Fine energy management method and system for wind power plant

Abstract

The invention relates to the technical field of wind energy management, and discloses a fine energy management method and system for a wind power plant, and the method comprises the following steps: S1, obtaining multi-source data, including real-time operation data, meteorological data and power grid load data of the wind power plant; s2, preprocessing the acquired multi-source data, including data cleaning, denoising, interpolation filling and standardization processing; s3, the generating capacity of the wind power plant is predicted based on a deep learning model, and the deep learning model is a hybrid model and comprises a long and short-term memory network and a convolutional neural network; and S4, using the generating capacity data output by the prediction model, and combining with the power grid load demand. Through deep learning, reinforcement learning and machine learning technologies, the power generation prediction precision of the wind power plant is improved, the scheduling and energy storage management strategy is optimized, and fault prediction and health monitoring of the wind turbine generator are realized, so that the operation efficiency and reliability of the wind power plant are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind energy management, and in particular to a refined energy management method and system for a wind farm. Background Art

[0002] As an important part of renewable energy, wind farms have significant volatility and uncertainty in their power generation. In order to improve the energy utilization efficiency of wind farms and reduce energy waste caused by factors such as weather, energy management of wind farms becomes particularly important.

[0003] Existing wind farm energy management methods improve the operating efficiency of wind farms by combining data prediction, optimized scheduling and energy storage management. For example, the wind power generation prediction method based on meteorological data and historical power generation data can accurately predict the power generation fluctuation of wind farms. Grid load management usually uses traditional linear optimization algorithms or model predictive control (MPC) to balance grid demand and wind power generation capacity.

[0004] Although existing technologies have improved the energy management efficiency of wind farms to a certain extent, there are still some shortcomings. First, existing wind power generation prediction methods mostly rely on a single model, which makes it difficult to fully capture the complex relationship between wind power generation and meteorological factors, resulting in prediction accuracy that needs to be improved. Second, traditional scheduling optimization methods are relatively static and difficult to respond to the rapid changes in wind farm and grid demand in real time, and fail to fully consider the dynamic scheduling of energy storage equipment in a wind power environment with greater volatility. In addition, existing fault prediction technologies mostly rely on the monitoring of single sensor data, which cannot effectively capture the early signs of potential failures of wind turbines, increasing the risk of equipment failure and maintenance costs. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a refined energy management method and system for a wind farm, which solves the problems of insufficient prediction accuracy, inflexible scheduling strategy optimization, and weak fault prediction capability in the prior art wind farm energy management method.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A wind farm refined energy management method, comprising the following steps:

[0007] S1. Obtain multi-source data, including real-time operation data of wind farms, meteorological data and grid load data;

[0008] S2. Preprocess the acquired multi-source data, including data cleaning, denoising, interpolation filling and standardization;

[0009] S3. predicting the power generation of the wind farm based on a deep learning model, where the deep learning model is a hybrid model including a long short-term memory network and a convolutional neural network;

[0010] S4. Using the power generation data output by the prediction model, combined with the load demand of the power grid, intelligent scheduling is performed, and a reinforcement learning algorithm is used to optimize the scheduling strategy;

[0011] S5. Based on the predicted power generation of the wind farm and the grid demand, optimize the charging and discharging management of energy storage equipment and use the reinforcement learning algorithm to adjust the energy storage charging and discharging strategy;

[0012] S6. Collect wind turbine health status data in real time through equipment sensors, use machine learning models to monitor equipment health and predict faults, and generate fault warning signals.

[0013] Preferably, the data preprocessing includes:

[0014] De-noise the real-time data collected by the sensor and use the Kalman filter algorithm or the mean filter algorithm to eliminate outliers;

[0015] Linear interpolation was used to fill in missing data, and the Z-score standardization method was used to normalize all data.

[0016] Preferably, the deep learning model includes:

[0017] The input of the LSTM network is historical wind farm power generation data and meteorological data, and the output is the wind farm power generation forecast value in the future short-term time period, which is used to capture the long-term dependency in time series data;

[0018] The input of the convolutional neural network is meteorological data, and local features are extracted through the convolution layer to extract the spatial features of meteorological data.

[0019] Preferably, the reinforcement learning scheduling optimization adopts a Q-learning algorithm, and the Q value update formula is:

[0020] Q(s t , a t )←Q(s t , a t )+α[r t +γmax a′ Q(s t+1 , a′)-Q(s t , a t )]

[0021] Among them, s t is the state at time t, a t For action, r tis the reward, γ is the discount factor, and α is the learning rate.

[0022] Preferably, the energy storage optimization management includes:

[0023] Based on the predicted wind power generation and grid load demand, reinforcement learning is used to optimize the charging and discharging strategies of energy storage devices to minimize energy storage losses and smooth out the power generation fluctuations of wind farms.

[0024] The reward function of the charging and discharging strategy is:

[0025]

[0026] Among them, P storage is the charging and discharging power of the energy storage device, D forecast For grid load forecasting.

[0027] Preferably, the machine learning model uses a support vector machine and a decision tree algorithm, and the formula for predicting wind turbine faults is:

[0028] P(fault)=classifier(S t )

[0029] Among them, S t is the sensor data of the wind turbine, and classifier is a classifier based on SVM or DT algorithm.

[0030] Preferably, during the scheduling optimization process, the optimization objective function of the intelligent scheduling strategy is:

[0031]

[0032] Among them, P t is the wind power generation at time t, P target is the grid load demand, C i is the operating cost of the i-th wind turbine, α, β, γ are the weighted coefficients of the scheduling objective function, T is the scheduling time period, and n is the number of wind turbines.

[0033] Preferably, the energy storage management optimization further includes the following contents:

[0034] Energy storage strategy model training: Through historical wind farm power generation data, grid load data, and charging and discharging data of energy storage equipment, reinforcement learning is used to train the energy storage management model to maximize the energy matching between wind farms and the grid;

[0035] Charging and discharging priority strategy: During the charging and discharging process of energy storage equipment, the charging and discharging priorities of energy storage equipment are set according to the fluctuations of grid load demand and wind power generation to balance grid demand;

[0036] Real-time scheduling adjustment: During actual operation, the system dynamically adjusts the charging and discharging status of the energy storage equipment according to changes in real-time power generation and load demand to cope with external disturbances.

[0037] Preferably, health monitoring and fault prediction further include:

[0038] Sensor data processing and feature extraction: feature extraction is performed through the sensor data of the wind turbine, and the fault characteristic signals of the wind turbine are extracted by wavelet transform and other methods to establish a fault diagnosis feature library;

[0039] Fault type classification: Use support vector machine or decision tree machine learning models to classify the health status of wind turbines and divide the status into multiple levels;

[0040] Dynamic fault prediction model update: Combine historical fault data with real-time operation data of wind turbines, and dynamically update the fault prediction model through incremental learning or transfer learning to continuously improve the accuracy of fault prediction.

[0041] The present invention also provides a wind farm refined energy management system, comprising:

[0042] Data acquisition module: used to collect wind farm operation data, meteorological data and power grid load data in real time;

[0043] Data processing module: used for denoising, standardization, interpolation and feature extraction of collected data;

[0044] Power generation prediction module: predicts wind farm power generation based on deep learning models LSTM and CNN;

[0045] Dispatch optimization module: optimizes the power generation dispatch strategy of wind farms based on reinforcement learning algorithm;

[0046] Energy storage management module: Based on the prediction results and grid load demand, reinforcement learning is used to optimize the charging and discharging strategies of energy storage equipment;

[0047] Health monitoring module: monitors the health status of wind turbines and predicts faults through sensor data and machine learning models.

[0048] The present invention provides a wind farm refined energy management method and system, which has the following beneficial effects:

[0049] 1. This invention can more accurately predict the power generation of wind farms, especially under complex meteorological conditions, by combining a hybrid deep learning model of long short-term memory network and convolutional neural network. This method significantly improves the accuracy of wind power generation prediction by capturing long-term dependencies and spatial characteristics in time series, providing reliable data support for intelligent scheduling and energy storage management.

[0050] 2. The present invention optimizes the power generation scheduling strategy of the wind farm and the charging and discharging strategy of the energy storage equipment by adopting a reinforcement learning algorithm, which can effectively balance the grid load and wind power fluctuations, reduce energy storage losses, and achieve the best match between wind power and grid demand. Through dynamic scheduling, the system can adapt to changes in the external environment in real time, thereby improving the overall energy efficiency and stability of the system.

[0051] 3. The present invention is based on machine learning algorithms, combined with feature extraction and dynamic update technology of sensor data, which can realize early diagnosis and real-time warning of wind turbine faults, thereby improving the accuracy of fault prediction and reducing equipment downtime and maintenance costs through precise maintenance decisions during the operation of wind farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of the method of the present invention;

[0053] Figure 2 It is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the specification 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.

[0055] Please refer to the attached Figure 1 The embodiment of the present invention provides a method for fine energy management of a wind farm, comprising the following steps:

[0056] S1. Acquire multi-source data, including real-time operation data of wind farms, meteorological data and grid load data. By acquiring multi-source data and monitoring the operation status of wind farms in real time, it is possible to fully grasp the power generation situation of wind farms and grid demand, and provide accurate data support for subsequent dispatch optimization, energy storage management, fault prediction, etc., to ensure refined and optimized management;

[0057] S2. Preprocess the acquired multi-source data, including data cleaning, denoising, interpolation filling and standardization. The effective implementation of data preprocessing ensures high-quality data input, thereby improving the accuracy of subsequent model prediction and scheduling optimization, avoiding the decline of system performance due to data anomalies or noise interference, and enhancing the stability and operability of the entire wind farm energy management method;

[0058] S3. Predict wind farm power generation based on a deep learning model. The deep learning model is a hybrid model that includes a long short-term memory network and a convolutional neural network. By combining the LSTM and CNN models, a more accurate wind power generation forecast can be achieved. This hybrid model can comprehensively consider time series and spatial characteristics, greatly improving the accuracy of wind power generation forecasts, and providing reliable input data for subsequent scheduling optimization, energy storage management, etc., thereby improving the overall operating efficiency of wind farms and the stability of the power system;

[0059] S4. Use the power generation data output by the prediction model, combined with the grid load demand, to conduct intelligent scheduling, and use the reinforcement learning algorithm to optimize the scheduling strategy. Through the reinforcement learning optimization scheduling strategy, the start and stop of the wind turbines and the charging and discharging state of the energy storage system can be automatically adjusted according to the actual power generation situation of the wind farm and the grid demand, so as to maximize the power generation efficiency of the wind farm and the grid load balance, reduce the impact of wind power fluctuations on the grid, and improve the stability of the grid and the economic benefits of the wind farm;

[0060] S5. Based on the predicted power generation and grid demand of the wind farm, the energy storage equipment charging and discharging optimization management is carried out, and the energy storage charging and discharging strategy is adjusted by the reinforcement learning algorithm. The charging and discharging strategy of the energy storage equipment is optimized, which not only improves the energy utilization rate of the wind farm, but also effectively smoothes the fluctuation of wind power generation, avoids the imbalance of grid load caused by wind speed fluctuation, improves the stability of the grid, reduces the loss of energy storage equipment, and improves the overall operation efficiency of the wind farm;

[0061] S6. Real-time data on the health status of wind turbines is collected through equipment sensors. Machine learning models are used to monitor the health of equipment and predict faults, generating fault warning signals. Accurate health monitoring and fault prediction can significantly reduce the unplanned downtime of wind turbines, reduce operation and maintenance costs, extend the service life of equipment, and ensure stable and efficient operation of wind farms. At the same time, fault warnings can arrange maintenance in advance to reduce economic losses caused by faults.

[0062] Please refer to the attached Figure 1 In a preferred embodiment of the present invention, data preprocessing includes:

[0063] De-noising the real-time data collected by the sensor, using the Kalman filter algorithm or the mean filter algorithm to eliminate outliers. The Kalman filter is based on the Bayesian theory. It can effectively suppress errors caused by environmental changes or sensor noise by continuously updating the estimated value using the system's dynamic model and observation model. The mean filter can effectively eliminate fluctuations caused by single abnormal data by calculating the average value of local data. The Kalman filter and the mean filter can effectively reduce the interference of noise on the data, improve the quality and reliability of sensor data, and remove outliers. It can effectively prevent erroneous predictions caused by data anomalies and enhance the stability and prediction ability of the wind farm energy management system.

[0064] Linear interpolation is used to fill in the missing data, and the Z-score standardization method is used to normalize all data. Linear interpolation assumes that the data change is linear and calculates the approximate value of the missing point based on the valid data before and after the missing data point. Specifically, given two known data points x 1 ,y 1 and x 2 ,y 2 The calculation formula for linear interpolation is:

[0065]

[0066] Among them, y is the estimated value of the missing data point, x is the interpolation position, and the principle of linear interpolation is based on the assumption that the data change conforms to the linear law between known data points. The missing data is predicted by the straight line equation. The linear interpolation method can effectively fill the gaps in the data and maintain the continuity of the data sequence. It provides complete input data for subsequent modeling and prediction. The missing data is repaired by interpolation method, which avoids information loss due to missing data and improves data utilization.

[0067] Please refer to the attached Figure 1 In a preferred embodiment of the present invention, the deep learning model includes:

[0068] The input of the long short-term memory network is the historical wind farm power generation data and meteorological data, and the output is the wind farm power generation forecast value in the future short-term time period, which is used to capture the long-term dependency in time series data. LSTM solves the gradient vanishing and gradient explosion problems that may occur in traditional RNN in long sequence learning by introducing "memory units", so that the model can effectively capture long-term dependencies. LSTM can effectively process the time series characteristics in wind farm power generation data and capture the long-term dependency between power generation and historical meteorological data. LSTM can autonomously learn complex time series patterns, reduce human intervention, and improve the automation level and intelligent decision-making ability of wind farm energy management systems;

[0069] The input of the convolutional neural network is meteorological data. Local features are extracted through the convolution layer to extract the spatial features of meteorological data. The convolution operation can identify local patterns related to wind speed changes or changing trends in wind farms from meteorological data. Therefore, the spatial features in meteorological data can be effectively extracted to help accurately predict the power generation of wind farms. Accurate meteorological data features can help the system better predict the power generation of wind farms, thereby achieving more refined scheduling and energy management.

[0070] Please refer to the attached Figure 1 In a preferred embodiment of the present invention, the reinforcement learning scheduling optimization adopts the Q-learning algorithm, and the Q value update formula is:

[0071] Q(s t , a t )←Q(s t , a t )+α[r t +γmax a′ Q(s t+1 , a′)-Q(s t , a t )]

[0072] Among them, s t is the state at time t, a t For action, r t is the reward, γ is the discount factor, α is the learning rate, and Q-learning is a model-free reinforcement learning algorithm that continuously updates the state-action value function Q through the balance between exploration and utilization, and finally converges to the optimal strategy. In each interaction with the environment, the Q-learning algorithm adjusts the action selection according to the current Q value update rule, so that the system can eventually select the optimal scheduling strategy through continuous learning. Q-learning can select the optimal action based on the current state, so that the power generation scheduling strategy of the wind farm can be optimized in real time according to the grid demand and power generation changes, improving the scheduling efficiency. At the same time, by continuously optimizing the scheduling decision, Q-learning can effectively reduce the power generation cost, energy storage loss and grid load fluctuation, thereby improving the economic benefits of the wind farm.

[0073] Please refer to the attached Figure 1 In a preferred embodiment of the present invention, energy storage optimization management includes:

[0074] Based on the predicted wind power generation and grid load demand, reinforcement learning is used to optimize the charging and discharging strategies of energy storage equipment to minimize energy storage losses and smooth out power generation fluctuations in wind farms. The optimization of the charging and discharging strategies of energy storage equipment relies on the core principle of reinforcement learning, that is, learning the optimal behavior strategy through interaction with the environment. By continuously adjusting the charging and discharging operations, the system can find the most suitable strategy in a dynamically changing environment. By continuously optimizing the Q value, the system gradually learns to select the optimal energy storage charging and discharging strategy under different operating conditions to achieve the goals of minimizing energy storage losses, smoothing power generation fluctuations and meeting grid demand.

[0075] The reward function of the charging and discharging strategy is:

[0076]

[0077] Among them, P storage is the charging and discharging power of the energy storage device, D forecast For grid load forecasting, the reward function can balance multiple optimization objectives, including power generation smoothing, energy storage efficiency and grid demand, thereby achieving global optimization of wind farm operation. Through the feedback mechanism of reinforcement learning, the system can flexibly adjust the charging and discharging strategies of energy storage equipment according to different wind farm conditions, thereby improving the adaptability of scheduling decisions.

[0078] Please refer to the attached Figure 1 In a preferred embodiment of the present invention, the machine learning model uses a support vector machine and a decision tree algorithm, and the formula for predicting wind turbine faults is:

[0079] P(fault)=classifier(S t )

[0080] Among them, S t is the sensor data of the wind turbine, and classifier is a classifier based on SVM or DT algorithm. SVM is a classification method based on maximum margin separation. Its goal is to find a hyperplane to separate samples of different categories and make the boundary as far away from the sample point as possible. Decision tree is a classification method based on tree structure. It recursively divides the data set by selecting the optimal feature partitioning data at each node. Through the combination of these two algorithms, the sensor data of the wind turbine can be used to efficiently predict faults, thereby reducing the downtime and maintenance costs of the wind turbine, thereby effectively improving the accuracy of fault prediction, reducing the occurrence of false alarms and missed alarms, timely discovering wind turbine faults, and reducing losses.

[0081] Please refer to the attached Figure 1 In a preferred embodiment of the present invention, during the scheduling optimization process, the optimization objective function of the intelligent scheduling strategy is:

[0082]

[0083] Among them, P t is the wind power generation at time t, P target is the grid load demand, C i is the operating cost of the i-th wind turbine, α, β, γ are the weighted coefficients of the scheduling objective function, T is the scheduling time period, and n is the number of wind turbines. Optimizing the objective function can reduce the fluctuation between wind power generation and grid load, so that the grid can receive the output of the wind farm more stably, thereby ensuring the reliability of power supply. Through precise adjustment of power generation, the frequency of use of energy storage equipment can be reduced, the energy storage loss can be reduced, and the additional loss caused by grid load fluctuation can be avoided.

[0084] Please refer to the attached Figure 1 In a preferred embodiment of the present invention, the energy storage management optimization further includes the following contents:

[0085] Energy storage strategy model training: Through historical wind farm power generation data, grid load data and energy storage equipment charging and discharging data, the energy storage management model is trained using reinforcement learning to maximize the energy matching between the wind farm and the grid. The core principle of the energy storage strategy model is to use reinforcement learning (such as deep Q network DQN) for autonomous learning and decision optimization. Reinforcement learning obtains the optimal strategy from actual operation through continuous exploration and trial and error. By optimizing the charging and discharging strategy of energy storage equipment, it can maximize the balance between the power generation of wind farms and the load demand of the grid, thereby improving the overall energy utilization efficiency of the system, while avoiding unnecessary charging or discharging, reducing the loss of energy storage equipment, and improving the utilization rate of the energy storage system;

[0086] Charging and discharging priority strategy: During the charging and discharging process of energy storage equipment, the charging and discharging priorities of energy storage equipment are set according to the fluctuations of grid load demand and wind power generation to balance grid demand. The charging and discharging priority strategy is optimized based on grid load forecast and wind power generation forecast. By reasonably scheduling the charging and discharging priorities of energy storage equipment, the system can balance wind farm power generation and grid load demand in different scenarios.

[0087] The core principle of this strategy is real-time decision-making based on energy dispatch. Priority allocation ensures that the energy storage system can provide support when electricity demand is tight, so that energy storage can be charged when wind power generation is in excess to achieve the optimal balance of the system. This strategy makes energy storage management more intelligent through prediction and dynamic adjustment, reduces the complexity of manual intervention, and ensures that energy storage equipment is charged and discharged at the most appropriate time, reducing energy loss and improving energy storage efficiency.

[0088] Real-time scheduling adjustment: During actual operation, the system dynamically adjusts the charging and discharging status of the energy storage equipment according to changes in real-time power generation and load demand to cope with external disturbances. The principle of real-time scheduling adjustment is dynamic scheduling based on real-time monitoring and prediction results. Through real-time monitoring of wind power generation, grid load and energy storage status, the system can quickly adjust the charging and discharging strategy. The scheduling process is real-time feedback, which can quickly respond to system changes and ensure that the energy storage equipment is always working in the most suitable state. By dynamically adjusting the charging and discharging status of the energy storage equipment in real time, the system can maximize the use of wind power resources and improve overall energy utilization efficiency.

[0089] Please refer to the attached Figure 1 In a preferred embodiment of the present invention, health monitoring and fault prediction further include:

[0090] Sensor data processing and feature extraction: Feature extraction is performed through the sensor data of the wind turbine, and wavelet transform and other methods are used to extract the fault characteristic signals of the wind turbine, and a fault diagnosis feature library is established. Wavelet transform is an analysis method that decomposes the signal into different frequency bands, which can effectively identify and extract the slight changes of the wind turbine before the failure. In the normal operation of the wind turbine, the sensor data shows regular fluctuations, and when a fault occurs, these fluctuations will be significantly abnormal. The use of wavelet transform can effectively capture these changes. The characteristic signals extracted by wavelet transform can capture abnormal changes in the early stages of wind turbine failure and provide early warning;

[0091] Fault type classification: Use support vector machine (SVM) or decision tree (DT) machine learning models to classify the health status of wind turbines and divide the status into multiple levels. Support vector machine (SVM) is a powerful supervised learning method that achieves data classification by maximizing the boundary interval. In wind turbine fault diagnosis, SVM can efficiently distinguish different types of fault signals. Decision tree (DT) divides input data into multiple categories through a series of "yes" or "no" judgments, and has strong interpretability. Through the training and application of machine learning models, the system can quickly identify the fault type of wind turbines and give diagnostic results in a timely manner;

[0092] Dynamic fault prediction model update: Combine historical fault data and real-time operation data of wind turbines to dynamically update the fault prediction model through incremental learning or transfer learning to continuously improve the accuracy of fault prediction. Incremental learning and transfer learning are two technologies that can enable the model to maintain its learning ability when data continues to grow. Incremental learning can avoid repeated training and gradually update the model by introducing new data. Transfer learning can quickly adapt to new tasks with the help of source task data and improve the generalization ability of the model.

[0093] Through the processing and feature extraction of sensor data, fault type classification and dynamic fault prediction model update, health monitoring and fault prediction provide strong support for the intelligent management of wind turbines. These technologies can provide early warning before a wind turbine fails, perform fault diagnosis in a timely manner, and help operation and maintenance personnel take preventive maintenance measures, thereby improving the operational efficiency of wind farms, reducing downtime and saving maintenance costs.

[0094] A wind farm refined energy management system described below and a wind farm refined energy management method described above may refer to each other.

[0095] Please refer to the attached Figure 2 The present invention also provides a wind farm refined energy management system, including:

[0096] Data acquisition module: used to collect wind farm operation data, meteorological data and power grid load data in real time;

[0097] Data processing module: used for denoising, standardization, interpolation and feature extraction of collected data;

[0098] Power generation prediction module: predicts wind farm power generation based on deep learning models LSTM and CNN;

[0099] Dispatch optimization module: optimizes the power generation dispatch strategy of wind farms based on reinforcement learning algorithm;

[0100] Energy storage management module: Based on the prediction results and grid load demand, reinforcement learning is used to optimize the charging and discharging strategies of energy storage equipment;

[0101] Health monitoring module: monitors the health status of wind turbines and predicts faults through sensor data and machine learning models.

[0102] The system of this embodiment can be executed by the above method embodiment, and its principles and technical effects are similar, which will not be repeated here.

[0103] 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 refined energy management method for a wind farm, characterized in that: The following steps are involved: S1. Obtain multi-source data, including real-time operation data of wind farms, meteorological data and grid load data; S2. Preprocess the acquired multi-source data, including data cleaning, denoising, interpolation filling and standardization; S3. predicting the power generation of the wind farm based on a deep learning model, where the deep learning model is a hybrid model including a long short-term memory network and a convolutional neural network; S4. Using the power generation data output by the prediction model, combined with the load demand of the power grid, intelligent scheduling is performed, and a reinforcement learning algorithm is used to optimize the scheduling strategy; S5. Based on the predicted power generation of the wind farm and the grid demand, optimize the charging and discharging management of energy storage equipment and use the reinforcement learning algorithm to adjust the energy storage charging and discharging strategy; S6. Collect wind turbine health status data in real time through equipment sensors, use machine learning models to monitor equipment health and predict faults, and generate fault warning signals.

2. A wind farm refined energy management method according to claim 1, characterized in that: The data preprocessing includes: De-noise the real-time data collected by the sensor and use the Kalman filter algorithm or the mean filter algorithm to eliminate outliers; Linear interpolation was used to fill in missing data, and the Z-score standardization method was used to normalize all data.

3. A wind farm refined energy management method according to claim 1, characterized in that: The deep learning model includes: The input of the LSTM network is historical wind farm power generation data and meteorological data, and the output is the wind farm power generation forecast value in the future short-term time period, which is used to capture the long-term dependency in time series data; The input of the convolutional neural network is meteorological data, and local features are extracted through the convolutional layer to extract the spatial features of meteorological data.

4. A wind farm refined energy management method according to claim 1, characterized in that: The reinforcement learning scheduling optimization adopts the Q-learning algorithm, and the Q value update formula is: Q(s t ,a t )←Q(s t ,a t )+α[r t +γmax a′ Q(s t+1 ,a′)-Q(s t ,a t )] Among them, s t is the state at time t, a t For action, r t is the reward, γ is the discount factor, and α is the learning rate.

5. A wind farm refined energy management method according to claim 1, characterized in that: The energy storage optimization management includes: Based on the predicted wind power generation and grid load demand, reinforcement learning is used to optimize the charging and discharging strategies of energy storage devices to minimize energy storage losses and smooth out the power generation fluctuations of wind farms. The reward function of the charging and discharging strategy is: Among them, P storage is the charging and discharging power of the energy storage device, D forecast For grid load forecasting.

6. A wind farm refined energy management method according to claim 1, characterized in that: The machine learning model uses a support vector machine and a decision tree algorithm, and the formula for predicting wind turbine faults is: P(fault)=classifier(S t ) Among them, S t is the sensor data of the wind turbine, and classifier is a classifier based on SVM or DT algorithm.

7. A wind farm refined energy management method according to claim 1, characterized in that: During the scheduling optimization process, the optimization objective function of the intelligent scheduling strategy is: Among them, P t is the wind power generation at time t, P target is the grid load demand, C i is the operating cost of the i-th wind turbine, α, β, γ are the weighted coefficients of the scheduling objective function, T is the scheduling time period, and n is the number of wind turbines.

8. A wind farm refined energy management method according to claim 1, characterized in that: The energy storage management optimization further includes the following contents: Energy storage strategy model training: Through historical wind farm power generation data, grid load data, and charging and discharging data of energy storage equipment, reinforcement learning is used to train the energy storage management model to maximize the energy matching between wind farms and the grid; Charging and discharging priority strategy: During the charging and discharging process of energy storage equipment, the charging and discharging priorities of energy storage equipment are set according to the fluctuations of grid load demand and wind power generation to balance grid demand; Real-time scheduling adjustment: During actual operation, the system dynamically adjusts the charging and discharging status of the energy storage equipment according to changes in real-time power generation and load demand to cope with external disturbances.

9. A wind farm refined energy management method according to claim 1, characterized in that: Health monitoring and failure prediction further include: Sensor data processing and feature extraction: feature extraction is performed through the sensor data of the wind turbine, and the fault characteristic signals of the wind turbine are extracted by wavelet transform and other methods to establish a fault diagnosis feature library; Fault type classification: Use support vector machine (SVM) or decision tree (DT) machine learning models to classify the health status of wind turbines and divide the status into multiple levels; Dynamic fault prediction model update: Combine historical fault data with real-time operation data of wind turbines, and dynamically update the fault prediction model through incremental learning or transfer learning to continuously improve the accuracy of fault prediction.

10. A refined energy management system for a wind farm, characterized in that: A wind farm refined energy management method according to any one of claims 1 to 9, comprising: Data acquisition module: used to collect wind farm operation data, meteorological data and power grid load data in real time; Data processing module: used for denoising, standardization, interpolation and feature extraction of collected data; Power generation prediction module: predicts wind farm power generation based on deep learning models LSTM and CNN; Dispatch optimization module: optimizes the power generation dispatch strategy of wind farms based on reinforcement learning algorithm; Energy storage management module: Based on the prediction results and grid load demand, reinforcement learning is used to optimize the charging and discharging strategies of energy storage equipment; Health monitoring module: monitors the health status of wind turbines and predicts faults through sensor data and machine learning models.

Images