Intelligent detection method and device for opening and closing coil of circuit breaker

Abstract

The invention discloses a circuit breaker opening and closing coil intelligent detection method and device, relates to the technical field of circuit breaker detection, adopts a composite machine learning model to construct an intelligent detection model, can identify weaker and more complex early fault features, improves the detection accuracy and sensitivity, and improves the detection accuracy. And then training an intelligent detection model by adopting a logarithmic spiral exploration strategy, a vortex path exploration strategy based on information entropy and a fractional order balance exploration strategy based on a damping sawtooth wave oscillation factor, so that the algorithm can be effectively prevented from falling into local optimum, and when facing data of circuit breakers of different models and different working conditions, the algorithm can be accurately detected. The training of the intelligent detection model can be effectively realized, so that the intelligent detection accuracy of the opening and closing coil of the circuit breaker is ensured.

Description

Technical Field

[0001] This application relates to the field of circuit breaker testing technology, specifically to an intelligent testing method and device for circuit breaker opening and closing coils. Background Technology

[0002] Circuit breakers are crucial protection and control devices in power systems, and their reliability directly affects the safe and stable operation of the power grid. The opening and closing coils, as the core driving components of the circuit breaker's operating mechanism, directly determine whether the circuit breaker can reliably execute opening and closing commands. When faults such as inter-turn short circuits, core jamming, or coil aging occur, their electrical characteristics (such as current waveforms) will change significantly. If these potential faults are not detected in time, it may lead to the circuit breaker refusing to operate or operating malfunctions, causing large-scale power outages or even system collapse. Traditional circuit breaker coil testing methods mainly rely on periodic offline tests, such as DC resistance testing and insulation resistance testing. These methods have the following shortcomings: 1) Offline nature: Requires the circuit breaker to be taken out of service, affecting power supply reliability; 2) Lag: Can only detect faults after they occur or have developed to a certain extent, failing to provide early warning; 3) Limited information: Can only reflect static resistance or insulation status, unable to capture transient characteristics during dynamic operation, and insensitive to early, weak faults.

[0003] With the development of online monitoring technology, fault diagnosis has become possible by collecting transient electrical data such as current and voltage of the opening and closing coils. Existing signal processing-based methods (such as wavelet transform and empirical mode decomposition) and simple machine learning methods (such as support vector machines and backpropagation neural networks) have improved diagnostic capabilities to some extent. However, these methods still have the following problems: the coil current waveform is a complex non-stationary and nonlinear signal containing rich spatiotemporal information. Traditional methods struggle to fully and automatically extract these deep features, resulting in limited diagnostic accuracy. While deep learning models have strong feature extraction capabilities, they have numerous parameters and a complex training process. Traditional optimization algorithms such as gradient descent are prone to getting trapped in local optima and have slow convergence speeds, making it difficult to find globally optimal model parameters, thus affecting the model's final performance and generalization ability. Summary of the Invention

[0004] The purpose of this application is to provide an intelligent detection method and device for circuit breaker opening and closing coils, which solves the problems of limited feature extraction capability and difficulty in model optimization in the prior art.

[0005] This application is achieved through the following technical solution:

[0006] The first aspect of this application provides an intelligent detection method for the opening and closing coils of a circuit breaker, including:

[0007] Historical electrical operating data of the circuit breaker's opening and closing coils are collected under various operating conditions, and training sample data is constructed based on the historical electrical operating data and its corresponding operating conditions; the operating conditions include normal conditions and abnormal conditions.

[0008] An intelligent detection model is constructed using a composite machine learning model, and the parameters of the intelligent detection model are initialized and encoded to obtain multiple encoded individuals;

[0009] Based on the training sample data, the coded individuals are trained using a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factor, in order to obtain the target coded individuals.

[0010] The target encoded individual is decoded to obtain the final parameters of the intelligent detection model, and the final parameters are applied to the intelligent detection model to obtain the trained intelligent detection model.

[0011] The electrical operation data of the circuit breaker's opening and closing coils to be identified is collected, and the trained intelligent detection model is used to detect the electrical operation data to be identified, so as to obtain the intelligent detection results of the circuit breaker's opening and closing coils.

[0012] In one possible design approach, historical electrical operating data of the circuit breaker's opening and closing coils are collected under various operating conditions, and training sample data is constructed based on the historical electrical operating data and its corresponding operating conditions, including:

[0013] Based on a preset data sampling frequency, historical electrical operation data of the circuit breaker's opening and closing coils are collected under various operating conditions; the historical electrical operation data includes current data and / or voltage data; each set of historical electrical operation data contains the same number of data points.

[0014] The historical electrical operation data is preprocessed to obtain preprocessed historical electrical operation data;

[0015] Sample data is constructed using the preprocessed historical electrical operation data, and sample labels are constructed using the operation status corresponding to the historical electrical operation data.

[0016] The sample data is associated with its corresponding sample label as training sample data, and multiple different training sample data are obtained.

[0017] In one possible design approach, a composite machine learning model is used to construct an intelligent detection model, and the parameters of the intelligent detection model are initialized and encoded to obtain multiple encoded individuals, including:

[0018] A composite machine learning model is constructed, consisting of a sequentially connected input layer, convolutional layer, pooling layer, BiGRU layer, attention mechanism layer, fully connected layer, and LSTM layer, to obtain an intelligent detection model.

[0019] The parameters of the intelligent detection model are randomly initialized between the upper limit and the lower limit, and the randomly initialized parameters are encoded into vectors to obtain encoded individuals. Multiple different encoded individuals are obtained repeatedly.

[0020] In one possible design approach, based on the training sample data, the encoded individual is trained using a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factors to obtain the target encoded individual, including:

[0021] Based on the training sample data, the fitness of the encoded individuals is obtained, and the encoded individual with the highest fitness is determined as the optimal individual;

[0022] Based on the optimal individual, a logarithmic spiral exploration strategy is used to guide the encoded individuals to obtain the guided encoded individuals;

[0023] A vortex path exploration strategy based on information entropy is adopted to conduct uncertain exploration of the coded individuals after guidance, and to obtain the coded individuals after uncertain exploration;

[0024] A fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factor is adopted to conduct equilibrium exploration on the encoded individuals after uncertain exploration, and to obtain the encoded individuals after equilibrium exploration.

[0025] Determine if the total number of iterations is greater than the preset maximum number of iterations. If so, obtain the target encoding individual based on the encoding individuals after the balancing exploration. Otherwise, proceed to the next training process based on the encoding individuals after the balancing exploration.

[0026] In one possible design approach, based on the optimal individual, a logarithmic spiral exploration strategy is used to guide the encoded individuals, obtaining the guided encoded individuals, including:

[0027] Based on the aforementioned optimal individual, the guiding information is obtained using a logarithmic spiral function as follows:

[0028] ;

[0029] in, Let i represent the i-th coded individual, where i = 1, 2, ..., N, and N represents the total number of coded individuals. Let represent the optimal individual, and t represent the current total number of iterations. express The corresponding guidance information, Represents the natural constant. Represents pi (π). Let represent a random number between (0, 1), and b represent the spiral shape constant;

[0030] The encoded individual obtained after guidance based on the guidance information is:

[0031] ;

[0032] in, This represents the encoded individual after the i-th guide.

[0033] In one possible design approach, an information entropy-based vortex path exploration strategy is employed to perform uncertain exploration on the coded individuals after guidance, obtaining the coded individuals after uncertain exploration, including:

[0034] The information entropy corresponding to the coded individual after obtaining the guidance is:

[0035] ;

[0036] ;

[0037] in, Represents information entropy. Let represent the probability parameter of information entropy occurrence for the encoded individual after the j-th guidance, where j = 1, 2, ..., N, and N represents the total number of encoded individuals. Represents the logarithmic function. This represents the fitness of the encoded individual after the j-th cipher.

[0038] Based on the information entropy, the information selection control factor is obtained as follows:

[0039] ;

[0040] in, This indicates the information selection control factor;

[0041] The dynamic step size control factor is obtained as follows:

[0042] ;

[0043] in, This represents the dynamic step size control factor. This represents the minimum value of the dynamic step size control factor. Represents the dynamic equilibrium factor. This represents the maximum number of iterations, and t represents the current total number of iterations. This represents the minimum fitness of the encoded individual after guidance. This represents the maximum fitness of the encoded individual after guidance;

[0044] Based on the information selection control factor and the dynamic step size control factor, an uncertain exploration is performed on the coded individuals after guidance, and the coded individuals after the uncertain exploration are obtained as follows:

[0045] ;

[0046] in, This represents the encoded individual after the j-th guide. This represents the encoded individual after the j-th uncertain exploration. Represents the optimal individual. This represents the encoded individual following the (j-1)th guide. This represents the encoded individual after the (j+1)th guide. Indicates the first learning factor. Indicates the second learning factor. Represents a random number between (0, 1). This represents a random number between (0, 1).

[0047] In one possible design approach, a fractional-order equilibrium exploration strategy based on a damped sawtooth wave oscillation factor is employed to perform equilibrium exploration on the encoded individuals after uncertain exploration, thereby obtaining the encoded individuals after equilibrium exploration, including...

[0048] The damped sawtooth wave oscillation factor is obtained as follows:

[0049]

[0050] in, This represents the damped sawtooth wave oscillation factor. Represents a random number between (0, 1). Represents a random number between (0,1). Represents a random number that follows a Gaussian distribution. Represents the sine function. Represents the cosine function. This represents the maximum number of iterations, and t represents the current total number of iterations. Indicates the control factor for the position of the oscillation peak;

[0051] Determine if the value of t / T is less than the damped sawtooth wave oscillation factor. If so, perform fractional-order local development on the coded individual after uncertain exploration to obtain the coded individual after balanced exploration. Otherwise, perform fractional-order global development on the coded individual after uncertain exploration to obtain the coded individual after balanced exploration.

[0052] In one possible design approach, fractional-order local development is performed on the encoded individuals after uncertain exploration to obtain encoded individuals after balanced exploration, including:

[0053] Obtain the average Euclidean distance for the encoded individuals after uncertain exploration, and determine the local fractional order based on the average Euclidean distance:

[0054]

[0055]

[0056] in, Let f denote the local fractional order, f denote the evolution factor, and e denote the natural constant. This represents the average Euclidean distance from the best individual to all other individuals. This represents the minimum value among all mean Euclidean distances. This represents the maximum value among all average Euclidean distances;

[0057] Based on the local fractional order, the exploration speed is obtained as follows:

[0058]

[0059] in, This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the t-th iteration. This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the (t+1)-th iteration. This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the (t-1)-th iteration. This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the (t-2)-th iteration. This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the (t-3)-th iteration; Indicates inertia weight, Represents a random number between (0,1). Represents the natural constant. This represents the encoded individual after the k-th uncertain exploration, where k = 1, 2, ..., N, and N represents the total number of encoded individuals. Represented as Other encoded individuals following random matching and uncertain exploration. express and The Euclidean distance between them Indicates the control coefficient for the degree of information interaction;

[0060] Based on the exploration speed, the encoded individual obtained after balanced exploration is:

[0061]

[0062] in, This represents the encoded individual after the k-th equilibrium exploration.

[0063] In one possible design approach, fractional-order global development is performed on the encoded individuals after uncertain exploration to obtain encoded individuals after balanced exploration, including:

[0064] For individual code after uncertain exploration, the global development parameters are obtained as follows:

[0065]

[0066] in, Represents global development parameters. Indicates the order of the global fraction. Let represent the d-th dimension parameter of the encoded individual after the nth uncertain exploration, where n = 1, 2, ..., N, N represents the total number of encoded individuals, and d = 1, 2, ..., D, D represents the total dimension of parameters in the encoded individuals. This represents the upper limit of the d-th dimension parameter. The maximum number of iterations is represented by t, and the current total number of iterations is represented by t.

[0067] Based on the global development parameters, the encoded individuals obtained after balanced exploration are:

[0068]

[0069] in, This represents the encoded individual after the k-th equilibrium exploration, and cos represents the cosine function. Represents pi (π). This represents the global development control coefficient, where s represents the global fractional order. The d-th dimension parameter represents the optimal individual.

[0070] Based on the same inventive concept, a second aspect of this application provides an intelligent detection device for the opening and closing coils of a circuit breaker, comprising:

[0071] The sample data acquisition module is used to collect historical electrical operation data of the circuit breaker's opening and closing coils under various operating conditions, and to construct training sample data based on the historical electrical operation data and its corresponding operating conditions; the operating conditions include normal conditions and abnormal conditions.

[0072] The model initialization module is used to construct an intelligent detection model using a composite machine learning model, and to initialize and encode the parameters of the intelligent detection model to obtain multiple encoded individuals;

[0073] The model optimization module is used to train the encoded individuals using a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factors, in order to obtain the target encoded individuals.

[0074] The parameter application module is used to decode the target encoded individual, obtain the final parameters of the intelligent detection model, and apply the final parameters to the intelligent detection model to obtain the trained intelligent detection model.

[0075] The intelligent detection module is used to collect electrical operation data of the circuit breaker's opening and closing coils to be identified, and to use the trained intelligent detection model to detect the electrical operation data to be identified, thereby obtaining the intelligent detection results of the circuit breaker's opening and closing coils.

[0076] Compared with the prior art, this application has the following advantages and beneficial effects:

[0077] This application provides an intelligent detection method and device for circuit breaker opening and closing coils. It employs a composite machine learning model to construct an intelligent detection model, which can identify weaker and more complex early fault characteristics, improving detection accuracy and sensitivity. Then, it uses a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order balance exploration strategy based on damped sawtooth wave oscillation factors to train the intelligent detection model. This effectively avoids the algorithm getting trapped in local optima. When faced with circuit breaker data of different models and operating conditions, it can effectively train the intelligent detection model, thereby ensuring the accuracy of intelligent detection of circuit breaker opening and closing coils. Attached Figure Description

[0078] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this application and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0079] Figure 1 A flowchart illustrating an intelligent detection method for circuit breaker opening and closing coils provided in this application embodiment;

[0080] Figure 2 A schematic diagram of the structure of an intelligent detection device for circuit breaker opening and closing coils provided in this application embodiment;

[0081] Figure 3 A schematic diagram of the structure of the damped sawtooth wave oscillation factor provided in the embodiments of this application;

[0082] The attached diagram shows the markings and corresponding component names:

[0083] 201-Sample data acquisition module, 202-Model initialization module, 203-Model optimization module, 204-Parameter application module, 205-Intelligent detection module. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0085] like Figure 1 As shown in the figure, this application provides an intelligent detection method for circuit breaker opening and closing coils, including:

[0086] S101. Collect historical electrical operation data of the circuit breaker's opening and closing coils under various operating conditions, and construct training sample data based on the historical electrical operation data and its corresponding operating conditions; the operating conditions include normal conditions and abnormal conditions.

[0087] At the substation site, high-precision current transformers and voltage sensors can be used to collect online current data of various circuit breaker models under different operating conditions at a sampling frequency of no less than 10kHz. These operating conditions include: normal operation, inter-turn short circuit, core jamming, coil aging, and low control voltage. The collected raw current data can be preprocessed, including: data denoising: using wavelet threshold denoising to filter out high-frequency electromagnetic interference noise; data normalization: normalizing all current data to the [0, 1] interval to eliminate dimensional influence; and data alignment and truncation: using the moment the opening / closing command is issued as a reference, a fixed length (e.g., 200ms) of current waveform is truncated to ensure that each sample data has consistent dimensions. The processed current waveform data is used as sample data, and its corresponding operating condition (normal / abnormal) is used as sample label to construct a large-scale training sample set.

[0088] S102. Construct an intelligent detection model using a composite machine learning model, and initialize and encode the parameters of the intelligent detection model to obtain multiple encoded individuals;

[0089] For example, an intelligent detection model can be built using a CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) model or a CNN-BP model. Then, all trainable parameters of this composite model (including convolutional kernel weights, biases, GRU / LSTM unit weights, etc.) are randomly initialized, and each set of parameters is encoded into a high-dimensional vector, called an encoded individual. M such encoded individuals (e.g., 50) are randomly generated to form the initial population.

[0090] S103. Based on the training sample data, the coded individuals are trained using a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factor to obtain the target coded individual.

[0091] By employing a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factors to train the intelligent detection model, the algorithm can effectively avoid getting trapped in local optima. When faced with circuit breaker data of different models and operating conditions, the intelligent detection model can be effectively trained, thereby ensuring the accuracy of intelligent detection of circuit breaker opening and closing coils.

[0092] S104. Decode the target encoded individual to obtain the final parameters of the intelligent detection model, and apply the final parameters to the intelligent detection model to obtain the trained intelligent detection model.

[0093] Decode the target encoded individual to obtain the final weights and bias parameters of the intelligent detection model, then solidify these parameters into the model to obtain the final trained model.

[0094] S105. Collect the electrical operation data to be identified of the circuit breaker's opening and closing coils, and use the trained intelligent detection model to detect the electrical operation data to be identified, thereby obtaining the intelligent detection results of the circuit breaker's opening and closing coils.

[0095] In practical applications, the current data of the opening and closing coils of the circuit breaker under test is collected in real time. After the same preprocessing as during training, the data is input into the final model. The model outputs a diagnostic result, such as normal, inter-turn short circuit, core jamming, etc., and gives the corresponding confidence level, realizing online, real-time and intelligent detection of coil status.

[0096] In one possible design approach, historical electrical operating data of the circuit breaker's opening and closing coils are collected under various operating conditions, and training sample data is constructed based on the historical electrical operating data and its corresponding operating conditions, including:

[0097] Based on a preset data sampling frequency, historical electrical operation data of the circuit breaker's opening and closing coils are collected under various operating conditions; the historical electrical operation data includes current data and / or voltage data; each set of historical electrical operation data contains the same number of data points.

[0098] The historical electrical operation data is preprocessed to obtain preprocessed historical electrical operation data;

[0099] Sample data is constructed using the preprocessed historical electrical operation data, and sample labels are constructed using the operation status corresponding to the historical electrical operation data.

[0100] The sample data is associated with its corresponding sample label as training sample data, and multiple different training sample data are obtained.

[0101] Optionally, feature extraction can be performed on historical electrical operation data during the sample data construction process. For example, if the historical electrical operation data is current data, the original coil current signal can be decomposed using the VMD (Variational Modal Decomposition) algorithm, and the correlation coefficient of each decomposed IMF (Intrinsic Mode Function) component can be calculated. Then, a threshold is calculated based on the correlation coefficient selection principle and compared with the calculation results. IMF components with calculation results greater than the threshold are considered to be IMF components with strong correlation to the original signal and are called effective components, which are directly retained. Components with calculation results less than the threshold are considered to be components with weak correlation to the original signal, but still have some correlation and cannot be directly discarded. The wavelet threshold has a significant advantage in processing local signal features, and the weakly correlated IMF components are processed a second time. Finally, the directly retained IMF components and the secondary processed IMF components are reconstructed to obtain the final denoising result, which can then be used as sample data.

[0102] In one possible design approach, a composite machine learning model is used to construct an intelligent detection model, and the parameters of the intelligent detection model are initialized and encoded to obtain multiple encoded individuals, including:

[0103] A composite machine learning model is constructed, consisting of a sequentially connected input layer, convolutional layer, pooling layer, BiGRU (Bidirectional Gated Recurrent Unit) layer, attention mechanism layer, fully connected layer, and LSTM layer, to obtain an intelligent detection model.

[0104] The parameters of the intelligent detection model are randomly initialized between the upper limit and the lower limit, and the randomly initialized parameters are encoded into vectors to obtain encoded individuals. Multiple different encoded individuals are obtained repeatedly.

[0105] The input layer receives normalized current waveform data.

[0106] Convolutional layers + pooling layers: One-dimensional convolutional kernels (e.g., kernel_size=5) are used to extract local morphological features of the current waveform, such as the peak current and the rate of change of current. Pooling layers (e.g., MaxPooling) are used to reduce dimensionality and enhance the robustness of the features.

[0107] BiGRU layer: a bidirectional gated cyclic unit that simultaneously captures the time-series dependence of current waveforms from both forward and backward directions, learning dynamic evolution patterns.

[0108] Attention mechanism layer: The time series features output by BiGRU are weighted and automatically focused on the most critical time segments for fault diagnosis (such as the core start-up stage and the latching action stage), suppressing interference from irrelevant information.

[0109] Fully connected layer: integrates the weighted feature vectors.

[0110] LSTM layer: The long short-term memory network is used again to perform deep temporal modeling on the integrated features, capturing more complex long-term dependencies and improving the model's sensitivity to weak and slowly changing faults.

[0111] Output layer: The Softmax activation function is used, and the probability distribution of the output coil in various states is shown.

[0112] In one possible design approach, based on the training sample data, the encoded individual is trained using a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factors to obtain the target encoded individual, including:

[0113] Based on the training sample data, the fitness of the encoded individuals is obtained, and the encoded individual with the highest fitness is determined as the optimal individual;

[0114] For example, we can take sample data as input and sample labels as expected output to obtain the root mean square loss function corresponding to the encoded individual. Then, we can take the reciprocal or a negative number of the root mean square loss function to obtain the fitness of the encoded individual. If we take the reciprocal, we can set a non-zero term in the denominator (such as 0.00010) to avoid the denominator being zero.

[0115] Based on the optimal individual, a logarithmic spiral exploration strategy is used to guide the encoded individuals to obtain the guided encoded individuals;

[0116] A vortex path exploration strategy based on information entropy is adopted to conduct uncertain exploration of the coded individuals after guidance, and to obtain the coded individuals after uncertain exploration;

[0117] A fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factor is adopted to conduct equilibrium exploration on the encoded individuals after uncertain exploration, and to obtain the encoded individuals after equilibrium exploration.

[0118] Determine if the total number of iterations is greater than the preset maximum number of iterations. If so, obtain the target encoding individual based on the encoding individuals after the balancing exploration. Otherwise, proceed to the next training process based on the encoding individuals after the balancing exploration.

[0119] Optionally, after each iteration of the encoded individual or after a change, limit-breaking processing can be performed on the encoded individual to ensure that the parameters are always within the valid range.

[0120] Existing optimization algorithms (such as particle swarm optimization and gradient descent) are prone to getting trapped in local optima during the training of intelligent detection models, and are also difficult to apply to data from different circuit breakers. As a result, the trained intelligent detection model cannot accurately achieve intelligent diagnosis. Therefore, this application provides a new training algorithm that uses a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factors to train coded individuals. This solves the problem of existing technologies getting trapped in local optima, improves training adaptability, and enables the intelligent detection model to adapt to data from various circuit breakers. It can find the optimal parameter combination in various scenarios and improve the accuracy of intelligent detection.

[0121] In one possible design approach, based on the optimal individual, a logarithmic spiral exploration strategy is used to guide the encoded individuals, obtaining the guided encoded individuals, including:

[0122] Based on the aforementioned optimal individual, the guiding information is obtained using a logarithmic spiral function as follows:

[0123] ;

[0124] in, Let i represent the i-th coded individual, where i = 1, 2, ..., N, and N represents the total number of coded individuals. Let represent the optimal individual, and t represent the current total number of iterations. express The corresponding guidance information, Represents the natural constant. Represents pi (π). This represents a random number between (0,1), and b represents the spiral shape constant (which can be set to 1 or 1.5).

[0125] The encoded individual obtained after guidance based on the guidance information is:

[0126] ;

[0127] in, This represents the encoded individual after the i-th guide.

[0128] By employing a logarithmic spiral exploration strategy to guide the encoded individual, it can explore the neighborhood of its own position in the solution space and move towards the known optimal position in a spiral manner. This approach ensures exploration accuracy while reducing the probability of getting trapped in local optima to some extent.

[0129] In one possible design approach, an information entropy-based vortex path exploration strategy is employed to perform uncertain exploration on the coded individuals after guidance, obtaining the coded individuals after uncertain exploration, including:

[0130] The information entropy corresponding to the coded individual after obtaining the guidance is:

[0131] ;

[0132] ;

[0133] in, Represents information entropy. Let represent the probability parameter of information entropy occurrence for the encoded individual after the j-th guidance, where j = 1, 2, ..., N, and N represents the total number of encoded individuals. Represents the logarithmic function. This represents the fitness of the encoded individual after the j-th cipher.

[0134] Based on the information entropy, the information selection control factor is obtained as follows:

[0135] ;

[0136] in, This indicates the information selection control factor;

[0137] The dynamic step size control factor is obtained as follows:

[0138] ;

[0139] in, This represents the dynamic step size control factor. This represents the minimum value of the dynamic step size control factor (which can be set to a constant term between 0.01 and 0.1). This represents the dynamic equilibrium factor (which can be set to a constant between (0,1), such as 0.5). This represents the maximum number of iterations, and t represents the current total number of iterations. This represents the minimum fitness of the encoded individual after guidance. This represents the maximum fitness of the encoded individual after guidance;

[0140] Based on the information selection control factor and the dynamic step size control factor, an uncertain exploration is performed on the coded individuals after guidance, and the coded individuals after the uncertain exploration are obtained as follows:

[0141] ;

[0142] in, This represents the encoded individual after the j-th guide. This represents the encoded individual after the j-th uncertain exploration. Represents the optimal individual. This represents the encoded individual following the (j-1)th guide. This represents the encoded individual after the (j+1)th guide. This represents the first learning factor (which can be set to 1 or 2). This represents the second learning factor (which can be set to 0.75 or 1.5). Represents a random number between (0,1). This represents a random number between (0, 1).

[0143] The vortex path exploration strategy based on information entropy significantly enhances the global search capability and the ability to escape local optima by introducing information entropy feedback, thus solving the problem of premature convergence in traditional optimization algorithms. It simultaneously generates a centripetal force to learn from the best individual and a tangential force to draw on information from neighboring individuals; the combination of these two forms a vortex-like search path, effectively enhancing the algorithm's ability to escape local optima.

[0144] In one possible design approach, a fractional-order equilibrium exploration strategy based on a damped sawtooth wave oscillation factor is employed to perform equilibrium exploration on the encoded individuals after uncertain exploration, thereby obtaining the encoded individuals after equilibrium exploration, including...

[0145] The damped sawtooth wave oscillation factor is obtained as follows:

[0146]

[0147] in, This represents the damped sawtooth wave oscillation factor. Represents a random number between (0,1). Represents a random number between (0,1). Represents a random number that follows a Gaussian distribution. Represents the sine function. Represents the cosine function. This represents the maximum number of iterations, and t represents the current total number of iterations. Indicates the control factor for the position of the oscillation peak;

[0148] Determine if the value of t / T is less than the damped sawtooth wave oscillation factor (since the value of t / T cannot be negative, we can determine if the value of t / T is less than the absolute value of the damped sawtooth wave oscillation factor, thus making the fluctuations in the algorithm more intense and improving the global exploration capability). If so, perform fractional-order local development on the encoded individual after uncertain exploration to obtain the encoded individual after balanced exploration; otherwise, perform fractional-order global development on the encoded individual after uncertain exploration to obtain the encoded individual after balanced exploration.

[0149] like Figure 3 As shown, the damped sawtooth wave oscillation factor fluctuates continuously at a certain equilibrium position with the number of iterations. The oscillation shape is sawtooth-like, and the amplitude of the oscillation gradually decreases, thus automatically balancing local exploitation and global exploitation in the search process. Furthermore, as the value of t / T increases in the later stages of the algorithm, it can gradually shift to local exploitation.

[0150] In one possible design approach, fractional-order local development is performed on the encoded individuals after uncertain exploration to obtain encoded individuals after balanced exploration, including:

[0151] Obtain the average Euclidean distance for the encoded individuals after uncertain exploration, and determine the local fractional order based on the average Euclidean distance:

[0152]

[0153]

[0154] in, Let f denote the local fractional order, f denote the evolution factor, and e denote the natural constant. This represents the average Euclidean distance from the best individual to all other individuals. This represents the minimum value among all mean Euclidean distances. This represents the maximum value among all average Euclidean distances;

[0155] For example, for any coded individual after uncertain exploration, we can obtain the Euclidean distance between this coded individual and all other coded individuals after uncertain exploration, and take the average to obtain the average Euclidean distance corresponding to the coded individual after uncertain exploration. The principle for obtaining the average Euclidean distance corresponding to the optimal individual is the same.

[0156] Based on the local fractional order, the exploration speed is obtained as follows:

[0157]

[0158] in, This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the t-th iteration. This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the (t+1)-th iteration. This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the (t-1)-th iteration. This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the (t-2)-th iteration. This represents the exploration speed of the encoded individual after the k-th uncertain exploration in the (t-3)-th iteration; This represents the inertia weight (which can be set to values ​​such as 0.9, 0.85, etc.). Represents a random number between (0,1). Represents the natural constant. This represents the encoded individual after the k-th uncertain exploration, where k = 1, 2, ..., N, and N represents the total number of encoded individuals. Represented as Other encoded individuals following random matching and uncertain exploration. express and The Euclidean distance between them Indicates the control coefficient for the degree of information interaction;

[0159] It is worth noting that, without , , , If the condition is met, the corresponding item will be set to zero to ensure the algorithm can run normally.

[0160] Based on the exploration speed, the encoded individual obtained after balanced exploration is:

[0161]

[0162] in, This represents the encoded individual after the k-th equilibrium exploration.

[0163] The local fractional order is dynamically adjusted based on the distance between the optimal individual and the population. If the optimal individual is far from the population, a large-scale exploration is conducted to increase the exploration speed; if it is close, a fine-grained search is performed. The update of the exploration speed not only considers the current and historical speeds but also introduces an exponential decay term for the distance between individuals, enhancing the effectiveness of information interaction and ensuring the diversity and convergence of the algorithm during training.

[0164] In one possible design approach, fractional-order global development is performed on the encoded individuals after uncertain exploration to obtain encoded individuals after balanced exploration, including:

[0165] For individual code after uncertain exploration, the global development parameters are obtained as follows:

[0166]

[0167] in, Represents global development parameters. Indicates the global fractional order (can be set to 3 or 4). Let represent the d-th dimension parameter of the encoded individual after the nth uncertain exploration, where n = 1, 2, ..., N, N represents the total number of encoded individuals, and d = 1, 2, ..., D, D represents the total dimension of parameters in the encoded individuals. This represents the upper limit of the d-th dimension parameter. The maximum number of iterations is represented by t, and the current total number of iterations is represented by t.

[0168] Based on the global development parameters, the encoded individuals obtained after balanced exploration are:

[0169]

[0170] in, This represents the encoded individual after the k-th equilibrium exploration, and cos represents the cosine function. Represents pi (π). This represents the global development control coefficient (which can be set to 0.4 or 0.5), and s represents the global fractional order. The d-th dimension parameter represents the optimal individual.

[0171] This fractional-order global development introduces fractional-order theory, defines global development parameters and global fractional order, and combines optimal individual information with random perturbation in the update formula. Furthermore, the perturbation range shrinks as the iteration progresses, ensuring that the algorithm can still explore new and potential solution space regions in the later stages.

[0172] The training algorithm provided in this application can effectively improve the training effect of the intelligent detection model, thereby enabling the intelligent detection model to have higher detection accuracy and ensuring the stable operation of the circuit breaker.

[0173] This application's embodiments form a complete automated closed loop from data acquisition, feature extraction, model optimization to final diagnosis, enabling intelligent circuit breaker detection. It eliminates reliance on manual feature engineering and expert experience, automatically learning fault modes from massive amounts of data. Combined with efficient optimization algorithms, it makes training complex deep models possible, paving the way for the large-scale engineering application of this technology in power systems and possessing significant practical implications for improving the intelligent operation and maintenance level of the power grid.

[0174] like Figure 2 As shown, based on the same inventive concept, this application provides an intelligent detection device for circuit breaker opening and closing coils, comprising:

[0175] The sample data acquisition module 201 is used to collect historical electrical operation data of the circuit breaker's opening and closing coils under various operating conditions, and to construct training sample data based on the historical electrical operation data and its corresponding operating conditions; the operating conditions include normal conditions and abnormal conditions.

[0176] The model initialization module 202 is used to construct an intelligent detection model using a composite machine learning model, and to initialize and encode the parameters of the intelligent detection model to obtain multiple encoded individuals;

[0177] The model optimization module 203 is used to train the encoded individual using a logarithmic spiral exploration strategy, an information entropy-based vortex path exploration strategy, and a fractional-order equilibrium exploration strategy based on damped sawtooth wave oscillation factor, so as to obtain the target encoded individual.

[0178] The parameter application module 204 is used to decode the target encoded individual, obtain the final parameters of the intelligent detection model, and apply the final parameters to the intelligent detection model to obtain the trained intelligent detection model.

[0179] The intelligent detection module 205 is used to collect the electrical operation data to be identified of the circuit breaker's opening and closing coils, and to use the trained intelligent detection model to detect the electrical operation data to be identified, thereby obtaining the intelligent detection results of the circuit breaker's opening and closing coils.

[0180] The intelligent detection device for circuit breaker opening and closing coils provided in this application embodiment can perform the above-described method embodiment. Its specific principle and beneficial effects are similar, and will not be repeated here.

[0181] Based on the same inventive concept, this application also provides an intelligent detection device for circuit breaker opening and closing coils, including a processor and a memory; the memory and the processor are interconnected via a bus.

[0182] The memory stores computer-executed instructions;

[0183] The processor executes computer execution instructions stored in the memory, causing the processor to execute an intelligent detection method for circuit breaker opening and closing coils as described in any embodiment of this application.

[0184] For specific examples, memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). Furthermore, the processor may include a main processor and coprocessors. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0185] This application provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed by a processor, they are used to implement the intelligent detection method for circuit breaker opening and closing coils described in any of the above embodiments.

[0186] This application embodiment can also provide a computer program product, including a computer program that, when executed by a processor, implements the intelligent detection method for circuit breaker opening and closing coils described in any of the above embodiments.

[0187] All or part of the steps in the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable memory. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned memory (storage medium) includes: read-only memory (ROM), RAM, flash memory, hard disk, solid-state drive, magnetic tape, floppy disk, optical disk, and any combination thereof.

[0188] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processing unit of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0189] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0190] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0191] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0192] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A circuit breaker opening and closing coil intelligent detection method, characterized in that, The application comprises the following steps: Collecting historical electrical operation data of the circuit breaker opening and closing coil under various operating states, and constructing training sample data according to the historical electrical operation data and the corresponding operating states; the operating states include normal states and abnormal states; Using a composite machine learning model to construct an intelligent detection model, and initializing the parameters of the intelligent detection model to obtain multiple coded individuals; Based on the training sample data, the coded individuals are trained using a logarithmic spiral exploration strategy, a vortex path exploration strategy based on information entropy, and a fractional order balance exploration strategy based on a damping sawtooth wave oscillation factor to obtain a target coded individual; Decoding the target coded individual to obtain the final parameters of the intelligent detection model, and applying the final parameters to the intelligent detection model to obtain the trained intelligent detection model; Collecting the to-be-identified electrical operation data of the circuit breaker opening and closing coil, and detecting the to-be-identified electrical operation data using the trained intelligent detection model to obtain the intelligent detection result of the circuit breaker opening and closing coil.

2. The circuit breaker opening and closing coil intelligent detection method according to claim 1, characterized in that, Collecting historical electrical operation data of the circuit breaker opening and closing coil under various operating states, and constructing training sample data according to the historical electrical operation data and the corresponding operating states, comprising: Collecting historical electrical operation data of the circuit breaker opening and closing coil under various operating states based on a preset data sampling frequency; the historical electrical operation data includes current data and / or voltage data; each piece of historical electrical operation data contains the same number of data; Pretreating the historical electrical operation data to obtain pretreated historical electrical operation data; Using the pretreated historical electrical operation data to construct sample data, and using the operating states corresponding to the historical electrical operation data to construct sample labels; Associating the sample data with its corresponding sample labels as training sample data, and obtaining multiple different training sample data.

3. The circuit breaker opening and closing coil intelligent detection method according to claim 1, characterized in that, Using a composite machine learning model to construct an intelligent detection model, and initializing the parameters of the intelligent detection model to obtain multiple coded individuals, comprising: Constructing a composite machine learning model comprising sequentially connected input layer, convolution layer, pooling layer, BiGRU layer, attention mechanism layer, fully connected layer and LSTM layer to obtain an intelligent detection model; Randomly initializing between the upper limit and the lower limit of the parameters of the intelligent detection model, and encoding the randomly initialized parameters into vectors to obtain coded individuals, and repeatedly obtaining multiple different coded individuals.

4. The circuit breaker opening and closing coil intelligent detection method according to claim 1, characterized in that, Based on the training sample data, the coded individuals are trained using a logarithmic spiral exploration strategy, a vortex path exploration strategy based on information entropy, and a fractional order balance exploration strategy based on a damping sawtooth wave oscillation factor to obtain a target coded individual, comprising: Based on the training sample data, the fitness of the coded individuals is obtained, and the coded individual with the maximum fitness is determined as the optimal individual; Based on the optimal individual, the coded individuals are guided using a logarithmic spiral exploration strategy to obtain guided coded individuals; The vortex path exploration strategy based on information entropy is used to perform uncertain exploration on the coded individual after the guidance, and coded individuals after uncertain exploration are obtained. The fractional order balance exploration strategy based on the damping sawtooth wave oscillation factor is used to perform balance exploration on the coded individual after the uncertain exploration, and coded individuals after balance exploration are obtained. It is judged whether the total iteration number is greater than the preset maximum iteration number, if yes, the target coded individual is obtained according to the coded individual after the balance exploration, otherwise, the coded individual after the balance exploration is taken as a basis to enter the next training process.

5. The circuit breaker opening and closing coil intelligent detection method according to claim 4, characterized in that, The logarithmic spiral exploration strategy is used to guide the coded individual based on the optimal individual, and the coded individual after the guidance is obtained, including: The logarithmic spiral function is used to obtain the guidance information based on the optimal individual, which is: ； wherein, denotes the i-th encoded individual, i = 1, 2, …, N, N denotes the total number of encoded individuals, denotes the optimal individual, t denotes the current total number of iterations, denotes the corresponding guidance information, denotes a natural constant, denotes the circle constant, denotes a random number between (0, 1), b denotes a spiral shape constant; The coded individual after the guidance is obtained according to the guidance information, which is: ； wherein, represents the encoded individual after the i-th steering.

6. The circuit breaker opening and closing coil intelligent detection method according to claim 4, characterized in that, The vortex path exploration strategy based on information entropy is used to perform uncertain exploration on the coded individual after the guidance, and coded individuals after uncertain exploration are obtained, including: The information entropy corresponding to the coded individual after the guidance is obtained, which is: ； ； wherein, denotes information entropy, denotes the information entropy occurrence probability parameter corresponding to the jth guided encoding individual, j = 1, 2, …, N, N denotes the total number of encoding individuals, denotes a logarithmic function, denotes the fitness corresponding to the jth guided encoding individual; According to the information entropy, the information selection quantity control factor is obtained, which is: ； wherein represents an information selection amount control factor; The dynamic step length control factor is obtained, which is: ； wherein, denotes a dynamic step control factor, denotes a minimum value of the dynamic step control factor, denotes a dynamic balance factor, denotes a maximum number of iterations, t denotes a current total number of iterations, denotes a minimum fitness corresponding to the coded individual after the guiding, denotes a maximum fitness corresponding to the coded individual after the guiding; According to the information selection quantity control factor and the dynamic step length control factor, the coded individual after the guidance is performed uncertain exploration, and the coded individual after uncertain exploration is obtained, which is: ； wherein, denotes the encoded individual after the jth guidance, denotes the encoded individual after the jth uncertain exploration, denotes the optimal individual, denotes the encoded individual after the j-1th guidance, denotes the encoded individual after the j+1th guidance, denotes the first learning factor, denotes the second learning factor, denotes a random number between (0,1), denotes a random number between (0,1).

7. The circuit breaker opening and closing coil intelligent detection method according to claim 4, characterized in that, The fractional order balance exploration strategy based on the damping sawtooth wave oscillation factor is used to perform balance exploration on the coded individual after the uncertain exploration, and coded individuals after balance exploration are obtained, including The damping sawtooth wave oscillation factor is obtained, which is: ； wherein, denotes a damped sawtooth wave oscillation factor, denotes a random number between (0, 1), denotes a random number between (0, 1), denotes a random number subject to a Gaussian distribution, denotes a sine function, denotes a cosine function, denotes a maximum number of iterations, t denotes a current total number of iterations, denotes an oscillation peak position control factor; It is judged whether the value of t / T is less than the damping sawtooth wave oscillation factor, if yes, the coded individual after the uncertain exploration is performed fractional order local development to obtain the coded individual after the balance exploration, otherwise, the coded individual after the uncertain exploration is performed fractional order global development to obtain the coded individual after the balance exploration.

8. The circuit breaker opening and closing coil intelligent detection method according to claim 7, characterized in that, The coded individual after the uncertain exploration is performed fractional order local development to obtain the coded individual after the balance exploration, including: The average Euclidean distance corresponding to the coded individual after the uncertain exploration is obtained, and the local fractional order order is obtained according to the average Euclidean distance, which is: ； ； wherein, represents a local fractional order, f represents an evolution factor, and e represents a natural constant represents the average Euclidean distance of the optimal individual to other individuals, represents the minimum value among all average Euclidean distances, represents the maximum value among all average Euclidean distances; According to the local fractional order order, the exploration speed is obtained, which is: ； wherein, denotes the exploration speed of the coded individual after the kth uncertain exploration in the tth iteration process, denotes the exploration speed of the coded individual after the kth uncertain exploration in the t+1th iteration process, denotes the exploration speed of the coded individual after the kth uncertain exploration in the t-1th iteration process, denotes the exploration speed of the coded individual after the kth uncertain exploration in the t-2th iteration process, denotes the exploration speed of the coded individual after the kth uncertain exploration in the t-3th iteration process; denotes the inertia weight, denotes a random number between (0, 1), denotes a natural constant, denotes the coded individual after the kth uncertain exploration, k = 1, 2, …, N, N denotes the total number of coded individuals, denotes that the coded individual after the other uncertain exploration of random matching, denotes the Euclidean distance between and denotes the information interaction degree control coefficient; According to the exploration speed, the coded individual after the balance exploration is obtained, which is: ； wherein, represents the kth encoded individual after the balance exploration.

9. The circuit breaker break opening and closing coil intelligent detection method according to claim 8, characterized in that, The coded individual after the uncertain exploration is performed fractional order global development to obtain the coded individual after the balance exploration, including: The global development parameter is obtained for the coded individual after the uncertain exploration, which is: ； wherein, denotes a global development parameter, denotes a total global fractional order, denotes the dth dimension parameter of the coded individual after the nth uncertain exploration, n = 1, 2, …, N, N denotes the total number of coded individuals, d = 1, 2, …, D, D denotes the total dimension of parameters in the coded individual, denotes the upper limit of the dth dimension parameter, denotes the maximum number of iterations, t denotes the current total number of iterations; According to the global development parameter, the coded individual after the balance exploration is obtained, which is: ； wherein, denotes the encoding individual after the kth equilibrium exploration, cos denotes the cosine function, denotes the circle constant, denotes the global development control coefficient, s denotes the global fractional order, denotes the dth dimensional parameter of the optimal individual.

10. A circuit breaker opening and closing coil intelligent detection device, characterized in that, including: The sample data acquisition module is used to collect the historical electrical operation data of the circuit breaker opening and closing coil under various operating states, and construct training sample data according to the historical electrical operation data and the corresponding operating states; the operating states include normal state and abnormal state; The model initialization module is used to construct an intelligent detection model by using a composite machine learning model, and initialize the parameters of the intelligent detection model to obtain a plurality of coded individuals; The model optimization module is configured to train the coded individuals by using a logarithmic spiral exploration strategy, a vortex path exploration strategy based on information entropy, and a fractional order balance exploration strategy based on a damping sawtooth wave oscillation factor, and obtain a target coded individual; The parameter application module is configured to decode the target coded individual, obtain final parameters of the intelligent detection model, apply the final parameters to the intelligent detection model, and obtain the trained intelligent detection model. The intelligent detection module is configured to collect to-be-recognized electrical operation data of the circuit breaker opening and closing coil, and detect the to-be-recognized electrical operation data by using the trained intelligent detection model to obtain an intelligent detection result of the circuit breaker opening and closing coil.

Images