A multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker

Abstract

This invention discloses a multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker, belonging to the field of power system phase selection and breaking technology. The invention includes the following steps: S1, Data preprocessing: Normalizing the raw short-circuit current data; S2, Hybrid model construction: Constructing a CNN-SE-GRU hybrid model based on the processed current data in S1, combining convolutional neural networks, attention mechanisms, and GRU networks; S3, CEO algorithm optimization: Optimizing the hyperparameters of the constructed hybrid model using the CEO algorithm to obtain a CEO-CNN-SE-GRU hybrid prediction model; S4, Model training: Training the hybrid prediction model to obtain a stable model; S5, Prediction execution: Calling the trained hybrid prediction model to perform multi-step prediction of the short-circuit current, predicting the zero-crossing point of the short-circuit current. Compared with existing technologies, this invention improves the speed and accuracy of multi-step prediction of the zero-crossing point of short-circuit current, while also exhibiting better generalization.

Description

Technical Field

[0001] This invention relates to the field of power system phase-selective disconnection technology, and in particular to a multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker, which provides theoretical support for phase-controlled disconnection of circuit breakers. Background Technology

[0002] In modern power system operation, rapid and accurate handling of short-circuit faults is crucial for ensuring power supply reliability and system safety and stability. After a short-circuit fault occurs, the diversity of fault types (such as three-phase, two-phase, and single-phase-to-ground faults), the differences in system impedance parameters, and the uncertainty of the short-circuit point's location in the transmission network cause complex nonlinear changes in the short-circuit current waveform. Simultaneously, noise interference and harmonic pollution in the power system further exacerbate the distortion of the short-circuit current waveform, making the accurate and rapid prediction of the short-circuit current's zero-crossing point a long-standing technical challenge in the power sector.

[0003] Traditional short-circuit current zero-crossing prediction methods include the Weighted Least Mean Square Error (WLMS) algorithm, which continuously adjusts filter coefficients to minimize the mean square value of the error signal; the improved half-wave Fourier algorithm, based on the Fourier transform principle, performs spectral analysis on the short-circuit current signal to extract the fundamental component and predict the zero-crossing point; and the Recursive Least Squares (RLS) algorithm, which uses the least squares criterion to update parameters recursively. However, these classic algorithms generally suffer from excessively long sampling times, typically requiring more than 10ms of sampling data to complete the prediction, and exhibiting relatively large prediction errors, generally around ±1ms, which is insufficient to meet the practical requirements of power systems for rapid response to short-circuit faults.

[0004] With the widespread application of deep learning technology in power system fault analysis, the Long Short-Term Memory (LSTM) algorithm, thanks to its unique gating mechanism, has made significant progress in predicting the zero-crossing point of short-circuit current, greatly reducing the sampling time to 3ms. Nevertheless, its prediction error remains within ±0.5ms. In high-voltage and ultra-high-voltage power systems where extremely high accuracy is required, this error range still cannot fully meet the practical engineering needs of rapid fault isolation and precise circuit breaker opening and closing.

[0005] In view of this, the present invention proposes a multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker, which has important theoretical value and practical significance for improving the efficiency of power system fault handling, reducing the risk of equipment damage, and ensuring the safe and stable operation of the power grid. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker to solve the problems mentioned in the background art. This invention effectively improves the prediction speed and accuracy of the zero-crossing point prediction of short-circuit current, has better stability and noise resistance, and can adapt to short-circuit fault conditions under different operating conditions.

[0007] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0008] A multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker includes the following steps:

[0009] S1. Data preprocessing: Normalize the raw short-circuit current data and scale it to the [0,1] interval while keeping the current waveform change trend unchanged. Obtain the time series of the short-circuit node and divide the obtained data into training set, test set and validation set.

[0010] S2. Hybrid Model Construction: Based on the current data processed in S1, a CNN-SE-GRU hybrid model is constructed by combining a convolutional neural network (CNN), an attention mechanism (SE), and a GRU network.

[0011] S3, CEO Algorithm Optimization: The CEO algorithm is used to optimize the hyperparameters of the hybrid model constructed in S2, and a CEO-CNN-SE-GRU hybrid prediction model with a collaborative optimization mechanism is obtained to improve the accuracy of short-circuit current zero-crossing prediction.

[0012] S4. Model Training: Train the hybrid prediction model obtained in S3 using the training set, and test the trained model using the test set to obtain a stable CEO-CNN-SE-GRU hybrid prediction model.

[0013] S5. Prediction Execution: Call the trained hybrid prediction model, use the data in the validation set as the model input, realize multi-step prediction of short-circuit current, and predict the zero-crossing point of short-circuit current.

[0014] Preferably, S2 specifically includes the following:

[0015] Convolutional Neural Network (CNN) Feature Extraction: The time series of the obtained short-circuit nodes are transformed into a form suitable for convolutional operations through sequence folding layers to construct a two-dimensional input structure for the CNN. The first convolutional layer uses a 3*1 convolutional kernel to extract 64-channel features, capturing local patterns and short-term dependencies in the time series; the second convolutional layer further extracts 128-channel features to capture higher-level temporal features.

[0016] Attention Mechanism (SE) Feature Enhancement: The attention mechanism (SE) is used to perform global average pooling on the feature sequence output by the first convolutional layer of the convolutional neural network (CNN) to compress the spatial dimension; a bottleneck design of 64-32-128 nodes is constructed using two fully connected layers, and a non-linearity is introduced by using the ReLU activation function after the first fully connected layer; the channel importance weights are calculated by using the Sigmoid activation function.

[0017] Feature weighting: The convolutional neural network (CNN) is combined with the attention mechanism (SE) to initially extract local features of the short-circuit current fault waveform and perform feature weighting;

[0018] Format conversion: The weighted features are restored to a temporal structure by defolding, and the multidimensional features are compressed into a one-dimensional vector by flattening operation;

[0019] GRU timing modeling: Timing modeling is performed using the GRU dual-gating mechanism to capture long-term dependencies in fault waveforms and output predicted waveforms;

[0020] Fitness function calculation: The root mean square error (RMSE) is calculated using the actual waveform and the predicted waveform and used as the fitness function of the optimization algorithm.

[0021] Preferably, the feature weighting specifically includes the following:

[0022] The expressions for convolutional layers and activation functions are:

[0023]

[0024] Where l represents the number of grid layers; M jj This represents the set of positions covered by the sliding window of the convolutional kernel on the input time series; This represents the output of the i-th neuron in layer l-1; This represents the weight matrix of the l-th layer; This represents the bias term for the j-th neuron in the l-th layer; Indicates the convolution operation; This represents the weighted input of the j-th neuron in layer l; The output feature is represented by f(·), which represents the activation function.

[0025] The functional representation of the channel importance differential weighting process is as follows:

[0026]

[0027] in, U represents the global average pooling result of the j-th channel in layer l; l This represents a vector consisting of all channel descriptors; H and W represent the height and width of the feature. This represents the feature value of the j-th channel in layer l at position (h, w); This represents the output features of the first fully connected layer; Indicates channel attention weights; and These represent the weight matrix and bias term of the fully connected layer, respectively. This represents the recalibrated feature sequence.

[0028] Preferably, the input for the GRU timing modeling is:

[0029]

[0030] Where, x t This indicates the information being entered at the current moment; "flatten" indicates the flattened layer.

[0031] The functional representation of the modeling process is as follows:

[0032]

[0033] h t =tanh(W h [r t ☉h t-1 x t ]+g h (10)

[0034] h t =z t ☉h t-1 +(1-z t )☉h t (11)

[0035] Among them, h t-1 Indicates the hidden state at the previous time step; h t Indicates the hidden state passed to the next time step; h tt For candidate hidden states, the activation function used is tanh; t z t These are the reset gate and the update gate, respectively; σ represents the Sigmoid function, which is r t z t The activation function; W and g are the weights and bias matrices of each control gate in the GRU network, respectively; ⊙ represents element-wise multiplication;

[0036] After modeling, the output of the fully connected layer is calculated using the hidden states of the GRU network:

[0037] y=ω out h T +b out (12)

[0038] Where y represents the output of the fully connected layer; ω out b out Represents the output layer weight matrix and bias vector; h T This represents the hidden state at the last time step of the sequence.

[0039] Preferably, S3 specifically includes the following:

[0040] The CEO optimization algorithm is used to optimize the parameters of the GRU hidden layer node number, Adam learning rate, and L2 regularization coefficient to keep the model in an optimal state. The specific implementation process is as follows:

[0041] S3.1, Mutation Operation

[0042] The CEO algorithm utilizes the chaotic mapping in equation (13) to provide a mutation direction for each individual, and designs a search framework for mutation operators as shown in equation (14):

[0043]

[0044] Among them, X t+1 , These represent the current individual and the mutated individual, respectively; 'a' represents the search step size; and 'dt' represents the evolutionary direction generated by the mapping.

[0045] S3.2, Cross Operation

[0046] After the mutation operation generates new individuals, the experimental vector is generated through a binomial crossover mechanism. in Dim represents the optimization dimension; The generation process is as follows:

[0047]

[0048] Where j = 1, 2, ..., Dim; j rand rand represents a random integer in the interval [1, Dim]. j (0,1] represents a value uniformly distributed between 0 and 1 for each j; C r Indicates the crossover rate;

[0049] Generation process and Consistent;

[0050] S3.3 Selection Operation: The CEO algorithm uses a greedy criterion to select the best individual from the n trial vectors obtained after mutation and crossover operations. The function is expressed as follows:

[0051]

[0052] In the formula, These are the test vectors. The optimal test vector.

[0053] Preferably, the prediction sampling window for the zero-crossing point of the short-circuit current in S5 is 2ms, containing 20 short-circuit current values; the prediction window is 36ms, containing 360 short-circuit current values.

[0054] The present invention further protects a computer device, the computer device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the instruction, program, code set or instruction set being loaded and executed by the processor to implement the above-mentioned multi-step prediction method for short-circuit current zero-crossing point based on a double-break circuit breaker.

[0055] The present invention further protects a computer-readable storage medium, characterized in that the computer-readable storage medium stores at least one instruction, at least one program, code set or instruction set, wherein the instruction, program, code set or instruction set is loaded and executed by a processor to implement the above-mentioned multi-step prediction method for the zero-crossing point of short-circuit current based on a double-breakpoint circuit breaker.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] (1) High prediction accuracy: The prediction error of the short-circuit current zero crossing point is controlled within ±0.4ms, and the average absolute error is 0.0775ms;

[0058] (2) Short sampling time: Only 2ms sampling time is needed to predict the short-circuit current waveform for the subsequent 36ms;

[0059] (3) Good stability: It exhibits stability under different aperiodic component amplitudes, attenuation constants, periodic component amplitudes, and fault phases;

[0060] (4) Strong noise resistance: It can maintain high prediction accuracy in a noise environment of 20-50dB, with a maximum average prediction error of only 0.32ms. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings involved in the embodiments are now briefly described. Obviously, the drawings in the following description are merely illustrative of some embodiments of the present invention. For those skilled in the art, other forms of drawings can be constructed based on these drawings without creative effort.

[0062] Figure 1 This is a diagram of the CNN-SE-GRU model framework proposed in Embodiment 1 of the present invention;

[0063] Figure 2 This is a diagram of the CEO-optimized CNN-SE-GRU model training framework proposed in Embodiment 1 of the present invention;

[0064] Figure 3 This is a schematic diagram of the simulation prediction results of the CEO-CNN-SE-GRU model proposed in Embodiment 1 of the present invention;

[0065] Figure 4 This is a comparison chart of the multi-model prediction results proposed in Embodiment 2 of the present invention;

[0066] Figure 5 This is a statistical chart of the prediction error of the CEO-CNN-SE-GRU model proposed in Embodiment 2 of the present invention;

[0067] Figure 6 This is a diagram illustrating the error impact analysis of the CEO-CNN-SE-GRU model proposed in Embodiment 2 of the present invention.

[0068] Figure 7 This is a noise impact analysis diagram of the CEO-CNN-SE-GRU model proposed in Embodiment 2 of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0070] This invention proposes a multi-step prediction method for the zero-crossing point of short-circuit current based on a dual-breakpoint circuit breaker. It organically integrates the local feature extraction advantages of convolutional neural networks (CNNs), the key feature enhancement characteristics of attention mechanisms, the temporal modeling capabilities of GRU networks, and the chaotic evolutionary optimization algorithm (CEO) to construct a CEO-CNN-SE-GRU hybrid prediction model with a collaborative optimization mechanism. The following description, along with relevant figures and specific examples, illustrates the proposed multi-step prediction method for the zero-crossing point of short-circuit current based on a dual-breakpoint circuit breaker.

[0071] Example 1:

[0072] This invention proposes a multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker, comprising:

[0073] Step 1: Data preprocessing: Normalize the original short-circuit current data and scale it to the [0,1] interval while keeping the current waveform trend unchanged to obtain the time series of the short-circuit node. Then, divide the obtained data into training set, test set and validation set.

[0074] Step 2, Hybrid Model Construction: Based on the processed current data in S1, a CNN-SE-GRU hybrid model is constructed by combining a Convolutional Neural Network (CNN), an attention mechanism (SE), and a GRU network; please refer to [link to relevant documentation]. Figure 1 Specifically, it includes the following:

[0075] Convolutional Neural Network (CNN) Feature Extraction: The time series of the obtained short-circuit nodes are transformed into a form suitable for convolutional operations through sequence folding layers to construct a two-dimensional input structure for the CNN. The first convolutional layer uses a 3*1 convolutional kernel to extract 64-channel features, capturing local patterns and short-term dependencies in the time series; the second convolutional layer further extracts 128-channel features to capture higher-level temporal features.

[0076] Attention Mechanism (SE) Feature Enhancement: The attention mechanism (SE) is used to perform global average pooling on the feature sequence output by the first convolutional layer of the convolutional neural network (CNN) to compress the spatial dimension; a bottleneck design of 64-32-128 nodes is constructed using two fully connected layers, and a non-linearity is introduced by using the ReLU activation function after the first fully connected layer; the channel importance weights are calculated by using the Sigmoid activation function.

[0077] Feature weighting: A combination of convolutional neural networks (CNN) and attention mechanisms (SE) is used to initially extract local features of the short-circuit current fault waveform and then perform feature weighting; specifically including:

[0078] The expressions for convolutional layers and activation functions are:

[0079]

[0080] Where l represents the number of grid layers; M jj This represents the set of positions covered by the sliding window of the convolutional kernel on the input time series; This represents the output of the i-th neuron in layer l-1; This represents the weight matrix of the l-th layer; This represents the bias term for the j-th neuron in the l-th layer; Indicates the convolution operation; This represents the weighted input of the j-th neuron in layer l; The output feature is represented by f(·), which represents the activation function.

[0081] The functional representation of the channel importance differential weighting process is as follows:

[0082]

[0083] in, U represents the global average pooling result of the j-th channel in layer l;l This represents a vector consisting of all channel descriptors; H and W represent the height and width of the feature. This represents the feature value of the j-th channel in layer l at position (h, w); This represents the output features of the first fully connected layer; Indicates channel attention weights; and These represent the weight matrix and bias term of the fully connected layer, respectively. This represents the recalibrated feature sequence;

[0084] Format conversion: The weighted features are restored to a temporal structure by defolding, and the multidimensional features are compressed into a one-dimensional vector by flattening operation;

[0085] GRU Temporal Modeling: Temporal modeling is performed using the GRU dual-gating mechanism to capture long-term dependencies in the fault waveform. A fully connected layer maps the GRU output to a 360-dimensional vector, and a regression layer calculates the final 360-step prediction output. The input to GRU temporal modeling is:

[0086]

[0087] Where, x t This indicates the information to be entered at the current time;

[0088] The functional representation of the modeling process is as follows:

[0089]

[0090] h t =tanh(W h [r t ☉h t-1 x t ]+g h (10)

[0091] h t =z t ☉h t-1 +(1-z t )☉h t (11)

[0092] Among them, h t-1 Indicates the hidden state at the previous time step; h t Indicates the hidden state passed to the next time step; h tt For candidate hidden states, the activation function used is tanh; t z t These are the reset gate and the update gate, respectively; σ represents the Sigmoid function, which is r t z tThe activation function; W and g are the weights and bias matrices of each control gate in the GRU network, respectively;

[0093] After modeling, the output of the fully connected layer is calculated using the hidden states of the GRU network:

[0094] y=ω out h T +b out (12)

[0095] Where y represents the output of the fully connected layer; ω out b out Represents the output layer weight matrix and bias vector; h T This represents the hidden state at the last time step of the sequence;

[0096] Fitness function calculation: The root mean square error (RMSE) is calculated using the actual waveform and the predicted waveform and used as the fitness function of the optimization algorithm.

[0097] Step 3: CEO Algorithm Optimization: The CEO algorithm, based on the traditional DE algorithm framework, generates chaotic sequences through exponential discrete memristor mapping (EMD), exhibiting hyperchaotic characteristics at k=2.66. The CEO algorithm diversifies the search space, avoiding local optima and search stagnation problems. Optimizing the hyperparameters of the CNN-SE-GRU hybrid model avoids the tedious process of manual parameter tuning. The fitness function is chosen as the RMSE calculated by the CNN-SE-GRU model. The initial chaotic search range is defined as [-0.5, -0.25] and [0.5, 0.25]. The three optimized hyperparameters are the learning rate of the Adam optimizer, the number of hidden layer nodes in the GRU, and the L2 regularization coefficient. (See also...) Figure 2 The specific implementation process is as follows:

[0098] First, initialize the population and randomly generate 6 sets of solutions. Then, input each solution into the CNN-SE-GRU hybrid model to calculate the initial population fitness and find the current optimal hyperparameter combination.

[0099] Then, individuals in the population are randomly paired for iterative optimization. In each iteration, the parameter space is first mapped to the chaotic space, and then the chaotic sequence is generated using EMD mapping. The chaotic sequence is then mapped back to the parameter space through inverse mapping.

[0100]

[0101] The mutation operation is performed using the current mutation strategy (Equation 18) combined with the optimal mutation strategy (Equation 19), and then a binary crossover operation is used to generate test vectors. A CNN-SE-GRU hybrid model is constructed using each test vector, and the fitness value is calculated. The selection operation retains the better combination of the initial fitness and the fitness of the test vectors. The global optimal solution is updated after each iteration. When the fitness change is less than 1 × 10⁻⁶ after 50 consecutive iterations, the optimal solution is achieved. -8 The algorithm terminates when the maximum number of iterations is reached. A hybrid model of CEO-CNN-SE-GRU is constructed using the optimal combination of hyperparameters. CEO, by combining chaos theory and evolutionary algorithms, balances global exploration and local exploitation capabilities, effectively improving the accuracy of short-circuit current zero-crossing point prediction.

[0102] Step 4, Model Training: Train the hybrid prediction model obtained in S3 using the training set, and test the trained model using the test set to obtain a stable CEO-CNN-SE-GRU hybrid prediction model.

[0103] Step 5, Prediction Execution: The prediction sampling window is set to 2ms, containing 20 short-circuit current values; the prediction window is set to 36ms, containing 360 short-circuit current values; the trained hybrid prediction model is called, and the data in the validation set is used as the model input to realize multi-step prediction of short-circuit current, predicting the zero-crossing point of the short-circuit current (results are represented in Table 1 and...). Figure 3 (As shown).

[0104] Table 1. Zero-point and extreme value errors of the CEO-CNN-SE-GRU multi-step prediction model.

[0105]

[0106] Table 1 shows the statistical analysis of the zero-point and peak-point errors of the predicted and actual short-circuit current waveforms. As shown in Table 1, the zero-point error is largest at t3, with an absolute error of 0.17 ms; the error is smallest at t2, at only 0.01 ms; the average absolute error and average relative error at the zero points are 0.0775 ms and 0.6075%, respectively; the relative error at the extreme value P2 is 5.09%, and the error at the time of occurrence of the extreme value is within 0.1 ms.

[0107] Example 2:

[0108] Based on Example 1, but with some differences, a control experiment and a characterization experiment were designed to characterize the performance of the proposed multi-step prediction method for short-circuit current zero-crossing point based on a double-breakpoint circuit breaker and the constructed CEO-CNN-SE-GRU hybrid prediction model. The characterization results are as follows: Figure 4-7 As shown.

[0109] Figure 4 This is a comparison chart of prediction results from multiple models, based on... Figure 4 It can be seen that the CEO-CNN-SE-GRU hybrid prediction model constructed in this example has the highest degree of fit with the true value and the highest prediction accuracy.

[0110] Combination Figure 5-7 It can be seen that the multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker proposed in this invention:

[0111] 1) High prediction accuracy: The prediction error of the short-circuit current zero crossing point is controlled within ±0.4ms, and the average absolute error is 0.0775ms;

[0112] 2) Short sampling time: Only 2ms sampling time is needed to predict the short-circuit current waveform for the subsequent 36ms;

[0113] 3) Good stability: It exhibits stability under different aperiodic component amplitudes, attenuation constants, periodic component amplitudes, and fault phases;

[0114] 4) Strong noise resistance: It can maintain high prediction accuracy in a noise environment of 20-50dB, with a maximum average prediction error of only 0.32ms.

[0115] To verify the generalization performance of the model, 140 sets of short-circuit current fault waveforms were randomly generated for verification, with the overall error controlled within 0.4 ms. By studying the effects of changes in factors such as fault phase angle, amplitude of each component, DC component attenuation constant, and signal-to-noise ratio on the model prediction error, the generalization ability of the hybrid prediction model was verified.

[0116] It should be noted that in this invention patent, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0117] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker, characterized in that, Includes the following steps: S1. Data preprocessing: Normalize the raw short-circuit current data and scale it to the [0,1] interval while keeping the current waveform change trend unchanged. Obtain the time series of the short-circuit node and divide the obtained data into training set, test set and validation set. S2. Hybrid Model Construction: Based on the current data processed in S1, a CNN-SE-GRU hybrid model is constructed by combining convolutional neural networks, attention mechanisms, and GRU networks; specifically, it includes the following: Convolutional Neural Network Feature Extraction: The time series of the obtained short-circuit nodes are transformed into a form suitable for convolution operations through a sequence folding layer, constructing a two-dimensional input structure for the convolutional neural network. The first convolutional layer uses 3... The first convolutional kernel extracts 64-channel features to capture local patterns and short-term dependencies in the time series; the second convolutional layer further extracts 128-channel features to capture higher-level temporal features. Attention Mechanism Feature Enhancement: The attention mechanism is used to perform global average pooling on the feature sequence output by the first convolutional layer of the convolutional neural network to compress the spatial dimension; a bottleneck design of 64-32-128 nodes is constructed using two fully connected layers, and a ReLU activation function is used after the first fully connected layer to introduce non-linearity; the channel importance weights are calculated using the Sigmoid activation function. Feature weighting: Combining convolutional neural networks with attention mechanisms, local features of short-circuit current fault waveforms are initially extracted and then weighted. Format conversion: The weighted features are restored to a temporal structure by defolding, and the multidimensional features are compressed into a one-dimensional vector by flattening operation; GRU timing modeling: Timing modeling is performed using the GRU dual-gating mechanism to capture long-term dependencies in fault waveforms and output predicted waveforms; Fitness function calculation: The root mean square error (RMSE) is calculated using the actual waveform and the predicted waveform as the fitness function of the optimization algorithm; S3, CEO Algorithm Optimization: The CEO algorithm is used to optimize the hyperparameters of the hybrid model constructed in S2, and a CEO-CNN-SE-GRU hybrid prediction model with a collaborative optimization mechanism is obtained to improve the accuracy of short-circuit current zero-crossing prediction. S4. Model Training: Train the hybrid prediction model obtained in S3 using the training set, and test the trained model using the test set to obtain a stable CEO-CNN-SE-GRU hybrid prediction model. S5. Prediction Execution: Call the trained hybrid prediction model, use the data in the validation set as the model input, realize multi-step prediction of short-circuit current, and predict the zero-crossing point of short-circuit current.

2. The multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker according to claim 1, characterized in that, The feature weighting specifically includes the following: The expressions for convolutional layers and activation functions are: (1) (2) in, l Indicates the number of grid layers; M j This represents the set of positions covered by the sliding window of the convolutional kernel on the input time series; express l -1st floor i The output of each neuron; Indicates the first l Layer weight matrix; Indicates the first l Layer j Bias terms for each neuron; Indicates the convolution operation; express l Layer j Weighted inputs to each neuron; This represents the output features after processing by the activation function; Indicates the activation function; The functional representation of the channel importance differential weighting process is as follows: (3) (4) (5) (6) in, express l Layer j Global average pooling results for each channel; This represents a vector consisting of all channel descriptors; H and W Indicates the height and width of the feature; express l Layer j Each channel is located at ( h , w The eigenvalue at () location; This represents the output features of the first fully connected layer; Indicates channel attention weights; , and , These represent the weight matrix and bias term of the fully connected layer, respectively. This represents the recalibrated feature sequence.

3. The multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker according to claim 2, characterized in that, The input for the GRU time series modeling is: (7) in, x t This indicates the information to be entered at the current time; flatten Indicates a flattened layer; The functional representation of the modeling process is as follows: (8) (9) (10) (11) in, h t-1 This indicates the hidden state at the previous moment; h t This indicates the hidden state that will be passed on to the next time step; h tt For candidate hidden states, the activation function used is tanh; r t , z t These are the reset door and the update door, respectively. σ The Sigmoid function is... r t , z t The activation function; ⊙ represents element-wise multiplication; After modeling, the output of the fully connected layer is calculated using the hidden states of the GRU network: (12) in, y Indicates the output of the fully connected layer; This represents the output layer weight matrix and bias vector; h T This represents the hidden state at the last time step of the sequence.

4. The multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker according to claim 3, characterized in that, S3 specifically includes the following: The CEO optimization algorithm is used to optimize the parameters of the GRU hidden layer node number, Adam learning rate, and L2 regularization coefficient to keep the model in an optimal state. The specific implementation process is as follows: S3.1, Mutation Operation The CEO algorithm utilizes the chaotic mapping in equation (13) to provide a mutation direction for each individual, and designs a search framework for mutation operators as shown in equation (14): (13) (14) Among them, among them, , These represent the current individual and the mutated individual, respectively; 'a' represents the search step size. d t Indicates the direction of evolution generated by the mapping; S3.2, Cross Operation After the mutation operation generates new individuals, the experimental vector is generated through a binomial crossover mechanism. x trial , y trial ,in x trial =( x trial , x trial ,…, x trial ), y trial =( y trial , y trial ,…, y trial ), Dimh Indicates the optimization dimension; x trial The generation process is as follows: (15) in, j =1,2,…,Dim; j rand Represents a random integer in the interval [1, Dim]. rand j (0,1] represents each j Randomly select values ​​that are uniformly distributed between 0 and 1; C r Indicates the crossover rate; y trial Generation process and x trial Consistent; S3.3 Selection Operation: The CEO algorithm uses a greedy criterion to select the best individual from the n trial vectors obtained after mutation and crossover operations. The function is expressed as follows: (16) (17) In the formula, x trial , y trial These are the test vectors. x trial , y trial The optimal test vector in the process.

5. The multi-step prediction method for the zero-crossing point of short-circuit current based on a double-break circuit breaker according to claim 4, characterized in that, The prediction sampling window for the zero-crossing point of the short-circuit current described in S5 is 2 ms, containing 20 short-circuit current values; the prediction window is 36 ms, containing 360 short-circuit current values.

6. A computer device, characterized in that, The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the instruction, program, code set, or instruction set is loaded and executed by the processor to implement the multi-step prediction method for short-circuit current zero-crossing point based on a dual-breakpoint circuit breaker as described in any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, at least one program, code set, or instruction set, which is loaded and executed by a processor to implement the multi-step prediction method for short-circuit current zero-crossing point based on a dual-breakpoint circuit breaker as described in any one of claims 1-5.

Images