A new energy operation system intelligent monitoring and management method

Abstract

The present invention relates to the technical field of new energy management, and discloses an intelligent monitoring and management method for a new energy operation system. The method first collects historical operation data of the new energy operation system, and performs preprocessing to form a feature data set. A state monitoring model based on a long short-term memory network LSTM is constructed to predict the future state of the system. The current configuration parameters and environmental data of the system are collected in real time, and the predicted state data is input into the model to obtain the predicted state data, which is compared with the actual monitoring data to calculate the state deviation. The present invention models the optimization process of the new energy operation system as a reinforcement learning problem, defines the state space, action space, reward function and execution strategy, and adopts a deep deterministic policy gradient DDPG algorithm to select the optimal action. Based on the output of the decision optimization model, the optimal optimization action is automatically executed or executed after confirmation by the management system to ensure the efficient, stable and safe operation of the new energy operation system.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy management, and in particular to an intelligent monitoring and management method for a new energy operation system. Background Art

[0002] With the transformation of the global energy structure and the popularization of the concept of sustainable development, new energy such as solar energy and wind energy have become an important part of the energy field. As the core of new energy power generation, energy storage and distribution, the stable, efficient and safe operation of the new energy operation system is of great significance to ensuring energy supply and reducing carbon emissions. However, due to its own complexity and uncertainty, such as changes in weather conditions, equipment aging, frequent failures, etc., the monitoring and management of the new energy system has brought great challenges. With the rapid development of the Internet of Things, big data and artificial intelligence technologies, it has become possible to use intelligent means to conduct comprehensive and real-time monitoring and management of new energy operation systems.

[0003] Although there are some new energy system monitoring technologies based on data analysis, most of these technologies focus on data collection and simple analysis, lack deep learning and prediction capabilities, and are difficult to accurately predict the future operating status of the system, nor can they automatically adjust optimization strategies according to the current status of the system and environmental changes. In addition, when dealing with complex and changeable new energy system data, existing technologies often have problems such as inaccurate data processing and poor model generalization ability, which seriously affects the accuracy and reliability of monitoring results. Summary of the invention

[0004] The purpose of the present invention is to provide a method for intelligent monitoring and management of a new energy operation system to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for intelligent monitoring and management of a new energy operation system, the method comprising:

[0006] S1: Collect historical operation data of the new energy operation system, including environmental data, system configuration parameters and fault records;

[0007] S2: Preprocess the data collected in S1 to form a feature data set for model training;

[0008] S3: Build a state monitoring model based on the long short-term memory network (LSTM) algorithm. The input of the model is the preprocessed feature data set, and the output is the state prediction of the new energy operation system in the future.

[0009] S4: real-time collection of current system configuration parameters and environmental data of the new energy operation system;

[0010] S5: input the data collected by S4 into the condition monitoring model and receive the predicted condition data output by the model;

[0011] S6: Monitor the new energy operation system in real time, obtain actual status data, and perform anomaly detection on the original monitoring data to remove abnormal values;

[0012] S7: Compare the actual state data obtained by S6 with the predicted state data output by the S5 model, and calculate the state deviation;

[0013] S8: Construct a decision optimization model to model the optimization process of the new energy operation system as a reinforcement learning problem. The specific definitions include:

[0014] State space: defined as the combination of real-time state data, environmental parameters and system configuration parameters of the new energy operation system;

[0015] Action space: defined as the set of optimization operations that can be performed;

[0016] Reward function: designed based on system efficiency, stability, safety, and energy consumption, combined with the state deviation calculated in S7, rewards operations that reduce deviations and penalizes operations that increase deviations;

[0017] Execution strategy: defined as the method of selecting the best action in the current state, using a strategy based on the deep deterministic policy gradient DDPG algorithm;

[0018] S9: Based on the output of the decision optimization model, the best optimization action is selected, and the decision result is automatically executed or transmitted to the operator through the management system for confirmation and execution.

[0019] Preferably, the status data includes power generation power, equipment temperature, voltage, current and frequency; the environmental parameters include ambient temperature, wind speed, wind direction, solar radiation intensity; and the configuration parameters include equipment model, capacity, connection mode and control strategy.

[0020] Preferably, the executable optimization operation set includes:

[0021] i) Adjust the operating parameters of power generation equipment;

[0022] ii) start or stop standby equipment;

[0023] iii) Adjust the charging and discharging strategies of energy storage devices;

[0024] iv) Optimize energy allocation strategy and adjust the output power of each device according to current energy demand and forecast data;

[0025] v) Adjust the equipment maintenance plan and perform maintenance in advance based on the equipment status prediction results.

[0026] Preferably, the network structure of the condition monitoring model specifically includes:

[0027] Input layer: receives feature data sets as input, which include time series status data, environmental parameters, system configuration data, and time characteristics;

[0028] Hidden layer: contains at least one LSTM hidden layer. Each LSTM hidden layer contains several LSTM units. Each LSTM unit has a forget gate, input gate and output gate. The hidden layer is followed by a ReLU activation function.

[0029] Output layer: Receives the output of the hidden layer and maps the output of the LSTM network to the predicted state data through a fully connected layer.

[0030] Preferably, the step of training the state monitoring model comprises:

[0031] S301: Divide the feature data set into a training set, a validation set, and a test set;

[0032] S302: Set the number of layers, number of units per layer, learning rate, batch size, and number of iterations of the LSTM network;

[0033] S303: Train the LSTM model using the training set, and update the model parameters through the back propagation algorithm and the Adam optimizer;

[0034] S304: Evaluate model performance on the validation set and adjust model parameters according to the evaluation results;

[0035] S305: Evaluate the performance of the final model on the test set;

[0036] S306: Save the trained model parameters.

[0037] Preferably, the calculation formula of the reward function R is:

[0038] R = w1 * R_eff + w2 * R_stab + w3 * R_safe + w4 * R_energy + w5 * R_dev

[0039] Among them, R_eff represents the reward item for system operation efficiency, which is calculated based on power generation efficiency and energy utilization rate; R_stab represents the reward item for system stability, which is calculated based on system failure rate and stable operation time; R_safe represents the reward item for system safety, which is calculated based on the frequency and severity of safety incidents; R_energy represents the reward item for energy consumption, which is calculated based on system energy consumption and energy saving; R_dev represents the reward item or penalty item for state deviation, and according to the state deviation calculated in S7, positive rewards are given to operations that reduce deviations, and penalties are given to operations that increase deviations; w1, w2, w3, w4, and w5 are the weight coefficients of each reward item, respectively.

[0040] Preferably, the step of training the decision optimization model includes:

[0041] F1: Initialize the actor network and the critic network;

[0042] F2: Initialize the experience playback buffer to store state transition experience;

[0043] F3: For each training round:

[0044] F3.1: Initialization state at the beginning of the round;

[0045] F3.2: Before the round ends, the following steps are executed in a loop: F3.2.1: The actor network generates actions based on the current state; F3.2.2: Execute the action, observe the reward and next state, and whether it terminates; F3.2.3: Store the experience in the experience replay buffer; F3.2.4: Randomly sample a batch of experience from the experience replay buffer to update the network parameters; F3.2.5: Use the sampled experience to update the parameters of the critic network to minimize the prediction error of the value function; F3.2.6: Use the sampled experience and the evaluation results of the critic network to update the parameters of the actor network to maximize the expected value of the cumulative reward; F3.2.7: Update the current state to the next state;

[0046] F3.3: When the round ends, reset the state to the new round start state;

[0047] F4: Regularly evaluate the performance of the decision optimization model and adjust the weight coefficient, learning rate and exploration noise of the reward function based on the evaluation results;

[0048] F5: Repeat the above steps until the preset number of training rounds is reached.

[0049] Preferably, the method for calculating the state deviation in S7 is:

[0050] S701: For each state parameter, calculate the absolute deviation between the predicted value and the actual value;

[0051] S702: Perform weighted summation on the absolute deviations of all state parameters to obtain a total deviation;

[0052] S703: Compare the total deviation with a preset deviation threshold to determine whether there is a deviation exceeding the threshold.

[0053] Preferably, in S2, the data preprocessing method includes: processing of missing values ​​and outliers and data normalization.

[0054] Preferably, the method for processing missing values ​​includes:

[0055] For numerical features, the median of the non-missing values ​​of the feature is used to fill the missing values;

[0056] For categorical features, the missing values ​​are filled with the mode of the non-missing values ​​of the feature;

[0057] For missing values ​​in time series data, the average value of the adjacent values ​​is used to fill them;

[0058] For features with obvious business meaning and few missing values, they are manually filled according to business logic;

[0059] Methods for dealing with outliers include:

[0060] For numerical features, the 3σ principle is used to identify and process outliers; specifically, the mean and standard deviation of the feature are calculated, and values ​​exceeding the mean ± 3 times the standard deviation are considered outliers.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] Through real-time monitoring and anomaly detection, this method can promptly discover abnormal behaviors in the system and calculate state deviations, providing a basis for taking corrective measures in a timely manner, thereby enhancing the stability of the system.

[0063] The decision optimization model incorporates system operating efficiency, stability, security and energy consumption into the reward function design, ensuring that the optimization actions not only improve system performance but also ensure system security.

[0064] This method constructs a decision optimization model based on reinforcement learning, models the optimization process of the new energy operation system as a reinforcement learning problem, and can automatically learn and find the optimal optimization strategy. The deep deterministic policy gradient (DDPG) algorithm is used as the execution strategy, which can efficiently explore and find the optimal solution in the continuous action space, realizing intelligent decision-making and automatic optimization.

[0065] By comprehensively considering historical operation data, real-time status data and environmental parameters, this method can adapt to the characteristics and changes of different new energy operation systems and improve the adaptability of the system. Both the state monitoring model and the decision optimization model are built based on deep learning algorithms, which have strong generalization capabilities and can handle complex and changeable new energy system data to ensure the accuracy and reliability of the monitoring results. Through intelligent monitoring and management, this method can promptly detect and handle problems in the system, reducing downtime and maintenance costs caused by failures. By optimizing the system operation strategy, energy utilization efficiency is improved, energy consumption and carbon emissions are reduced, which is in line with the concept of sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a working principle diagram of a new energy operation system intelligent monitoring and management method according to the present invention;

[0067] Figure 2 Flowchart for the construction of condition monitoring model. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0069] See also Figure 1-2 The present invention provides a technical solution: a method for intelligent monitoring and management of a new energy operation system, the method comprising:

[0070] S1. Data collection: Collect historical operation data of the new energy operation system, covering environmental data (such as temperature, wind speed, irradiance, etc.), system configuration parameters (such as equipment model, capacity, connection method, etc.) and fault records (including fault type, occurrence time, handling measures, etc.).

[0071] S2. Data preprocessing: Perform preprocessing operations such as cleaning, deduplication, and normalization on the collected data to form a structured and standardized feature data set, providing a basis for subsequent model training.

[0072] S3. Construction of condition monitoring model: The condition monitoring model is constructed based on the long short-term memory network (LSTM) algorithm. The LSTM algorithm is good at processing time series data and can capture long-term dependencies between data. The preprocessed feature data set is used as input to train the LSTM model so that it can predict the state of the new energy operation system in the future, including but not limited to the equipment operation status, energy output, energy storage level, etc.

[0073] S4. Real-time data collection: Real-time collection of current system configuration parameters and environmental data of the new energy operation system to ensure the timeliness and accuracy of the data.

[0074] S5. State prediction: Input the real-time collected data into the trained state monitoring model to obtain the predicted state data output by the model, providing a basis for subsequent monitoring and optimization.

[0075] S6. Real-time monitoring and anomaly detection: Monitor the new energy operation system in real time and obtain actual status data, including but not limited to equipment operation parameters, energy flow data, etc. Perform anomaly detection on the original monitoring data, identify and eliminate abnormal values, and ensure data reliability.

[0076] S7. State deviation calculation: Compare the actual state data obtained by real-time monitoring with the predicted state data predicted by the model, calculate the state deviation, and evaluate the difference between the actual operating state of the system and the expected state.

[0077] S8. Construction of decision optimization model, modeling the optimization process of new energy operation system as a reinforcement learning problem. Specific definitions include:

[0078] State space: It is composed of real-time state data, environmental parameters and system configuration parameters of the new energy operation system, which fully reflects the current status of the system.

[0079] Action space: contains all the optimization operations that can be performed.

[0080] Reward function: Based on the system's operating efficiency, stability, safety, and energy consumption, positive rewards are given to operations that reduce state deviations, and negative rewards are given to operations that increase deviations.

[0081] Execution strategy: Adopt a strategy based on the Deep Deterministic Policy Gradient (DDPG) algorithm, which can efficiently explore and find the optimal action in the continuous action space.

[0082] S8. Optimization decision and execution: Based on the output of the decision optimization model, select the best optimization action. This action aims to minimize state deviation and improve system operation efficiency. The decision results are automatically executed or transmitted to the operator through the management system for confirmation and execution to ensure the effective implementation of the optimization measures.

[0083] The present invention will be further described below in conjunction with Examples 1 to 3: Embodiment 1:

[0084] This embodiment is used to describe in detail the construction and training process of the state monitoring model. The state monitoring model is built based on the long short-term memory network (LSTM) and is used to predict the future state of the new energy operation system. The following is the specific network structure and training steps of the model:

[0085] The network structure of the condition monitoring model includes:

[0086] Input layer: The input layer receives feature data sets as input. Feature data sets include time series status data (such as power generation, equipment temperature, etc.), environmental parameters (such as ambient temperature, wind speed, etc.), system configuration data (such as equipment model, capacity, etc.) and time features (such as hours, dates, etc.). After preprocessing, these feature data are input into the LSTM network in a certain format.

[0087] Hidden layer: The hidden layer contains at least one LSTM hidden layer. Each LSTM hidden layer consists of several LSTM units. Each LSTM unit has a forget gate, input gate, and output gate to control the flow and storage of information. LSTM units can capture long-term dependencies in time series data and are suitable for processing state prediction problems of new energy operation systems. The hidden layer is followed by a ReLU (Rectified Linear Unit) activation function to increase the nonlinear expression ability of the network.

[0088] Output layer: The output layer receives the output of the hidden layer and maps the output of the LSTM network to the predicted state data through a fully connected layer. The predicted state data can include key indicators such as future power generation and equipment temperature.

[0089] Steps to train a condition monitoring model:

[0090] S301 Data division: Divide the feature data set into training set, validation set and test set. The training set is used to train the model, the validation set is used to evaluate the model performance and adjust the model parameters, and the test set is used to evaluate the performance of the final model.

[0091] S302 Parameter setting: Set the hyperparameters of the LSTM network, such as the number of layers, number of units per layer, learning rate, batch size, and number of iterations. The selection of these parameters will affect the training effect and performance of the model.

[0092] S303 Model training: Use the training set to train the LSTM model. Calculate the gradient of the loss function through the back-propagation algorithm, and use the Adam optimizer to update the model parameters to minimize the loss function.

[0093] During the training process, monitor the changes in the loss function to ensure that the model gradually converges.

[0094] S304 Model Validation: Evaluate the performance of the model on the validation set, such as accuracy, mean square error, etc. According to the evaluation results, adjust the model's hyperparameters, such as increasing the number of LSTM layers, adjusting the learning rate, etc., to improve model performance.

[0095] S305 Model Testing: Evaluate the performance of the final model on the test set to verify the generalization ability and prediction accuracy of the model, ensuring that the model can also show good prediction results on unseen data.

[0096] S306 Model saving: Save the trained model parameters, including the weights and biases of the LSTM network. These parameters can be used for subsequent model loading and prediction tasks. Embodiment 2:

[0097] The decision optimization model uses the deep deterministic policy gradient (DDPG) algorithm to optimize the system's operating strategy by training the actor network and the critic network; the reward function of this model is calculated as follows:

[0098] A. Calculation of reward function: The reward function R is used to evaluate the system's operating performance in different states and guide the actor network to generate optimized actions. The calculation formula of the reward function R is:

[0099] R = w1 * R_eff + w2 * R_stab + w3 * R_safe + w4 * R_energy + w5 * R_dev

[0100] in:

[0101] R_eff represents the bonus item of system operation efficiency, which is calculated based on power generation efficiency and energy utilization rate, and reflects the energy conversion efficiency of the system in the current state;

[0102] R_stab represents the bonus item for system stability, which is calculated based on the system failure rate and stable operation time, reflecting the reliability and stability of the system;

[0103] R_safe represents the reward item for system security, which is calculated based on the frequency and severity of security incidents and reflects the security assurance capability of the system;

[0104] R_energy represents the energy consumption reward, which is calculated based on the system energy consumption and energy saving, and encourages the system to reduce energy consumption and improve energy efficiency;

[0105] R_dev represents the reward or penalty term for state deviation, which is calculated based on the deviation between the actual state of the system and the expected state. Positive rewards are given to operations that reduce the deviation, and penalties are given to operations that increase the deviation.

[0106] w1, w2, w3, w4, and w5 are the weight coefficients of each reward item, which are used to adjust the proportion of each reward item in the total reward.

[0107] The training steps of the decision optimization model include:

[0108] F1. Network initialization: Initialize the actor network and the critic network. The actor network is responsible for generating optimized actions, and the critic network is responsible for evaluating the value of state-action pairs.

[0109] F2. Experience replay buffer initialization: Initialize the experience replay buffer to store state transition experience, including the current state, executed actions, rewards obtained, next state, and whether to terminate.

[0110] F3. Training round loop, for each training round, perform the following steps:

[0111] F3.1: Initialize the state at the beginning of the round as the input of the actor network.

[0112] F3.2: Before the round ends, the following sub-steps are executed in a loop:

[0113] F3.2.1: The actor network generates actions based on the current state and outputs them to the system for execution.

[0114] F3.2.2: After executing the action, observe the reward and next state returned by the system, and whether the termination condition is met.

[0115] F3.2.3: Store the current state, action, reward, next state, and whether to terminate information as an experience in the experience replay buffer.

[0116] F3.2.4: Randomly sample a batch of experiences from the experience replay buffer to update the network parameters.

[0117] F3.2.5: Using the sampled experience, update the parameters of the critic network to minimize the value function prediction error.

[0118] F3.2.6: Using the sampled experience and the evaluation results of the critic network, update the parameters of the actor network to maximize the expected value of the cumulative reward.

[0119] F3.2.7: Update the current state to the next state and continue the next cycle.

[0120] F3.3: When the round ends, reset the state to the new round start state and continue training for the next round.

[0121] F4. Model performance evaluation and parameter adjustment: Regularly evaluate the performance of the decision optimization model, such as cumulative rewards, system stability, energy efficiency and other indicators. According to the evaluation results, adjust the weight coefficients w1, w2, w3, w4, w5 of the reward function, as well as parameters such as learning rate and exploration noise to optimize model performance.

[0122] F5. Training termination condition: Repeat the above steps until the preset number of training rounds is reached or other termination conditions are met, such as the model performance reaches the preset target. Embodiment 3:

[0123] The method for calculating the state deviation comprises the following steps:

[0124] S701. Absolute deviation calculation: For each state parameter, calculate the absolute deviation between its predicted value and actual value. The predicted value can be obtained through the state monitoring model or historical data trend analysis, and the actual value is the data collected by the system in real time.

[0125] S702. Weighted summation: Perform weighted summation on the absolute deviations of all state parameters to obtain the total deviation. The weight coefficient is set according to the degree of influence of each state parameter on the overall performance of the system to ensure that the deviation of key parameters accounts for a larger proportion of the total deviation.

[0126] S703. Deviation threshold comparison: Compare the calculated total deviation with the preset deviation threshold. If the total deviation exceeds the preset threshold, it is determined that there is a deviation exceeding the threshold, and further analysis of the cause is required and corresponding adjustment measures are taken; if the total deviation is within the threshold range, the system status is considered normal.

[0127] In step S2, the data preprocessing method includes: processing of missing values ​​and outliers and data normalization; wherein the missing value processing includes:

[0128] Processing of missing values ​​of numerical features: For numerical features, the median of the non-missing values ​​of the feature is used to fill the missing values ​​to ensure the continuity and stability of the data.

[0129] Missing value processing of categorical features: For categorical features, the mode of the non-missing values ​​of the feature is used to fill the missing values ​​to reflect the main distribution trend of the feature in the data set.

[0130] Missing value processing of time series data: For missing values ​​in time series data, the average value of the adjacent values ​​is used to fill them in order to maintain the continuity and smoothness of the time series.

[0131] Handling missing values ​​for features with business significance: For features with obvious business significance and few missing values, they are manually filled in according to business logic to ensure the accuracy and completeness of the data.

[0132] Outlier processing includes: numerical feature outlier processing. For numerical features, the 3σ principle is used to identify and process outliers. Specifically, the mean and standard deviation of the feature are first calculated, and then the value exceeding the mean ±3 times the standard deviation is considered an outlier. For the identified outliers, they can be eliminated, replaced or corrected according to the actual situation to ensure the rationality and reliability of the data.

[0133] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0134] Although 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 the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent monitoring and management of a new energy operation system, characterized in that: The method comprises: S1: Collect historical operation data of the new energy operation system, including environmental data, system configuration parameters and fault records; S2: Preprocess the data collected in S1 to form a feature data set for model training; S3: Build a state monitoring model based on the long short-term memory network (LSTM) algorithm. The input of the model is the preprocessed feature data set, and the output is the state prediction of the new energy operation system in the future. S4: real-time collection of current system configuration parameters and environmental data of the new energy operation system; S5: input the data collected by S4 into the condition monitoring model and receive the predicted condition data output by the model; S6: Monitor the new energy operation system in real time, obtain actual status data, and perform anomaly detection on the original monitoring data to remove abnormal values; S7: Compare the actual state data obtained by S6 with the predicted state data output by the S5 model, and calculate the state deviation; S8: Construct a decision optimization model to model the optimization process of the new energy operation system as a reinforcement learning problem. The specific definitions include: State space: defined as the combination of real-time state data, environmental parameters and system configuration parameters of the new energy operation system; Action space: defined as the set of optimization operations that can be performed; Reward function: designed based on system efficiency, stability, safety, and energy consumption, combined with the state deviation calculated in S7, rewards operations that reduce deviations and penalizes operations that increase deviations; Execution strategy: defined as the method of selecting the best action in the current state, using a strategy based on the deep deterministic policy gradient DDPG algorithm; S9: Based on the output of the decision optimization model, the best optimization action is selected, and the decision result is automatically executed or transmitted to the operator through the management system for confirmation and execution; The executable optimization operation set includes: i) Adjust the operating parameters of power generation equipment; ii) start or stop standby equipment; iii) Adjust the charging and discharging strategies of energy storage devices; iv) Optimize energy allocation strategy and adjust the output power of each device according to current energy demand and forecast data; v) Adjust the equipment maintenance plan and perform maintenance in advance based on the equipment status prediction results; The calculation formula of the reward function R is: R = w1 * R_eff + w2 * R_stab + w3 * R_safe + w4 * R_energy + w5 * R_dev Among them, R_eff represents the reward item of system operation efficiency, which is calculated according to power generation efficiency and energy utilization rate; R_stab represents the reward item of system stability, which is calculated according to system failure rate and stable operation time; R_safe represents the reward item of system safety, which is calculated according to the frequency and severity of safety incidents; R_energy represents the reward item of energy consumption, which is calculated according to system energy consumption and energy saving; R_dev represents the reward item or penalty item of state deviation, according to the state deviation calculated in S7, positive rewards are given to operations that reduce deviations, and penalties are given to operations that increase deviations; w1, w2, w3, w4, and w5 are the weight coefficients of each reward item respectively; The steps to train a decision optimization model include: F1: Initialize the actor network and the critic network; F2: Initialize the experience playback buffer to store state transition experience; F3: For each training round: F3.1: Initialization state at the beginning of the round; F3.2: Before the round ends, the following steps are executed in a loop: F3.2.1: The actor network generates actions based on the current state; F3.2.2: Execute the action, observe the reward and next state, and whether it terminates; F3.2.3: Store the experience in the experience replay buffer; F3.2.4: Randomly sample a batch of experience from the experience replay buffer to update the network parameters; F3.2.5: Use the sampled experience to update the parameters of the critic network to minimize the prediction error of the value function; F3.2.6: Use the sampled experience and the evaluation results of the critic network to update the parameters of the actor network to maximize the expected value of the cumulative reward; F3.2.7: Update the current state to the next state; F3.3: When the round ends, reset the state to the new round start state; F4: Regularly evaluate the performance of the decision optimization model and adjust the weight coefficient, learning rate and exploration noise of the reward function based on the evaluation results; F5: Repeat the above steps until the preset number of training rounds is reached.

2. According to claim 1, a new energy operation system intelligent monitoring and management method is characterized in that: The status data includes power generation, equipment temperature, voltage, current and frequency; the environmental parameters include ambient temperature, wind speed, wind direction, solar radiation intensity; the configuration parameters include equipment model, capacity, connection method and control strategy.

3. According to claim 1, a new energy operation system intelligent monitoring and management method is characterized in that ,The network structure of the state monitoring model specifically includes: Input layer: receives feature data sets as input, which include time series state data, environmental parameters, system configuration data, and time characteristics; Hidden layer: contains at least one LSTM hidden layer. Each LSTM hidden layer contains several LSTM units. Each LSTM unit has a forget gate, input gate and output gate. The hidden layer is followed by a ReLU activation function. Output layer: Receives the output of the hidden layer and maps the output of the LSTM network to the predicted state data through a fully connected layer.

4. The intelligent monitoring and management method for a new energy operation system according to claim 3 is characterized in that: The steps to train a condition monitoring model include: S301: Divide the feature data set into a training set, a validation set and a test set; S302: Set the number of layers, number of units per layer, learning rate, batch size, and number of iterations of the LSTM network; S303: Train the LSTM model using the training set, and update the model parameters through the back propagation algorithm and the Adam optimizer; S304: Evaluate model performance on the validation set and adjust model parameters according to the evaluation results; S305: Evaluate the performance of the final model on the test set; S306: Save the trained model parameters.

5. The intelligent monitoring and management method for a new energy operation system according to claim 1 is characterized in that: The method for calculating the state deviation in S7 is: S701: For each state parameter, calculate the absolute deviation between the predicted value and the actual value; S702: Perform weighted summation on the absolute deviations of all state parameters to obtain a total deviation; S703: Compare the total deviation with a preset deviation threshold to determine whether there is a deviation exceeding the threshold.

6. The intelligent monitoring and management method for a new energy operation system according to claim 1 is characterized in that: In S2, the data preprocessing method includes: processing of missing values ​​and outliers and data normalization.

7. The intelligent monitoring and management method for a new energy operation system according to claim 6 is characterized in that: Methods for handling missing values ​​include: For numerical features, the median of the non-missing values ​​of the feature is used to fill the missing values; For categorical features, the missing values ​​are filled with the mode of the non-missing values ​​of the feature; For missing values ​​in time series data, the average value of the adjacent values ​​is used to fill them; For features with obvious business meaning and few missing values, they are manually filled according to business logic; Methods for dealing with outliers include: For numerical features, the 3σ principle is used to identify and process outliers; specifically, the mean and standard deviation of the feature are calculated, and values ​​exceeding the mean ± 3 times the standard deviation are considered outliers.

Images