components: an encoder and a decoder. The main goal of an autoencoder is to reconstruct the input data, thereby obtaining a compressed representation of the data through unsupervised learning. The autoencoder learns to reconstruct its input data. The reconstruction error is used as a measure of anomaly. The reconstruction error is calculated as follows:

[0128]

[0129] Where ci represents the input data, c represents the reconstructed data, and m represents several features. If the reconstruction error is higher than a predefined threshold, an anomaly is detected.

[0130] (6) Logistic regression

[0131] Logistic regression is a supervised machine learning algorithm that represents the relationship between one or more independent variables (features) and the likelihood of a specific outcome occurring. It is often used to solve binary classification problems, such as determining whether an email is spam, whether a customer will buy a product, etc. The logistic regression model predicts the probability of the positive class using a logistic function, as shown below:

[0132]

[0133] Where C represents the input features and ω represents the model parameters.

[0134] (7)FFNN

[0135] A feed-forward neural network (FFNN) is an artificial neural network in which information flows in one direction: from the input layer to one or more hidden layers and then to the output layer. In the experiments, we used three layers: The first layer is the input layer, which receives the input data. Each node in this layer represents a feature of the input. There is also a hidden layer, which is an intermediate layer between the input and output layers, where processing is done. Each node in the hidden layer is linked to each node in the previous and next layers. The output layer is the last layer. The predictions of the MLP are obtained by forward propagation as shown below:

[0136] b 1 =CD (1) +g (1)

[0137] f (1) =ReLU(b (1) )

[0138] b (2) =f (1) C (2) +g (2)

[0139] f (2) =ReLU(b (2) )

[0140] b (L) =f (L-1) C (L) +g (L)

[0141] e=Softmax(b (L) )

[0142] Where C represents the input data, D (i) and g (i) represents the weight and bias of the i-th layer, f (i) represents the activation of the i-th layer, and e represents the predicted output.

[0143] Specifically, a detailed comparative evaluation of the algorithms based on anomaly detection parameters is presented below, and a detailed comparison table is given below. Selecting the most appropriate anomaly detection method from the options listed in Table 3 depends on the user's personal requirements and priorities. Each method has distinct advantages and disadvantages, and the best choice depends on various aspects such as the characteristics of the data, the relative importance of precision and recall, and computational efficiency. Common algorithms used in anomaly detection are: isolation forest, random forest, SVM, decision tree, logistic regression, and autoencoder. Table 3 below shows the performance metrics of these algorithms. Isolation forest and feedforward neural network (FFNN) show poor F1 score, precision, and recall performance. The isolation forest algorithm has an F1 score of 0.68, which indicates difficulty in reliably detecting anomalies, despite its accuracy of 0.76. Similarly, FFNN has an F1 score of 0.53, indicating that its ability to effectively generalize to unfamiliar data or accurately distinguish anomalies from common events is limited.

[0144] Table 3 Evaluation indicators

[0145]

[0146] Autoencoders, logistic regression, and support vector machines (SVMs) performed similarly across all evaluation metrics. The autoencoder achieved a high F1 score of 0.95, indicating that it is very effective in detecting anomalies while maintaining high levels of precision and recall. Logistic regression and SVM showed high robustness in correctly detecting anomalies, with an F1 score of 0.98. However, logistic regression had slightly worse model performance compared to SVM, indicating that it may have difficulty managing complex anomaly patterns. Among the ensemble methods, the choice between using decision trees and logistic regression depended on their performance in the evaluation metrics, as shown in Table 3.

[0147] The closer the ROC (AUC value) is to 1, the more likely the model can distinguish between positive and negative examples, with a higher true positive rate and lower false positive rate at various thresholds. As shown in Table 3, an AUC of 1 means that the model has a 100% chance of correctly identifying positive and negative examples.

[0148] The ensemble method (a combination of decision trees and logistic regression) performs well in all metrics: F1 score, precision, and recall are 1, while Kappa score and model accuracy are 0.99. These statistics clearly show that the ensemble method can detect anomalies with excellent results.

[0149] Step 4: Collect the operating data of the new energy power grid in real time and input it into the optimal deep learning model obtained in step 3 to detect false data, and take corresponding response measures based on the detection results.

[0150] In a further preferred but non-limiting embodiment, in step 4, taking corresponding response measures according to the detection results is specifically:

[0151] Determine whether there is a false data attack based on the model output results. If an attack is detected, the early warning mechanism is triggered and the attack event is recorded. Based on the characteristics of the attack, the power grid operation parameters are adjusted to reduce the impact of the attack and ensure safe and stable operation of the system.

[0152] Embodiment 2 of the present invention provides a new energy grid false data detection system based on deep learning, which is used to implement the new energy grid false data detection method based on deep learning described in embodiment 1. The system includes:

[0153] Dataset acquisition module, algorithm integration module, model building module and model application module;

[0154] The data set acquisition module is used to acquire the time series data set of the energy monitoring system; the algorithm integration module is used to integrate logistic regression and decision tree into a deep learning model; the model construction module is used to import the time series data set after step processing into the integrated deep learning model, perform FDIA attack simulation, analyze the impact of false data injection attack on the data set, and adjust the deep learning model accordingly according to the simulation results to obtain the optimal model; the model application module is used to collect the operation data of the new energy power grid in real time and input it into the optimal deep learning model for false data detection, and take corresponding response measures according to the detection results.

[0155] Embodiment 3 of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, a method for detecting false data in a new energy power grid based on deep learning according to Embodiment 1 is implemented.

[0156] Embodiment 4 of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements a new energy power grid false data detection method based on deep learning according to Embodiment 1.

[0157] The specific implementation methods of the present invention are not exhaustive and do not constitute a limitation on the scope of protection of the claims. Based on the inspiration gained from the embodiments of the present invention, those skilled in the art can think of other substantially equivalent alternatives without creative work, all of which are within the scope of protection of the present invention.

Claims

1. A new energy grid false data detection method based on deep learning, characterized in that: The steps include: Collect time series data sets from energy monitoring systems, including energy consumption, production, and system performance parameters; Integrate logistic regression and decision trees into a single deep learning model; Import the time series data set into the integrated deep learning model, perform FDIA attack simulation, analyze the impact of false data injection attacks on the data set, and adjust the deep learning model accordingly based on the simulation results to obtain the optimal model; The operating data of the new energy power grid is collected in real time and input into the optimal deep learning model for false data detection, and corresponding response measures are taken according to the detection results.

2. According to a method for detecting false data in a new energy grid based on deep learning according to claim 1, it is characterized in that: The time series data set includes the timestamp of the recorded data, the energy consumed or generated daily, the energy consumption or production in the past week, the energy consumption or production in the past month, the energy consumption or production in the past year, the voltage measured in the three-phase power grid, the current measured in the three-phase power grid, and the power measured through the three-phase power grid.

3. According to a method for detecting false data in a new energy grid based on deep learning according to claim 1, it is characterized in that: The integration of logistic regression and decision tree into a deep learning model is specifically as follows: Denote the prediction of logistic regression as Clr and the prediction of decision tree as Edt; Logistic regression outputs the probability of binary classification, between 0 and 1, and these probabilities are thresholded to obtain binary predictions; Decision trees provide binary predictions directly based on features; The predictions of logistic regression and decision trees are combined using a majority voting scheme to integrate logistic regression and decision trees into one deep learning model.

4. The method for detecting false data of a new energy power grid based on deep learning according to claim 3 is characterized in that: The integrated prediction GEn can be expressed as: GEn=MV(Clr,Edt) Where Clr is the prediction from the logistic regression, Edt is the prediction from the decision tree, and MV(.) is the function that returns the majority class in its input.

5. The method for detecting false data of a new energy power grid based on deep learning according to claim 1 is characterized in that: The method of importing the time series data set into the integrated deep learning model, performing FDIA attack simulation, analyzing the impact of the false data injection attack on the data set, and adjusting the deep learning model accordingly according to the simulation results to obtain the optimal model specifically includes: Import the dataset; Simulate FDIA attacks, generate simulated attack data and inject imported data sets; Perform data preprocessing on the dataset injected with simulated attack data; Select features from the processed dataset for training the deep learning model and increase the feature value by 2.0 times; Use the processed data and features to train deep learning models to detect attack data; Evaluate the performance of the trained model and adjust the model accordingly based on the evaluation results to obtain the deep learning model with the best detection effect.

6. The method for detecting false data of a new energy power grid based on deep learning according to claim 4 is characterized in that: The data preprocessing of the data set injected with the simulated attack data specifically includes: The time series dataset is processed by SMOTEENN, which combines oversampling of the minority class with SMOTE and uses ENN to clean the dataset.

7. The method for detecting false data of a new energy power grid based on deep learning according to claim 1, characterized in that: The specific response measures taken according to the detection results are: Determine whether there is a false data attack based on the model output results. If an attack is detected, the early warning mechanism is triggered and the attack event is recorded. Based on the characteristics of the attack, the power grid operation parameters are adjusted to reduce the impact of the attack and ensure safe and stable operation of the system.

8. A new energy grid false data detection system based on deep learning, based on a new energy grid false data detection method based on deep learning according to any one of claims 1 to 7, characterized in that: The system includes: Dataset acquisition module, algorithm integration module, model building module and model application module; The data set acquisition module is used to acquire the time series data set of the energy monitoring system; the algorithm integration module is used to integrate logistic regression and decision tree into a deep learning model; the model construction module is used to import the time series data set after step processing into the integrated deep learning model, perform FDIA attack simulation, analyze the impact of false data injection attack on the data set, and adjust the deep learning model accordingly according to the simulation results to obtain the optimal model; the model application module is used to collect the operation data of the new energy power grid in real time and input it into the optimal deep learning model for false data detection, and take corresponding response measures according to the detection results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into the processor, a method for detecting false data in a new energy power grid based on deep learning is implemented according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a method for detecting false data in a new energy power grid based on deep learning is implemented according to any one of claims 1 to 7.

Images