Power distribution cabinet standby regulation and control system based on machine vision

Abstract

The invention relates to the technical field of power distribution cabinet equipment monitoring, discloses a power distribution cabinet standby regulation and control system based on machine vision, and aims to improve the stability and operation efficiency of a power system. The system collects equipment images in the power distribution cabinet in real time through a high-definition camera, performs equipment identification and state monitoring by using a convolutional neural network (CNN) algorithm, and predicts the future state trend of the equipment in combination with a long short-term memory (LSTM) algorithm. And the regulation and control decision module intelligently generates a regulation and control instruction according to the current and future states of the equipment by adopting a deep reinforcement learning algorithm. And the execution control module is responsible for receiving the instruction and controlling the power distribution cabinet equipment. According to the invention, intelligent monitoring and accurate regulation and control of power distribution cabinet equipment are realized, the automation level and stable operation capability of a power system are effectively improved, the manual inspection cost is reduced, the timeliness and accuracy of fault handling are improved, and powerful support is provided for intelligent management of the power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution cabinet equipment monitoring, and in particular to a power distribution cabinet equipment control system based on machine vision. Background Art

[0002] With the rapid development of science and technology, power equipment and systems are becoming increasingly intelligent and automated. As a crucial component of power systems, the operating status of distribution cabinets (DPCs) significantly impacts the stable operation of the entire system. Traditional methods for monitoring and controlling DPC equipment rely primarily on manual inspections and operations, which are inefficient and make it difficult to identify and resolve problems in a timely manner.

[0003] In recent years, the development of machine vision technology has opened up new possibilities for intelligent monitoring and control of distribution cabinet equipment. Machine vision technology uses high-definition cameras to capture real-time device images and, through image processing and recognition algorithms, automatically identifies and monitors device status. However, existing machine vision-based distribution cabinet equipment monitoring systems are mostly limited to real-time monitoring of device status and lack the ability to intelligently control it. Traditional device control methods rely primarily on pre-set rules and thresholds for judgment and decision-making. This approach is unable to adapt to the complex operating conditions and changing environmental factors of distribution cabinet equipment, making it difficult to achieve precise and efficient control.

[0004] How to combine machine vision technology and advanced machine learning algorithms to achieve intelligent monitoring and precise control of equipment inside distribution cabinets and improve the stable operation of power systems has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of the present invention is to provide a distribution cabinet control system based on machine vision to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a distribution cabinet control system based on machine vision, the system comprising:

[0007] The image acquisition module is used to collect image data of the equipment inside the power distribution cabinet in real time through a high-definition camera, and pre-process the image data to form a standardized image data set;

[0008] The device identification module is used to build a device identification model. This model is based on the convolutional neural network (CNN) algorithm, which extracts features and classifies devices from preprocessed image data, and outputs device type, location, and status information.

[0009] The condition monitoring module is used to build a condition monitoring model. This model combines the long short-term memory network (LSTM) algorithm to perform time series analysis on the condition information output by the equipment identification model and predict the future condition trend of the equipment.

[0010] The control decision module is used to generate equipment control instructions based on the output of the condition monitoring model. Specifically, it uses a deep reinforcement learning algorithm to build a control strategy model, defines a state evaluation function, quantitatively evaluates the control results based on the current state and future state trends of the equipment, and dynamically adjusts the generated control instructions based on the evaluation results.

[0011] The execution control module is used to receive control instructions and perform corresponding operation control on the equipment inside the distribution cabinet.

[0012] Preferably, the device identification model is constructed as follows:

[0013] Define the CNN model structure, including convolutional layers, pooling layers, fully connected layers, and output layers; the convolutional layers are used to extract image features, the pooling layers are used to reduce feature dimensions, the fully connected layers are used to integrate feature information, and the output layer uses the softmax function to output the probability distribution of device type, location, and status.

[0014] Preferably, the step of training the device recognition model includes:

[0015] Collect training data sets, including standard images of various types of equipment and their labels. The labels include equipment type, location, and status information.

[0016] Initialize the model parameters, set the learning rate, batch size and training rounds;

[0017] The training data set is input into the model, and the loss between the model output and the true label is calculated through forward propagation. The loss function adopts the cross entropy loss function;

[0018] The gradient of the loss function with respect to the model parameters is calculated through back propagation, and the stochastic gradient descent (SGD) algorithm is used to update the model parameters.

[0019] Evaluate the recognition accuracy of the trained model on the validation set, and adjust the model structure and parameters based on the evaluation results.

[0020] Preferably, the state monitoring model is constructed as follows:

[0021] Define the LSTM model structure, including the input layer, LSTM layer, fully connected layer, and output layer; the LSTM layer is used to process time series data and capture the changing patterns of device status, and the fully connected layer is used to map the output of the LSTM layer to trend predictions of the device's future status.

[0022] Preferably, the training step of the condition monitoring model includes:

[0023] Collect training datasets, including labels of historical status sequences of devices and their future status trends;

[0024] Initialize the parameters of the LSTM model, set the learning rate, sequence length and training rounds;

[0025] The training data set is input into the LSTM model, and the loss between the model output and the true label is calculated through forward propagation. The loss function adopts the mean square error loss function;

[0026] The gradient of the loss function with respect to the model parameters is calculated through back propagation, and the Adam optimization algorithm is used to update the model parameters;

[0027] Evaluate the prediction accuracy of the trained LSTM model on the test set, and adjust the model structure and parameters based on the evaluation results.

[0028] Preferably, the state evaluation function is defined as follows:

[0029] S(st,at)=w1*C(st,at)+w2*T(st,at)

[0030] Where st represents the current state of the device, including device type, location, and status information; at represents the control action taken; w1 and w2 are the weight coefficients of the device state importance and future state trend, respectively, and w1 + w2 = 1; C(st,at) and T(st,at) represent the device state importance evaluation value and future state trend evaluation value after taking action at in state st, respectively.

[0031] Preferably, the control decision module uses the deep deterministic policy gradient DDPG algorithm in the deep reinforcement learning algorithm to train the control strategy model.

[0032] Preferably, the step of training the control strategy model includes:

[0033] Step 1: Set the hyperparameters of the DDPG algorithm, including the learning rate, discount factor, soft update coefficient, and noise attenuation rate; initialize the actor network and critic network, both of which use deep neural network structures. The actor network is used to output the control action under the current state, and the critic network is used to estimate the value of the action;

[0034] Step 2: Define an experience replay buffer to store samples of the interaction between the control process and the environment. Each sample contains the current state, the action taken, the immediate reward obtained, the next state, and a flag indicating whether it is terminated.

[0035] Step 3: The device identification module and the state monitoring module obtain the current state and future state trends of the device in real time and convert them into state representations. Actions are selected based on the current actor network and executed in the distribution cabinet device. Feedback from the environment, including new state information and immediate rewards, is observed. The interaction data is stored in the experience replay buffer.

[0036] Step 4: Randomly draw a batch of samples from the experience replay buffer; use the sample data to update the critic network and optimize the network parameters by minimizing the error between the predicted value and the true value; use the sample data and the output of the critic network to update the actor network and maximize the expected value of the cumulative reward through the policy gradient method;

[0037] Step 5: Repeat steps 3 to 4 until the preset training rounds are reached or the performance of the actor network is stable; during the training process, the parameters of the actor network and the critic network are gradually adjusted according to the soft update coefficient.

[0038] Preferably, the execution control module includes a device control submodule and a feedback monitoring submodule; the device control submodule is used to control the switching, adjustment and protection operations of the equipment inside the distribution cabinet according to the control instructions; the feedback monitoring submodule is used to monitor the response of the equipment in real time and transmit the feedback information to the control decision module for subsequent processing.

[0039] Preferably, the image acquisition module further includes an image enhancement submodule for performing brightness adjustment, contrast enhancement and edge sharpening processing on the acquired image data.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The system uses high-definition cameras to collect real-time image data of equipment inside the distribution cabinet. Utilizing convolutional neural network (CNN) algorithms, it accurately identifies and monitors the status of these devices in real time. This significantly improves the efficiency and accuracy of equipment monitoring while reducing reliance on manual inspections. Combined with the long-short-term memory (LSTM) algorithm, the system performs time series analysis of equipment status to predict future trends. This predictive capability helps identify potential problems in advance, providing data support for preventive maintenance and reducing unplanned downtime.

[0042] The control strategy model constructed using a deep reinforcement learning algorithm enables intelligent control based on the current state and future trends of equipment. Compared to traditional methods based on preset rules and thresholds, this dynamic control approach is more adaptable to the complex operating conditions and changing environmental factors of equipment within distribution cabinets, improving control accuracy and efficiency. Through real-time monitoring, prediction, and intelligent control, this system can promptly identify and address issues with equipment within distribution cabinets, effectively preventing potential failures and significantly improving the stable operation of the power system. The system's high degree of automation can significantly reduce the frequency of manual inspections and operations, thereby saving labor costs and improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a working principle diagram of the power distribution cabinet control system of the present invention;

[0044] Figure 2 Diagram of the steps for building and training a device recognition model;

[0045] Figure 3 A diagram showing the steps for training a control strategy model. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] See also Figure 1-3 The present invention provides a technical solution: a distribution cabinet control system based on machine vision, the system comprising:

[0048] Image Acquisition Module: This module uses high-definition cameras installed inside the power distribution cabinet to capture real-time image data from equipment. These cameras can be configured for either timed or continuous shooting, ensuring real-time monitoring of equipment status. The captured image data then undergoes preprocessing, including noise reduction, contrast enhancement, and size normalization, to form a standardized image dataset for use by subsequent modules.

[0049] Device Identification Module: In this module, a device identification model is constructed using a convolutional neural network (CNN) algorithm. Trained on a large dataset of preprocessed images, this model automatically extracts features from images and accurately classifies devices. The model's output includes device type, location, and current status, which is then passed to the next module for further analysis.

[0050] Condition Monitoring Module: The Condition Monitoring Module receives information from the Device Identification Module and, combined with the Long Short-Term Memory (LSTM) algorithm, builds a model capable of processing time series data. This model analyzes historical device status data and predicts future trends, such as changes in key parameters like temperature and pressure. This predictive capability is crucial for timely identification of potential issues and prevention of failures.

[0051] Control Decision Module: The control decision module is the core component of the system. It uses a deep reinforcement learning algorithm to build a control strategy model. This model generates control instructions based on the output of the condition monitoring module, namely the current state of the equipment and its future state trends. To achieve this, we define a state evaluation function that quantitatively assesses the impact of different control strategies on the equipment state. Based on this evaluation result, the model dynamically adjusts the generated control instructions to ensure stable operation of the power system.

[0052] Execution Control Module: This module receives commands from the control and decision-making module and converts them into specific control signals to operate the equipment within the distribution cabinet. These control signals may include instructions to turn devices on and off, adjust device parameters, and so on. These signals are designed to ensure optimal equipment operation and prevent failures.

[0053] The present invention will be further described below in conjunction with Examples 1 to 3:

[0054] Example 1:

[0055] The device identification model described in this invention is used to accurately identify the type, location, and status of devices within a power distribution cabinet. The following is a specific implementation method for constructing and training the device identification model:

[0056] 1. Build the device recognition model and define the CNN model structure, including:

[0057] Convolutional layers: Convolutional layers are the foundation of CNN models, extracting local features from the input image through convolution operations. In the device recognition model, multiple convolutional layers are designed, each using convolution kernels of different sizes to capture feature information at different scales.

[0058] Pooling layer: The pooling layer is located after the convolutional layer and is used to reduce the dimensionality of the feature map, reduce the amount of computation, and enhance the robustness of the model. In the device recognition model, the maximum pooling operation is used to retain the salient features in the feature map.

[0059] Fully connected layers: These layers integrate previously extracted features to form a global feature representation. In device recognition models, an appropriate number of fully connected layers are designed to ensure effective feature transfer and integration.

[0060] Output layer: The output layer uses a softmax function to convert the output of the fully connected layer into a probability distribution of device type, location, and status. Through the softmax function, the model can output the predicted probability of each category, facilitating subsequent classification decisions.

[0061] Training steps for the device recognition model:

[0062] Collect a training dataset: We collected standard images of various types of PDC equipment, ensuring clear image quality and diverse angles to cover a wide range of scenarios encountered in real-world applications. We annotated each image and generated a corresponding label file. This label file contains information about the device type, location, and status, which is used to monitor the model training process.

[0063] Initialize model parameters: Use an appropriate initialization method, such as He initialization or Xavier initialization, to ensure that model parameters are within a reasonable range of values at the beginning of training. Configure hyperparameters such as the learning rate, batch size, and number of training rounds, as these parameters directly affect the training performance and convergence speed of the model.

[0064] Forward Propagation and Loss Calculation: The training dataset is fed into the model in batches, and the model's output at each layer is calculated through forward propagation. At the output layer, the cross-entropy loss function is used to calculate the loss between the model's predictions and the true labels. The cross-entropy loss function measures the difference between the predicted and true probability distributions and is suitable for multi-classification problems.

[0065] Backpropagation and parameter updating: The backpropagation algorithm calculates the gradient of the loss function with respect to the model parameters. Backpropagation propagates errors layer by layer, enabling the model to learn the mapping from input to output. The stochastic gradient descent (SGD) algorithm or its variants (such as Adam and RMSprop) is used to update the model parameters based on this gradient information. Through continuous iterative optimization, the model's loss on the training dataset is gradually reduced.

[0066] Model Evaluation and Adjustment: Evaluate the recognition accuracy of the trained model on an independent validation set. By comparing the evaluation results under different model structures, parameter configurations, and training strategies, select the best-performing model as the final device recognition model. Based on the evaluation results, adjust the model structure and parameters, such as increasing the number of convolutional layers, changing the convolution kernel size, or adjusting the learning rate, to further improve the model's recognition performance.

[0067] Example 2:

[0068] The condition monitoring model ensures that the system can accurately predict the key components of the future status trend of the equipment. Its network structure specifically includes:

[0069] Input layer: This layer receives time series data of device status. This data is preprocessed into standardized and normalized sequences, which serve as input to the LSTM model.

[0070] LSTM layer: The LSTM layer is the core of the model, capable of memorizing long-term dependencies. Through the gating mechanism within the LSTM unit, the model can capture the changing patterns of device states over time, effectively processing time series data.

[0071] Fully connected layer: The fully connected layer linearly transforms the output of the LSTM layer, mapping it to a trend prediction of the device's future state. This layer helps the model learn and understand the high-level features of the LSTM layer output and convert them into specific prediction values.

[0072] Output layer: The output layer generates the model's predictions about the future state of the device. These predictions can be continuous values, representing the future trend of a certain device parameter, such as temperature or pressure.

[0073] The training steps of the condition monitoring model include:

[0074] Collect historical status data for distribution cabinet equipment, including but not limited to time series records of key parameters such as temperature, humidity, voltage, and current. Preprocess the data, including removing outliers, filling missing values, and standardizing and normalizing the data, to ensure data quality and consistency.

[0075] The preprocessed time series data is divided into a training set and a test set. A label indicating the future state trend is generated for each time series sample in the training set. These labels can be the state value of the device at a certain point in the future, or a trend indicator of state change.

[0076] Initialize the LSTM model's weights and bias parameters. These parameters can be initialized using random initialization or a specific initialization strategy. Set an appropriate learning rate, sequence length (i.e., time step), and number of training epochs. These hyperparameters will affect the model's training speed and performance.

[0077] The training dataset is fed into the LSTM model in batches, and the model's output at each time step is calculated through forward propagation. The mean squared error (MSE) loss function is used to measure the difference between the model's predictions and the true labels. The MSE loss function is suitable for continuous-valued prediction tasks and reflects the average deviation between the predicted and actual values. The gradient of the loss function with respect to the model parameters is calculated through backpropagation, and the Adam optimization algorithm is used to update these parameters. The Adam optimization algorithm combines the principles of momentum and RMSprop to adaptively adjust the learning rate and accelerate the model's convergence.

[0078] Evaluate the prediction accuracy of the trained LSTM model on an independent test set. Evaluate the model's performance by comparing the agreement between the model's predictions and the actual observations. Based on the evaluation results, adjust the model's structure and parameters. This may include increasing or decreasing the number of LSTM layers, adjusting the sequence length, modifying the learning rate, and other strategies to further improve the model's predictive power.

[0079] Example 3:

[0080] The following is the definition of the state evaluation function and the specific implementation method of the control strategy model training in the control decision module:

[0081] The state evaluation function S(st,at) is used to comprehensively evaluate the current state and future state trend of the equipment to guide control decisions. The specific definition is as follows:

[0082] S(st,at)=w1*C(st,at)+w2*T(st,at)

[0083] Where: st represents the current state of the device, including device type, location, and status information, which is obtained in real time by the device identification module. at represents the control action taken by the system, such as turning the device on or off or adjusting device parameters. w1 and w2 are weight coefficients, representing the weights of the device status importance and future status trend, respectively, satisfying w1 + w2 = 1. These weights can be adjusted according to actual needs to balance the impact of the current state and future trends. C(st,at) represents the device status importance assessment value after taking action at in state st, calculated using preset assessment rules or models. T(st,at) represents the future status trend assessment value after taking action at in state st, predicted by the status monitoring module.

[0084] The control decision module uses the Deep Deterministic Policy Gradient (DDPG) algorithm to train the control policy model. The specific steps are as follows:

[0085] Set the hyperparameters of the DDPG algorithm: Initialize the learning rate, discount factor, soft update coefficient, and noise attenuation rate. These parameters have a significant impact on the training speed and convergence of the algorithm.

[0086] The actor network and critic network are constructed using deep neural network structures. The actor network is responsible for outputting control actions based on the current state, while the critic network evaluates the actions taken and gives reward predictions.

[0087] Define an experience replay buffer: Create an experience replay buffer to store sample data generated by interacting with the environment. Each sample includes the current state, the action performed, the immediate reward obtained, the next state after the transition, and a flag indicating whether it has terminated.

[0088] Real-time interaction and data collection: The device identification module and the state monitoring module acquire the current and future state trends of the PDC equipment in real time and convert them into state representations suitable for input into the neural network. Control actions are selected based on the output of the actor network and executed on the PDC equipment. Environmental feedback, including the new state of the equipment and immediate rewards, is observed and stored in the experience replay buffer.

[0089] Network Update: Randomly sample a batch of sample data from the experience replay buffer. Use this batch of sample data to update the critic network, optimizing network parameters by minimizing the error between predicted rewards and actual rewards. Use the updated critic network and sample data to update the actor network, optimizing the control policy by maximizing the expected value of the cumulative reward using policy gradient ascent.

[0090] Iterative training and parameter adjustment: Repeat the above interaction and network update steps until the preset training rounds are reached or the performance of the actor network stabilizes. During the training process, the parameters of the actor network and the critic network are gradually adjusted according to the soft update coefficient to ensure the stability and continuity of learning.

[0091] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0092] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A power distribution cabinet control system based on machine vision, characterized in that: The system comprises: The image acquisition module is used to collect image data of the equipment inside the power distribution cabinet in real time through a high-definition camera, and pre-process the image data to form a standardized image data set; The device identification module is used to build a device identification model. This model is based on the convolutional neural network (CNN) algorithm, which extracts features and classifies devices from preprocessed image data, and outputs device type, location, and status information. The condition monitoring module is used to build a condition monitoring model. This model combines the long short-term memory network (LSTM) algorithm to perform time series analysis on the condition information output by the equipment identification model and predict the future condition trend of the equipment. The control decision module is used to generate equipment control instructions based on the output of the condition monitoring model. Specifically, it uses a deep reinforcement learning algorithm to build a control strategy model, defines a state evaluation function, quantitatively evaluates the control results based on the current state and future state trends of the equipment, and dynamically adjusts the generated control instructions based on the evaluation results. The execution control module is used to receive control instructions and perform corresponding operation control on the equipment inside the distribution cabinet.

2. A power distribution cabinet control system based on machine vision according to claim 1, characterized in that: The device identification model is constructed as follows: Define the CNN model structure, including convolutional layers, pooling layers, fully connected layers, and output layers; the convolutional layers are used to extract image features, the pooling layers are used to reduce feature dimensions, the fully connected layers are used to integrate feature information, and the output layer uses the softmax function to output the probability distribution of device type, location, and status.

3. The power distribution cabinet control system based on machine vision according to claim 2, characterized in that: The training steps of the device recognition model include: Collect training data sets, including standard images of various types of equipment and their labels. The labels include equipment type, location, and status information. Initialize the model parameters, set the learning rate, batch size and training rounds; The training data set is input into the model, and the loss between the model output and the true label is calculated through forward propagation. The loss function adopts the cross entropy loss function; The gradient of the loss function with respect to the model parameters is calculated through back propagation, and the stochastic gradient descent (SGD) algorithm is used to update the model parameters. Evaluate the recognition accuracy of the trained model on the validation set, and adjust the model structure and parameters based on the evaluation results.

4. The power distribution cabinet control system based on machine vision according to claim 1 is characterized in that: The state monitoring model is constructed as follows: Define the LSTM model structure, including the input layer, LSTM layer, fully connected layer, and output layer; the LSTM layer is used to process time series data and capture the changing patterns of device status, and the fully connected layer is used to map the output of the LSTM layer to trend predictions of the device's future status.

5. The power distribution cabinet control system based on machine vision according to claim 4 is characterized in that: The training steps of the condition monitoring model include: Collect training datasets, including labels of historical status sequences of devices and their future status trends; Initialize the parameters of the LSTM model, set the learning rate, sequence length and training rounds; The training data set is input into the LSTM model, and the loss between the model output and the true label is calculated through forward propagation. The loss function adopts the mean square error loss function; The gradient of the loss function with respect to the model parameters is calculated through back propagation, and the Adam optimization algorithm is used to update the model parameters; Evaluate the prediction accuracy of the trained LSTM model on the test set, and adjust the model structure and parameters based on the evaluation results.

6. The power distribution cabinet control system based on machine vision according to claim 1, characterized in that: The state evaluation function is defined as follows: S(st,at)=w1*C(st,at)+w2*T(st,at) Where st represents the current state of the device, including device type, location, and status information; at represents the control action taken; w1 and w2 are the weight coefficients of the device state importance and future state trend, respectively, and w1 + w2 = 1; C(st,at) and T(st,at) represent the device state importance evaluation value and future state trend evaluation value after taking action at in state st, respectively.

7. The power distribution cabinet control system based on machine vision according to claim 6, characterized in that: The control decision module uses the deep deterministic policy gradient DDPG algorithm in the deep reinforcement learning algorithm to train the control strategy model.

8. The power distribution cabinet control system based on machine vision according to claim 7, characterized in that: The steps for training a control strategy model include: Step 1: Set the hyperparameters of the DDPG algorithm, including the learning rate, discount factor, soft update coefficient, and noise attenuation rate; initialize the actor network and critic network, both of which use deep neural network structures. The actor network is used to output the control action under the current state, and the critic network is used to estimate the value of the action; Step 2: Define an experience replay buffer to store samples of the interaction between the control process and the environment. Each sample contains the current state, the action taken, the immediate reward obtained, the next state, and a flag indicating whether it is terminated. Step 3: The device identification module and the state monitoring module obtain the current state and future state trends of the device in real time and convert them into state representations. Actions are selected based on the current actor network and executed in the distribution cabinet device. Feedback from the environment, including new state information and immediate rewards, is observed. The interaction data is stored in the experience replay buffer. Step 4: Randomly draw a batch of samples from the experience replay buffer; use the sample data to update the critic network and optimize the network parameters by minimizing the error between the predicted value and the true value; use the sample data and the output of the critic network to update the actor network and maximize the expected value of the cumulative reward through the policy gradient method; Step 5: Repeat steps 3 to 4 until the preset training rounds are reached or the performance of the actor network is stable; during the training process, the parameters of the actor network and the critic network are gradually adjusted according to the soft update coefficient.

9. The power distribution cabinet control system based on machine vision according to claim 1, characterized in that: The execution control module includes a device control submodule and a feedback monitoring submodule; the device control submodule is used to control the switching, adjustment and protection operations of the equipment inside the distribution cabinet according to the control instructions; the feedback monitoring submodule is used to monitor the response of the equipment in real time and transmit the feedback information to the control decision module for subsequent processing.

10. The power distribution cabinet control system based on machine vision according to claim 1, characterized in that: The image acquisition module also includes an image enhancement submodule for performing brightness adjustment, contrast enhancement and edge sharpening processing on the acquired image data.

Images