Intelligent molded case circuit breaker control method and system based on big data

Abstract

The invention relates to the technical field of molded case circuit breaker control, and discloses an intelligent molded case circuit breaker control method and system based on big data, and the method comprises the steps: collecting the operation data of a circuit breaker and the related data of a circuit system in real time through a plurality of sensors disposed on the molded case circuit breaker, and obtaining multi-source heterogeneous data; constructing a deep generative adversarial network to carry out abnormal data detection on the collected multi-source heterogeneous data, and processing to obtain preprocessed data; constructing a risk assessment model by adopting a CNN-LSTM-Transform neural network, inputting the preprocessed data into the risk assessment model, and outputting a risk assessment result; according to a risk assessment result, combining an operation environment and a load condition of a current circuit, and based on an adaptive threshold decision algorithm and fuzzy logic control, dynamically adjusting an action threshold of the circuit breaker, and formulating a corresponding control strategy; according to the invention, the control intelligence level and reliability of the circuit breaker are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of molded case circuit breaker control, and specifically relates to an intelligent molded case circuit breaker control method and system based on big data. Background Technique

[0002] The molded case circuit breaker plays a crucial role in protecting the circuit. It can cut off the circuit in time when faults such as overload and short circuit occur in the circuit to prevent the expansion of accidents. However, the traditional control method of the molded case circuit breaker mainly relies on simple threshold judgment. For example, when the current exceeds the set threshold, the circuit breaker operates to cut off the circuit. This control method has obvious limitations. On the one hand, the setting of the threshold is usually fixed and cannot be dynamically adjusted according to the actual operation of the circuit, resulting in misoperation or non-operation in some complex power consumption environments, affecting the normal operation and protection effect of the circuit. On the other hand, the traditional method does not make full use of a large amount of historical data and real-time monitoring data, cannot comprehensively analyze and predict the operation state of the circuit, and is difficult to discover potential fault hazards in advance, reducing the reliability and safety of the circuit system. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design an intelligent molded case circuit breaker control method and system based on big data.

[0004] The first aspect of the present invention provides an intelligent molded case circuit breaker control method based on big data, and the method includes the following steps: Real-time collect the operation data of the circuit breaker and the relevant data of the circuit system through multiple sensors set on the molded case circuit breaker to obtain multi-source heterogeneous data; Construct a deep generative adversarial network to detect abnormal data in the collected multi-source heterogeneous data, and obtain preprocessed data after processing; Adopt a CNN-LSTM-Transformer neural network to construct a risk assessment model, and input the preprocessed data into the risk assessment model to output a risk assessment result; According to the risk assessment result, combined with the current operation environment and load condition of the circuit, dynamically adjust the action threshold of the circuit breaker based on the adaptive threshold decision algorithm and fuzzy logic control, and formulate corresponding control strategies.

[0005] Optionally, in the first implementation manner of the first aspect of the present invention, the constructing a deep generative adversarial network to detect abnormal data in the collected multi-source heterogeneous data, and obtaining preprocessed data after processing includes: Perform sliding window filtering on the multi-source heterogeneous data to remove pulse noise and baseline drift, and use median filtering to process current mutation points; Perform Z-score normalization on sensor data with different dimensions, and use the dynamic time warping algorithm to align time series data with different sampling rates; Detect abnormal data in the cleaned data through a deep generative adversarial network. The generator learns the normal data distribution pattern, and the discriminator identifies abnormal data deviating from the manifold structure; Calculate the anomaly score output by the discriminator based on the Mahalanobis distance, determine the level of abnormal data, and perform anomaly processing according to the corresponding level to obtain preprocessed data.

[0006] Optionally, in the second implementation manner of the first aspect of the present invention, the generator adopts a variational autoencoder structure, including 3 layers of bidirectional LSTM and 2 layers of fully connected layers, with 256 neurons in each layer of bidirectional LSTM; The discriminator adopts a spatio-temporal convolutional network structure, including 8 dilated causal convolutional blocks, and identifies abnormal data distributions through multi-scale time feature extraction.

[0007] Optionally, in the third implementation manner of the first aspect of the present invention, the CNN module in the risk assessment model includes a 1D convolutional layer and an attention mechanism. The 1D convolutional layer is provided with 3 convolutional blocks with convolutional kernel sizes of 11, 7, and 5 respectively, a stride of 1, and each convolutional block is followed by BatchNorm and the LeakyReLU activation function. The attention mechanism adopts the fusion of channel attention and spatial attention; The LSTM module in the risk assessment model includes a bidirectional LSTM layer and an attention gating mechanism, where the bidirectional LSTM layer includes 2 layers of structures, with 128 neurons in each layer; The Transformer module in the risk assessment model includes a multi-head self-attention mechanism and position encoding, where the multi-head self-attention mechanism includes 8 heads and a hidden dimension of 256.

[0008] Optionally, in the fourth implementation manner of the first aspect of the present invention, the risk assessment model is constructed using a CNN-LSTM-Transformer neural network, and the preprocessed data is input into the risk assessment model to output a risk assessment result, including: Concatenate the local features extracted by the CNN module, the temporal features extracted by the LSTM module, and the correlation features extracted by the Transformer module, and reduce the dimension through a fully connected layer; Adopt a 3-layer MLP structure to output a 5-dimensional risk vector, which respectively represents overload risk, short-circuit risk, poor contact risk, insulation aging risk, and environmental adaptability risk; Convert the risk vector into a probability distribution through the Softmax activation function, and output the risk assessment result.

[0009] Optionally, in the fifth implementation manner of the first aspect of the present invention, according to the risk assessment result, combining the operating environment and load conditions of the current circuit, based on the adaptive threshold decision algorithm and fuzzy logic control, dynamically adjust the action threshold of the circuit breaker, and formulate corresponding control strategies, including: Fuse and process the risk assessment result, the operating environment and load conditions of the current circuit, and integrate them into a 28-dimensional state vector, and input it into the reinforcement learning Actor network, and the Actor network outputs a discrete action set; Convert the risk probability, load change rate, and environmental adaptability score into fuzzy linguistic variables, divide the fuzzy sets through the triangular membership function, perform logical reasoning based on the preset fuzzy rules, and use the centroid method to convert the fuzzy output into specific values to obtain the threshold adjustment coefficient, warning response time, and trip energy compensation coefficient; Weightedly fuse the actions output by the reinforcement learning and the adjustment coefficients output by the fuzzy logic to form the final comprehensive adjustment parameter; Set the overload long-time delay threshold, short-circuit short-time delay threshold, and short-circuit instantaneous threshold based on the comprehensive adjustment parameter, and formulate corresponding control strategies.

[0010] Optionally, in the sixth implementation manner of the first aspect of the present invention, at each decision-making, the Actor network calculates the probability distribution of each action according to the state vector, selects the action with the highest long-term cumulative reward, and the Critic network synchronously evaluates the action value. By comparing the actual reward with the estimated reward, the parameters of the Actor network are optimized in reverse.

[0011] The second aspect of the present invention provides an intelligent molded case circuit breaker control system based on big data. The system includes: An acquisition module, configured to collect the operation data of the circuit breaker and the relevant data of the circuit system in real time through a plurality of sensors arranged on the molded case circuit breaker to obtain multi-source heterogeneous data; A preprocessing module, configured to construct a deep generative adversarial network to detect abnormal data in the collected multi-source heterogeneous data, and obtain preprocessed data after processing; A risk assessment module, configured to construct a risk assessment model using a CNN-LSTM-Transformer neural network, and input the preprocessed data into the risk assessment model to output a risk assessment result; A dynamic adjustment module, configured to dynamically adjust the action threshold of the circuit breaker according to the risk assessment result, combine the operating environment and load conditions of the current circuit, and based on the adaptive threshold decision algorithm and fuzzy logic control, and formulate corresponding control strategies.

[0012] The third aspect of the present invention provides an intelligent molded case circuit breaker control device based on big data. The intelligent molded case circuit breaker control device based on big data includes a memory and at least one processor, and instructions are stored in the memory; the at least one processor calls the instructions in the memory to enable the intelligent molded case circuit breaker control device based on big data to execute each step of the intelligent molded case circuit breaker control method based on big data as described in any one of the above.

[0013] The fourth aspect of the present invention provides a computer-readable storage medium, and instructions are stored on the computer-readable storage medium. When the instructions are executed by a processor, each step of the intelligent molded case circuit breaker control method based on big data as described in any one of the above is implemented.

[0014] In the technical solution provided by the present invention, multiple sensors arranged on the molded case circuit breaker are used to collect the operation data of the circuit breaker and the relevant data of the circuit system in real time to obtain multi-source heterogeneous data; a deep generative adversarial network is constructed to detect abnormal data in the collected multi-source heterogeneous data, and the preprocessed data is obtained after processing; a risk assessment model is constructed by using a CNN-LSTM-Transformer neural network, and the preprocessed data is input into the risk assessment model to output a risk assessment result; according to the risk assessment result, combined with the operation environment and load conditions of the current circuit, based on an adaptive threshold decision algorithm and fuzzy logic control, the action threshold of the circuit breaker is dynamically adjusted, and corresponding control strategies are formulated; through the application of big data technology, the present invention realizes the comprehensive collection, processing and analysis of the operation data of the molded case circuit breaker, can dynamically adjust the action threshold according to the operation conditions of the actual circuit, formulate more accurate control strategies, improve the control intelligence level and reliability of the circuit breaker, can discover potential fault hazards in advance, realize the preventive maintenance of the circuit system, reduce the failure rate and maintenance cost of the equipment, and has significant economic and social benefits. Description of the Drawings

[0015] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention.

[0016] Figure 1 It is a flowchart of the intelligent molded case circuit breaker control method based on big data provided by the embodiment of the present invention; Figure 2 It is a schematic structural diagram of the intelligent molded case circuit breaker control system based on big data provided by the embodiment of the present invention; Figure 3 It is a schematic structural diagram of the intelligent molded case circuit breaker control device based on big data provided by the embodiment of the present invention. Detailed implementation manners

[0017] In the description of the present invention, the terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and the above drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described here can be implemented in an order different from that shown or described here. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or device that includes a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention will be described below. Please refer to Figure 1 The flowchart of the intelligent molded case circuit breaker control method based on big data provided by the embodiments of the present invention. The method specifically includes the following steps: Step 101: Real-time collect the operation data of the circuit breaker and the relevant data of the circuit system through multiple sensors arranged on the molded case circuit breaker to obtain multi-source heterogeneous data; In this embodiment, the current sensor uses a Hall effect current sensor, which is installed at the incoming line end and the outgoing line end of the molded case circuit breaker and is used to collect data such as real-time current value, current waveform, and current harmonic components. The accuracy reaches 0.5% FS, and the sampling frequency is 10 kHz to accurately reflect the current change in the circuit; the voltage sensor selects a capacitive voltage sensor, which is also installed at the incoming line end and the outgoing line end, and can collect data such as real-time voltage value, voltage waveform, voltage fluctuation amplitude, and voltage sag and swell. The measurement accuracy is 0.5%, and the sampling frequency is 10 kHz to ensure accurate monitoring of the circuit voltage state; the temperature sensor installs a thermocouple temperature sensor at the easily heated parts such as the contacts, coils, and arc extinguishing chambers of the molded case circuit breaker to collect the temperature data of each part in real time. The accuracy is ±1°C, and the sampling frequency is 1 Hz to timely detect abnormal heating caused by reasons such as poor contact and overload; humidity sensor: The humidity sensor is installed in the cavity inside the circuit breaker to collect the relative humidity data of the internal environment. The accuracy is ±3% RH, and the sampling frequency is 1 Hz, considering the influence of humidity on the insulation performance of the circuit breaker; the vibration sensor installs an acceleration vibration sensor on the outer shell of the circuit breaker to collect the vibration signal during the operation of the circuit breaker. The frequency response range is 0.5 Hz - 10 kHz, and the sampling frequency is 20 kHz. By analyzing the vibration data, the fastening degree and operation state of the internal components of the circuit breaker can be judged; Each sensor is connected to the microprocessor of the molded case circuit breaker through a dedicated data acquisition module. The acquisition module uses a high-speed A / D converter to convert the analog signal output by the sensor into a digital signal, and transmits the data to the central data processing unit through the CAN bus or Ethernet. During the data transmission process, CRC checksum and encryption technologies are used to ensure the integrity and security of the data.

[0019] Step 102: Construct a deep generative adversarial network to detect abnormal data in the collected multi-source heterogeneous data, and obtain preprocessed data after processing. In this embodiment, sliding window filtering is performed on the multi-source heterogeneous data to remove impulse noise and baseline drift, and median filtering is used to process current mutation points; Z-score normalization is performed on sensor data with different dimensions, and the dynamic time warping algorithm is used to align time series data with different sampling rates; the deep generative adversarial network is used to detect abnormal data in the cleaned data, the generator learns the normal data distribution pattern, and the discriminator identifies abnormal data deviating from the manifold structure; the Mahalanobis distance is used to calculate the abnormal score output by the discriminator, determine the level of abnormal data, and perform abnormal processing according to the corresponding level to obtain preprocessed data.

[0020] In this embodiment, the generator adopts a variational autoencoder structure, which includes 3 layers of bidirectional LSTM and 2 layers of fully connected layers. Random noise is introduced through the reparameterization trick to enhance the robustness of the model and generate synthetic samples consistent with the real data distribution, with 256 neurons in each layer of bidirectional LSTM; the discriminator adopts a spatio-temporal convolutional network structure, which includes 8 dilated causal convolutional blocks, and identifies abnormal data distributions through multi-scale time feature extraction.

[0021] In this embodiment, the detected abnormal data is processed at three levels: Level I anomaly (score > 7σ): directly eliminate and trigger the sensor self-check process; Level II anomaly (5σ < score ≤ 7σ): mark as suspicious data and use interpolation method for data repair; Level III anomaly (3σ < score ≤ 5σ): retain the data but increase the weight decay factor to reduce its impact on subsequent analysis.

[0022] Step 103: Use a CNN-LSTM-Transformer neural network to construct a risk assessment model, and input the preprocessed data into the risk assessment model to output a risk assessment result. In this embodiment, the CNN module includes a 1D convolutional layer and an attention mechanism. The 1D convolutional layer is provided with 3 convolutional blocks, and the convolutional kernel sizes are 11, 7, and 5 respectively, with a stride of 1. After each convolutional block, there are a BatchNorm and a LeakyReLU activation function. The attention mechanism adopts the fusion of channel attention and spatial attention to enhance the attention to key features, and is used to extract local patterns such as the frequency domain features of the current waveform and the spatial features of temperature changes; The LSTM module includes a bidirectional LSTM layer and an attention gating mechanism. The bidirectional LSTM layer includes 2 layers of structures, with 128 neurons in each layer, capturing the long-term and short-term dependencies of the data. The attention gating mechanism weights the hidden states output by the LSTM to highlight the important time step features and processes the trend and periodic features in the time series data; The Transformer module includes a multi-head self-attention mechanism and a positional encoding. The multi-head self-attention mechanism includes 8 heads and a hidden dimension of 256, capturing the non-local associations and complex interactions in the data; The positional encoding adopts a sine-cosine positional encoding to retain the time series information and process the correlation and co-variation patterns between multi-sensor data.

[0023] In this embodiment, the local features extracted by the CNN module, the time series features extracted by the LSTM module, and the correlation features extracted by the Transformer module are tensor concatenated and dimension-reduced through a fully connected layer; A 3-layer MLP structure is used to output a 5-dimensional risk vector, which respectively represents the overload risk, short circuit risk, poor contact risk, insulation aging risk, and environmental adaptability risk; The risk vector is converted into a probability distribution through a Softmax activation function to output the risk assessment result.

[0024] In this embodiment, in the design of the CNN module, an architecture of a 1D convolutional layer and a fusion attention mechanism is adopted, which is specifically for the extraction of local features of time series and spatial distribution data such as current waveforms and temperature sequences. The 1D convolutional layer contains 3 cascaded convolutional blocks, and convolutional kernels with sizes of 11, 7, and 5 are used in sequence, corresponding to different scales of feature perception ranges: the 11-kernel captures long-period frequency components, such as the low-frequency oscillation of current harmonics, the 7-kernel focuses on medium-frequency dynamic changes, such as the transition process of voltage sags, and the 5-kernel extracts high-frequency details, such as the current spike at the moment of contact of the contact; The BatchNorm layer accelerates the training convergence by normalizing the input distribution, and the LeakyReLU activation function alleviates the dead neuron problem of ReLU by retaining the negative gradient, ensuring the stability of the non-linear transformation of features; Channel attention performs global average pooling and global max pooling on the feature map output by the 1D convolution to generate channel-level mean and extreme value descriptors respectively; extracts inter-channel dependencies through a shared two-layer fully-connected network, with the number of input channels → number of channels / 2 → number of channels, and generates channel weights after sigmoid activation to achieve channel-level enhancement of the frequency-domain features of the current waveform; Spatial attention performs average and max pooling on the feature map in the channel dimension, and generates spatial weights through a 1D convolution of 7×1 after concatenation, focusing on spatial distribution anomalies of the temperature sensor array such as the temperature of a certain contact being significantly higher than adjacent positions or local impact features of vibration signals such as high-frequency vibration points caused by bolt loosening; Multiply the channel weights and spatial weights element-wise and superimpose them on the original feature map to form an output that enhances local patterns. Capture signal features of different granularities through multi-scale convolutional kernels, and combine the attention mechanism to dynamically allocate feature weights, effectively improving the sensitivity to local abnormal patterns such as the frequency-domain features of the current waveform and the spatial correlation of temperature changes, and providing a refined feature representation for risk assessment.

[0025] In this embodiment, the LSTM module adopts a combined architecture of a bidirectional LSTM layer and an attention gating mechanism, which is specifically designed to handle the long-term and short-term dependencies in the time series features extracted by the CNN. The bidirectional LSTM layer is designed as a two-layer stacked structure, with each layer containing 128 neurons, capturing the historical and future context information of the time series data through forward and backward propagation simultaneously; The first layer of bidirectional LSTM inputs the feature sequence output by the CNN into the forward and backward LSTM units respectively. Each time step outputs a 256-dimensional hidden state, 128 dimensions forward and 128 dimensions backward, capturing short-term dynamic features such as current fluctuations and temperature change rates; The second layer of bidirectional LSTM receives the output sequence of the first layer and further mines the long-term dependence patterns across time steps, such as daily / weekly load cycles and equipment aging trends, and outputs the dimension; For the hidden state sequence of the bidirectional LSTM, generate attention weights through a fully-connected layer, multiply the attention weights and the hidden state of the corresponding time step element-wise to obtain weighted features, making the model focus on key time steps such as current mutation points before short circuits and abnormal acceleration segments of temperature rise. Perform global average pooling on the weighted feature sequence to generate the final time series feature vector, which retains the trend and periodic features screened by attention; capture the multi-scale dependencies of time series data through the bidirectional information flow of the bidirectional LSTM, and combine the attention gating mechanism to automatically identify fault precursor features and periodic patterns, significantly improving the model's ability to model non-stationary time series data.

[0026] In this embodiment, the Transformer module adopts a combined architecture of multi-head self-attention mechanism and sine-cosine position encoding, which is specifically used to process the non-local correlations and co-variation patterns in the temporal features extracted by the LSTM. The multi-head self-attention mechanism is set to 8 heads, and the hidden dimension is 256, realizing feature interaction modeling; The temporal features output by the LSTM are projected into the query matrix Q, the key matrix K, and the value matrix V respectively. The dimension of each head is 32. For each head, the scaled dot-product attention is calculated to capture the interaction patterns in specific subspaces such as the phase relationship between different phase currents and the coupling effect of temperature and vibration. The outputs of the 8 heads are concatenated and then mapped back to 256 dimensions through a fully connected layer, forming a feature representation that fuses multi-scale dependency relationships, effectively capturing complex patterns such as the co-mutation of multi-sensor data during short-circuit faults and the non-linear correlation between temperature and current under overload conditions; The temporal position information is injected into the Transformer input, and the absolute position information is converted into sine-cosine waves of different frequencies, enabling the model to learn temporal dependencies such as during the device startup / stop phase and the temporal patterns of periodic load changes. The position encoding vector is added element-wise to the Transformer input features to ensure that the model retains the original temporal order when processing non-local correlations.

[0027] In this embodiment, the loss function uses a weighted cross-entropy loss function, assigning different weights to different types of risks. The fault risk weight is set to 5 times that of the normal state; the optimizer uses the AdamW optimizer with an initial learning rate of 1e-4, and a cosine annealing learning rate scheduler with a period of 20 epochs; the L2 regularization coefficient is 1e-5, the Dropout rate is 0.3, and the parameter momentum of the BatchNorm layer is 0.9. Training stops when the validation set loss does not decrease for 10 consecutive epochs, and the best model is saved.

[0028] Step 104: According to the risk assessment results, combined with the current operating environment and load conditions of the circuit, based on the adaptive threshold decision algorithm and fuzzy logic control, dynamically adjust the action threshold of the circuit breaker and formulate corresponding control strategies.

[0029] In this embodiment, the risk assessment results, the operating environment of the current circuit, and the load conditions are integrated and processed, and integrated into a 28-dimensional state vector, which is input into the reinforcement learning Actor network. The Actor network outputs a discrete action set; the risk probability, load change rate, and environmental adaptability score are converted into fuzzy linguistic variables, the fuzzy sets are divided through the triangular membership function, logical reasoning is performed based on the preset fuzzy rules, and the centroid method is used to convert the fuzzy output into a specific value to obtain the threshold adjustment coefficient, warning response time, and trip energy compensation coefficient; the actions output by the reinforcement learning and the adjustment coefficients output by the fuzzy logic are weighted and integrated to form the final comprehensive adjustment parameter; based on the comprehensive adjustment parameter, the overload long-time delay threshold, short-circuit short-time delay threshold, and short-circuit instantaneous threshold are set, and the corresponding control strategy is formulated.

[0030] In this embodiment, during each decision-making, the Actor network calculates the probability distribution of each action according to the state vector, selects the action with the highest long-term cumulative reward, and the Critic network synchronously evaluates the action value. By comparing the actual reward with the predicted reward, the Actor network parameters are optimized in reverse.

[0031] In this embodiment, the design of the reward function takes into account safety, reliability, and economy. For example, when a real short-circuit fault is detected and the threshold is set reasonably, the reward value is +100; if a false trip occurs due to the threshold being set too sensitively, the reward value is -50; when operating stably for a long time, a reward of +10 to +30 is given according to the load balance degree. Through this mechanism, the model gradually learns the optimal threshold adjustment strategy under different working conditions.

[0032] In this embodiment, the action thresholds of the molded case circuit breaker are continuously adjustable, such as the overload protection threshold, short-circuit trip threshold, etc. However, directly optimizing the continuous space will cause the complexity of the policy search to increase exponentially. By dividing the continuous threshold range into several discrete gears, for example, dividing the overload threshold into three gears: low, medium, and high, or abstracting the control strategy into a limited number of discrete operations, such as increasing the threshold by 5%, decreasing the threshold by 10%, or maintaining the current threshold, the problem can be transformed into an optimization problem in the discrete action space, significantly reducing the computational burden of the reinforcement learning algorithm and making it more suitable for real-time control scenarios.

[0033] In this embodiment, through the Mamdani reasoning method, the outputs of multiple rules are aggregated to generate a fuzzy control quantity. The fuzzy rules are as follows: Rule 1: If the overload risk is high and the load change rate is high, the threshold adjustment coefficient is reduced by 15%, and the warning response time is 200 ms; Rule 2: If the insulation aging risk is medium and the environmental adaptability is poor, the threshold adjustment coefficient is reduced by 5%, and the trip energy compensation is increased.

[0034] In this embodiment, the overload long-time delay threshold is adjusted within the range of 0.7 to 1.5 times the rated current according to the load type and risk level, such as motor / resistive load, and the adjustment step does not exceed 5% of the rated current; The short-circuit short-time delay threshold is dynamically corrected within the range of 5 to 10 times the rated current based on the characteristics of the real-time short-circuit current waveform, such as the rising rate, to avoid misjudgment caused by transient shocks; The short-circuit instantaneous threshold is temporarily increased by 10% to 20% for high-frequency transient faults, such as lightning strikes, to suppress non-fault tripping.

[0035] In this embodiment, the early warning strategy triggers four-level early warnings, green / yellow / orange / red, according to the risk probability. For example, when an orange warning is issued, local audible and visual alarms + cloud push are used to remind the maintenance personnel to check for potential hazards; The tripping strategy gives priority to handling emergency risks, such as short-circuit faults with a red warning, triggering instantaneous tripping, with a response time < 30 ms. For progressive risks, such as overload, an inverse time tripping curve is adopted. The greater the load, the shorter the tripping time, to avoid frequent power outages affecting power supply continuity; The compensation strategy automatically increases the tripping energy compensation for harsh environments or aging equipment scenarios to ensure reliable opening of the contacts and reduce the fire risk caused by abnormal contact resistance.

[0036] In this embodiment, the adjusted threshold parameters and control strategies are encoded into Modbus protocol instructions and are sent to the circuit breaker terminal in real time through the edge computing gateway to update the internal protection logic parameters. The instructions include the effective time, adjustment range, and strategy validity period. For example, the validity period of a temporary threshold adjustment is 4 hours, and the default value is restored after expiration; After the strategy is executed, the circuit breaker action data and circuit parameters are continuously collected. The action data includes the tripping time, the number of times the contacts are opened and closed. By comparing the actual response with the expected effect, such as whether the tripping is correct under the set threshold, the execution error is calculated. If misoperation or missed operation occurs continuously three times, the model retraining process is automatically triggered to update the reinforcement learning strategy and the fuzzy rule base.

[0037] Please refer to Figure 2 , the structural schematic diagram of the intelligent molded case circuit breaker control system based on big data provided by the embodiment of the present invention. The system includes: An acquisition module for collecting the operation data of the circuit breaker and the relevant data of the circuit system in real time through a plurality of sensors arranged on the molded case circuit breaker to obtain multi-source heterogeneous data; A preprocessing module for constructing a deep generative adversarial network to detect abnormal data in the collected multi-source heterogeneous data and obtaining preprocessed data after processing; A risk assessment module, which is used to construct a risk assessment model by using a CNN-LSTM-Transformer neural network, and input the preprocessed data into the risk assessment model to output a risk assessment result; A dynamic adjustment module, which is used to dynamically adjust the action threshold of the circuit breaker and formulate corresponding control strategies according to the risk assessment result, in combination with the operating environment and load conditions of the current circuit, based on an adaptive threshold decision algorithm and fuzzy logic control.

[0038] Figure 3 FIG. is a schematic structural diagram of an intelligent molded case circuit breaker control device based on big data provided by an embodiment of the present invention. The intelligent molded case circuit breaker control device 300 based on big data may vary greatly due to configuration or performance differences, and may include one or more processors (central processing units, CPUs) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 for storing application programs 333 or data 332 (for example, one or more mass storage devices). Among them, the memory 320 and the storage media 330 may be transient storage or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the intelligent molded case circuit breaker control device 300 based on big data. Further, the processor 310 may be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the intelligent molded case circuit breaker control device 300 to implement the method provided in the above embodiment.

[0039] The intelligent molded case circuit breaker control device 300 based on big data may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating devices 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that Figure 3 The structure of the intelligent molded case circuit breaker control device based on big data shown does not constitute a limitation on the computer device provided by the present invention, and may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements.

[0040] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer is caused to execute the steps of the intelligent molded case circuit breaker control method based on big data provided in the above various embodiments.

[0041] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described devices, apparatuses, or units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0042] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0043] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent molded case circuit breaker control method based on big data, characterized in that, The method includes the following steps: Multiple sensors installed on the molded case circuit breaker are used to collect the operation data of the circuit breaker and the relevant data of the circuit system in real time, obtaining multi-source heterogeneous data; A deep generative adversarial network is constructed to detect abnormal data in the collected multi-source heterogeneous data, and the preprocessed data is obtained after processing; A risk assessment model is constructed using a CNN-LSTM-Transformer neural network, and the preprocessed data is input into the risk assessment model to output a risk assessment result; According to the risk assessment result, combined with the current operating environment and load conditions of the circuit, based on the adaptive threshold decision algorithm and fuzzy logic control, the action threshold of the circuit breaker is dynamically adjusted, and corresponding control strategies are formulated.

2. The intelligent molded case circuit breaker control method based on big data according to claim 1, characterized in that, The construction of the deep generative adversarial network to detect abnormal data in the collected multi-source heterogeneous data and obtain the preprocessed data after processing includes: Sliding window filtering is performed on the multi-source heterogeneous data to remove impulse noise and baseline drift, and median filtering is used to process current mutation points; Z-score normalization is performed on sensor data with different dimensions, and the dynamic time warping algorithm is used to align time series data with different sampling rates; The deep generative adversarial network is used to detect abnormal data in the cleaned data. The generator learns the normal data distribution pattern, and the discriminator identifies abnormal data deviating from the manifold structure; Based on the Mahalanobis distance, the anomaly score output by the discriminator is calculated, the level of the abnormal data is determined, and abnormal processing is performed according to the corresponding level to obtain the preprocessed data.

3. The intelligent molded case circuit breaker control method based on big data according to claim 2, wherein The generator adopts a variational autoencoder structure, including 3 layers of bidirectional LSTM and 2 layers of fully connected layers, with 256 neurons in each layer of bidirectional LSTM; The discriminator adopts a spatio-temporal convolutional network structure, including 8 dilated causal convolutional blocks, and identifies abnormal data distributions through multi-scale time feature extraction.

4. The intelligent molded case circuit breaker control method based on big data according to claim 1, wherein The CNN module in the risk assessment model includes a 1D convolutional layer and an attention mechanism. The 1D convolutional layer is provided with 3 convolutional blocks with convolutional kernel sizes of 11, 7, and 5 respectively, a stride of 1, and a BatchNorm and LeakyReLU activation function are connected after each convolutional block. The attention mechanism adopts the fusion of channel attention and spatial attention; The LSTM module in the risk assessment model includes a bidirectional LSTM layer and an attention gating mechanism, where the bidirectional LSTM layer includes 2 layers, with 128 neurons in each layer; The Transformer module in the risk assessment model includes a multi-head self-attention mechanism and position encoding, where the multi-head self-attention mechanism includes 8 heads and a hidden dimension of 256.

5. The intelligent molded case circuit breaker control method based on big data according to claim 1, characterized in that, The construction of the risk assessment model using a CNN-LSTM-Transformer neural network and inputting the preprocessed data into the risk assessment model to output a risk assessment result includes: The local features extracted by the CNN module, the temporal features extracted by the LSTM module, and the correlation features extracted by the Transformer module are tensor concatenated and dimension-reduced through a fully connected layer; Adopt a 3-layer MLP structure to output a 5-dimensional risk vector, which respectively represents overload risk, short-circuit risk, poor contact risk, insulation aging risk, and environmental adaptability risk; Convert the risk vector into a probability distribution through the Softmax activation function, and output the risk assessment result.

6. The intelligent molded case circuit breaker control method based on big data according to claim 1, characterized in that According to the risk assessment result, combined with the operating environment and load conditions of the current circuit, based on the adaptive threshold decision algorithm and fuzzy logic control, dynamically adjust the action threshold of the circuit breaker, and formulate corresponding control strategies, including: Fuse and process the risk assessment result, the operating environment and load conditions of the current circuit, and integrate them into a 28-dimensional state vector, and input it into the reinforcement learning Actor network. The Actor network outputs a discrete action set; Convert the risk probability, load change rate, and environmental adaptability score into fuzzy linguistic variables, divide the fuzzy set through the triangular membership function, perform logical reasoning based on the preset fuzzy rules, and use the centroid method to convert the fuzzy output into a specific value to obtain the threshold adjustment coefficient, warning response time, and tripping energy compensation coefficient; Perform weighted fusion on the actions output by reinforcement learning and the adjustment coefficients output by fuzzy logic to form the final comprehensive adjustment parameter; Set the overload long-time delay threshold, short-circuit short-time delay threshold, and short-circuit instantaneous threshold based on the comprehensive adjustment parameter, and formulate corresponding control strategies.

7. The intelligent molded case circuit breaker control method based on big data according to claim 6, wherein During each decision-making, the Actor network calculates the probability distribution of each action according to the state vector, selects the action with the highest long-term cumulative reward, and the Critic network synchronously evaluates the action value. By comparing the actual reward with the predicted reward, the Actor network parameters are optimized in reverse.

8. An intelligent molded case circuit breaker control system based on big data, characterized in that, The system includes: A collection module for collecting the operation data of the circuit breaker and the relevant data of the circuit system in real time through multiple sensors arranged on the molded case circuit breaker to obtain multi-source heterogeneous data; A preprocessing module for constructing a deep generative adversarial network to detect abnormal data in the collected multi-source heterogeneous data, and obtaining preprocessed data after processing; A risk assessment module for constructing a risk assessment model using a CNN-LSTM-Transformer neural network, and inputting the preprocessed data into the risk assessment model to output a risk assessment result; A dynamic adjustment module for dynamically adjusting the action threshold of the circuit breaker and formulating corresponding control strategies based on the risk assessment result, combined with the operating environment and load conditions of the current circuit, based on the adaptive threshold decision algorithm and fuzzy logic control.

9. An intelligent molded case circuit breaker control device based on big data, characterized in that, The intelligent molded case circuit breaker control device based on big data includes a memory and at least one processor, and instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the intelligent molded case circuit breaker control device based on big data executes each step of the intelligent molded case circuit breaker control method according to any one of claims 1-7.

10. A computer-readable storage medium, on which instructions are stored, characterized in that, When the instructions are executed by the processor, each step of the intelligent molded case circuit breaker control method according to any one of claims 1-7 is implemented.

Images