Wind power generation efficiency optimization system based on big data
Through the big data optimization system, using technologies such as space-time map deep learning and fuzzy logic control, the wind power generation parameters and energy storage strategies are dynamically adjusted, and the problems of unstable power output, low conversion efficiency and high maintenance costs of the wind power system are solved, and the wind power generation system is realized. More efficient and stable wind power generation and reduced maintenance costs.
Patent Information
- Application Number
- CN202410975359.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing wind power systems face problems such as instability in power output, low conversion efficiency and high maintenance costs, which are difficult to respond quickly to grid demands, and have high maintenance complexity and cost.
Adopting a wind power efficiency optimization system based on big data, using the data acquisition and management module, data analysis and prediction module, dynamic control and optimization module and energy storage and management module, the space-time graph deep learning algorithm, fuzzy logic controller and genetic algorithm , dynamically adjust the operating parameters of wind power generation and the charging and discharging strategies of energy storage systems, and optimize the power output and maintenance plan.
It improves the power output stability and conversion efficiency of the wind power system, reduces maintenance costs, enhances the flexibility and operational efficiency of the system, and ensures the balance of power supply and demand and equipment health management.
Smart Images

Figure BDA0004954373150000041 
Figure BDA0004954373150000043 
Figure BDA0004954373150000051
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wind power generation, and more specifically to a wind power generation efficiency optimization system based on big data. Background Art
[0002] Under the urgent background of global energy transformation and environmental protection, wind power generation, as a clean and renewable energy form, is gradually becoming an important part of the energy structure of countries around the world. Since the oil crisis in the 1970s, wind energy technology has experienced a leap from laboratory to commercial application. Especially in recent decades, thanks to scientific and technological progress and policy support, the cost of wind power generation has dropped significantly and the efficiency has been significantly improved. It has become one of the most competitive new energy sources in the world. At present, with the increasing prominence of climate change issues, countries have increased their investment in wind energy, striving to accelerate the low-carbon transformation of the energy system and achieve sustainable development goals by expanding the scale of wind power generation.
[0003] Modern wind power generation systems are mainly composed of core components such as wind turbines, transmission systems, generators, control systems and grid interfaces. The design of turbine blades follows the principles of aerodynamics and can effectively capture wind energy and convert it into mechanical energy; the transmission system transfers the rotational kinetic energy of the turbine to the generator, which then converts the mechanical energy into electrical energy. The control system is the "brain" of the wind power generation system, responsible for monitoring environmental conditions and adjusting operating parameters to achieve optimal energy conversion efficiency. However, despite continuous technological advances, existing wind power generation systems still face some challenges.
[0004] First, the instability of power output is an inherent problem of wind power generation. The natural fluctuation of wind speed means that wind power generation is difficult to predict and control. This intermittent output is in sharp contrast to the grid's demand for stable and continuous power supply, which brings great challenges to power dispatching. Secondly, the flexibility of wind power generation systems is limited. On the one hand, the start-up and shutdown of wind turbines depend on a specific wind speed range, and it is impossible to adjust the output immediately according to real-time power demand; on the other hand, wind power generation lacks a rapid response mechanism, and it is difficult to quickly increase or decrease production capacity when grid demand changes. This shortcoming is particularly obvious during peak power periods, limiting the potential of wind energy as a baseload power source. Furthermore, high maintenance costs are also a problem that cannot be ignored in wind power generation. Wind turbines are large in structure and are mostly located in remote areas. Daily inspections and fault repairs are not only costly, but also difficult to implement. Offshore wind power projects face additional challenges such as wave erosion and salt spray corrosion, which further increase the complexity and cost of maintenance operations.
[0005] Therefore, the present invention discloses a wind power generation efficiency optimization system based on big data. Summary of the invention
[0006] In view of the shortcomings of the prior art, the present invention discloses a wind power generation efficiency optimization system based on big data. The present invention can improve the power output stability, conversion efficiency and economic benefits of the existing wind power generation system, while significantly reducing maintenance costs.
[0007] In order to achieve the above technical effects, the present invention adopts the following technical solutions:
[0008] A wind power generation efficiency optimization system based on big data, comprising:
[0009] A data acquisition and management module is used to collect and manage real-time and historical data of the wind farm; the data acquisition and management module collects environmental data and equipment status data in real time through a sensor network, and transmits the collected environmental data and equipment status data to a central data management platform through a wireless communication network; the central data management platform sends the data to a data lake for storage; wherein the environmental data at least includes wind speed, wind direction, temperature and humidity; the equipment status data at least includes generator speed, power output, blade angle, bearing temperature, vibration data, as well as voltage, current and frequency;
[0010] A data analysis and prediction module, which is used to train a prediction model based on the historical data stored in the data lake, and analyze and predict the real-time data collected by the data acquisition and management module; the prediction model predicts wind power generation trends and equipment failures through a deep learning algorithm of a spatiotemporal graph; the wind power generation trends include future wind speed, wind direction and power generation;
[0011] A dynamic control and optimization module, used to dynamically adjust the wind power generation operating parameters according to the prediction results of the data analysis and prediction module; based on the prediction model output, the dynamic control and optimization module dynamically adjusts the wind power generation operating parameters through a fuzzy logic controller, and dynamically adjusts the power output to match the grid demand through a mixed integer linear programming method; the wind power generation operating parameters include but are not limited to pitch angle and generator speed;
[0012] The energy storage and management module is used to smooth the power output through the energy storage system according to the prediction results of the data analysis and prediction module; the energy storage and management module integrates the lithium-ion battery energy storage system with the wind farm through the energy storage system integration method, and monitors the battery status through the battery management system BMS; based on the power generation and grid demand predicted by the data analysis and prediction module, the energy storage and management module also optimizes the charging and discharging strategy of the energy storage system through a genetic algorithm.
[0013] As a further technical solution of the present invention,
[0014] The central data management platform includes a data preprocessing module, a data management module and a security and compliance module; the data preprocessing module includes a data receiving unit, a data cleaning unit and a data formatting unit; the data receiving unit receives real-time data streams from the wind farm sensor network through the message publishing / subscribing MQTT protocol; the data cleaning unit identifies and removes outliers through the interquartile range (IQR) method; the data formatting unit ensures the consistency and standardization of data structure through data mapping; the data management module stores and queries data with timestamps through a time series database, and ensures the transparency and traceability of data assets through a metadata registry and metadata management APIs; the security and compliance module automatically evaluates the legality of the data processing process through log auditing and compliance checking algorithms; the compliance checking algorithm interprets contracts, policy documents and regulatory requirements through natural language processing to ensure that business processes and documents comply with regulations, and performs logical judgments based on input data through a rule engine; the rule engine identifies potential violations or operations through logical reasoning and condition matching.
[0015] As a further technical solution of the present invention,
[0016] The security compliance module also ensures the security of data during transmission and storage through symmetric encryption and asymmetric encryption methods, and ensures the security of data access through attribute access control.
[0017] As a further technical solution of the present invention,
[0018] The workflow of the data analysis and prediction module is as follows:
[0019] Step 401: data integration and preprocessing;
[0020] Acquire historical data from the data lake and receive real-time data streams at the same time, and perform data cleaning, format unification, and missing value processing through a distributed data processing system; the distributed data processing system improves data quality through data cleaning algorithms and data conversion methods;
[0021] Step 402: feature engineering;
[0022] Features were extracted from the preprocessed data through principal component analysis, recursive feature elimination, and feature importance scoring for training the prediction model;
[0023] Step 403: model training;
[0024] A prediction model is trained using historical data through a spatiotemporal graph deep learning algorithm to predict wind power generation trends and equipment failures; the spatiotemporal graph deep learning algorithm captures the spatial characteristics of wind field distribution through a convolutional neural network (CNN) and captures time series dependencies through a long short-term memory (LSTM) network, and trains a model to predict future wind speed, wind direction, and power generation;
[0025] Step 404: real-time data analysis;
[0026] Push real-time data to the prediction model through the real-time data stream processing framework, and use the model to output the prediction results;
[0027] Step 405: output and apply the prediction results;
[0028] Feedback the prediction results to the dynamic control and optimization module through the message queue for adjusting the wind turbine operating parameters; the prediction results at least include future wind speed, wind direction and power generation; the operating parameters at least include blade angle and generator speed;
[0029] Step 406: Model verification and update;
[0030] The accuracy and stability of the prediction model are continuously verified through cross-validation and rolling window validation methods. Based on new historical data, the model is periodically updated through online learning methods to adapt to changes in the environment and equipment.
[0031] As a further technical solution of the present invention,
[0032] The working method of the spatiotemporal graph deep learning algorithm in the wind power generation efficiency optimization system is as follows:
[0033] Step 501, data preprocessing and feature engineering;
[0034] The missing values of wind speed data are filled by time series interpolation method, and the wind speed data is standardized by wind speed standardization to construct the wind power data feature set;
[0035] Step 502: construct a space-time graph structure;
[0036] Based on the wind power data feature set, a weighted matrix of spatial distances between nodes is defined by a spatial distance weighted matrix construction formula, and a graph structure is constructed to reflect the spatial relationship between the wind power station and its surrounding environment. The spatial distance weighted matrix construction formula expresses spatial proximity through an exponential decay function; the formula expression of the spatial distance weighted matrix construction formula is:
[0037]
[0038] In formula (1), W igis the spatial distance weight between node i and node g, d ig Represents the geographical distance between node i and node g, which is used to calculate the spatial weight. Represents the attenuation coefficient of the control space influence range, which is used to determine the influence of spatial distance on weight;
[0039] Step 503: spatial feature learning;
[0040] The spatial features between nodes are extracted from the constructed graph structure through the graph convolution network update formula to reflect the correlation between wind speed and wind direction at different locations in the wind power station; the formula expression of the graph convolution network update formula is:
[0041]
[0042] In formula (2), B i (l+1) is the updated feature representation of node i at the (l+1) layer, which is used to capture spatial dependencies; B i (l) represents the feature representation of node j in the lth layer, which is used for aggregation calculation; C(i) represents the neighbor set of node i, which is used to aggregate neighborhood information; N(l) represents the weight matrix of the lth layer, which is used to learn feature transformation; α represents the activation function, which is used to introduce nonlinearity and enhance the expression ability of the model; C ig Represents the normalization factor, which is used to ensure the smoothness of the aggregation operation;
[0043] Step 504: time series modeling;
[0044] The time dynamic characteristics of wind speed changes are captured by the long short-term memory network update function to predict future wind speed and wind direction; the formula expression of the long short-term memory network update function is:
[0045]
[0046] In formula (3), C t represents the unit state, which is used to capture long-term dependencies in the sequence; x is the input vector of the current time step, which carries immediate information; b c Represents the bias term, which is used to adjust the degree of door opening; f t 、i t and t Respectively represent the gating mechanisms of forgetting, input and output information flows, which are used to determine the retention, update and output status of information; W c represents the weight matrix, which is used to learn the mapping between input and state; * represents element-wise multiplication; h t Represents the hidden state, which is passed to the next time step as the output of the model;
[0047] Step 505: integrating spatiotemporal features;
[0048] The final prediction result is generated by combining the spatial features and the time series features through the spatiotemporal feature fusion formula; the spatiotemporal feature fusion formula fuses the H output of the graph convolutional network through weighted summation. (L) and h output by the LSTM network t , generate the comprehensive spatiotemporal feature vector Z t ; The formula expression of the spatiotemporal feature fusion formula is:
[0049]
[0050] In formula (4), Z t represents the fused spatiotemporal feature vector, w s and w t are the weights of spatial features and temporal features, respectively, used to balance the contributions of the two; H (L) represents the output of the last layer of the graph convolutional network, including spatial features; h t Represents the output of the last layer of the long short-term memory network, including time features; W fision represents the weight matrix of the fusion network, which is used to learn feature weights; b fusior Represents the bias term of the fusion network, which is used to adjust the benchmark for weight calculation;
[0051] Step 506: prediction and post-processing;
[0052] Based on the fusion of spatiotemporal features Z t , power generation prediction will be performed through the multi-layer perceptron network MLP.
[0053] As a further technical solution of the present invention,
[0054] The workflow steps of the fuzzy logic controller in the wind power generation efficiency optimization system are as follows:
[0055] Step 601, fuzzification of input variables;
[0056] The input signal is converted into a fuzzy set through a membership function; the input signal includes at least the actual wind speed, wind direction and predicted power generation; the fuzzy set includes three sets of low wind speed, medium wind speed and high wind speed; the membership function adopts the shape of triangle, trapezoid and Gaussian function;
[0057] Step 602: Establishment of fuzzy rules;
[0058] Define the decision logic of the fuzzy logic controller through expert systems and historical data;
[0059] Step 603: fuzzy reasoning;
[0060] According to the current input variables and fuzzy rule base, fuzzy reasoning is performed through the fuzzy reasoning method Sugeno to calculate the output fuzzy set; the fuzzy reasoning method obtains the fuzzy set of the output variable through the activation, aggregation and defuzzification of fuzzy rules;
[0061] Step 604, defuzzification of the output;
[0062] The fuzzy set obtained by fuzzy reasoning is converted into a numerical value as an actual control signal through the center of gravity method, the maximum membership method or the weighted average method;
[0063] Step 605: Dynamic adjustment and optimization;
[0064] The control signal obtained by defuzzification is used as input to adjust the pitch angle and generator speed through the wind turbine control system to optimize the power generation efficiency in real time. In addition, the power output is dynamically adjusted through the mixed integer linear programming method to achieve the balance of the power system.
[0065] Step 606: real-time feedback and correction;
[0066] The fuzzy rules and membership functions are learned and adjusted online based on real-time monitoring system performance and feedback through minimum mean square error and recursive least squares method.
[0067] As a further technical solution of the present invention,
[0068] The mixed integer linear programming method first constructs an efficiency optimization model containing integer variables and continuous variables through an objective function; then, the mixed integer linear programming method explores the solution space through branch and bound method, cutting plane method and branch and price method, and uses linear relaxation to approximate the optimal solution, gradually narrowing the search range of integer solutions until the global optimal solution is found or the preset termination criteria are met; wherein, the branch and bound method improves the solution efficiency through solution space partitioning and boundary condition pruning; the cutting plane method approximates the integer solution through constraint conditions to avoid redundant search paths; the constraint conditions include but are not limited to power grid power balance, wind turbine operating limits, equipment maintenance cycle and electricity market price fluctuations; according to the global optimal solution obtained through exploration, the mixed integer linear programming method obtains real-time wind speed, wind direction, power generation forecast and power grid demand data from the data analysis and prediction module through a data stream processing framework, and dynamically updates the efficiency optimization model parameters.
[0069] As a further technical solution of the present invention,
[0070] The energy storage and management module includes an energy storage system integrated control submodule, a battery management submodule and an energy storage strategy optimization submodule; the energy storage system integrated control submodule includes an energy storage system interface unit and a wind farm integration unit; the energy storage system interface unit exchanges data with the lithium-ion battery energy storage system through a standardized communication protocol; the wind farm integration unit connects the energy storage system to the AC power grid of the wind farm through a power electronic converter; the battery management submodule includes a battery status monitoring unit, a battery balancing control unit and a fault diagnosis and protection unit; the battery status monitoring unit continuously monitors at least the voltage, current and temperature parameters of the battery through a sensor array; the battery balancing control unit adopts a passive balancing strategy, adjusts the voltage difference between the cells in the battery pack through a balancing circuit, and prolongs the battery life; the fault diagnosis and protection unit identifies battery faults through Kalman filtering, and prevents battery overcharging, over-discharging and overheating through a protection circuit; the energy storage strategy optimization submodule generates a charging and discharging strategy of the energy storage system through a genetic algorithm based on the wind farm power generation and grid demand predicted by the data analysis and prediction module, and dynamically adjusts the charging and discharging plan of the energy storage system through a fuzzy logic control strategy.
[0071] As a further technical solution of the present invention,
[0072] The working steps of the genetic algorithm are:
[0073] Step 901: Convert decision variables into genes on chromosomes through binary coding; the decision variables include but are not limited to charging and discharging power and time parameters;
[0074] Step 902: randomly generate an initial working strategy through a random number generator to form a population;
[0075] Step 903: Evaluate the effectiveness of each strategy based on the goal of minimizing cost or maximizing benefit through a fitness function;
[0076] Step 904: select individuals according to fitness values by proportional selection, tournament selection and roulette selection methods to enter the mating pool and prepare for genetic operations;
[0077] Step 905: select some or all features from the parent strategy through the crossover operator and combine them to generate a new offspring strategy;
[0078] Step 906: Increase the diversity of strategies through mutation operators to prevent the algorithm from converging to a local optimal solution too early;
[0079] Step 907: Replace part of the old population with the newly generated offspring strategy by overlapping generations;
[0080] Step 908, checking the algorithm stopping condition through the maximum number of iterations, fitness threshold and population diversity level;
[0081] Step 909, repeating steps 903 to 907 until the convergence condition is met;
[0082] Step 910: When the algorithm ends, the individual with the highest fitness value is identified and output from the current population.
[0083] As a further technical solution of the present invention,
[0084] The wind power generation efficiency optimization system based on big data also includes a decision support and optimization strategy module, which is used to provide operation strategy suggestions, optimize asset allocation, maintenance plans and market trading strategies; based on the data analysis results, the decision support and optimization strategy module comprehensively considers power generation, equipment health and energy storage status through a particle swarm optimization algorithm, optimizes asset allocation and maintenance plans, and combines the predicted power generation and market prices to formulate market trading strategies using a reinforcement learning algorithm.
[0085] Positive beneficial effects:
[0086] The present invention predicts the trend of wind power generation through a deep learning algorithm of a spatiotemporal graph. The system can know the changes in future wind speed, wind direction and power generation in advance, so that the dynamic control and optimization module can timely adjust the wind power generation operating parameters, such as the pitch angle and the generator speed, to maintain a stable power output, reduce the power fluctuation caused by wind fluctuations, and improve the power quality and reliability. In addition, the efficient prediction capability of the data analysis and prediction module allows the system to plan the optimal operating state in advance, avoiding unnecessary energy loss. At the same time, the dynamic control and optimization module dynamically adjusts the power output through a mixed integer linear programming method to ensure that power production is accurately matched with grid demand, reducing energy waste and improving the overall conversion efficiency of wind power generation. Secondly, through the early warning of equipment failures by the prediction model, the energy storage and management module can optimize the charging and discharging strategy of the energy storage system, avoid battery loss caused by frequent charging and discharging due to power fluctuations, extend the service life of the energy storage system, and reduce maintenance costs. At the same time, the prediction model can also guide the maintenance team to perform preventive maintenance, reduce the occurrence of sudden failures, and further save maintenance costs. The application of fuzzy logic controller and genetic algorithm enables the system to intelligently cope with complex and changing operating environments, and realizes refined management of wind power generation through dynamic adjustment and optimization strategies. In addition, the decision support provided by the data analysis and prediction module provides operators with data-based strategy suggestions, reduces the risk of decision-making errors, and enhances the overall operational efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0087] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:
[0088] Figure 1 A framework diagram of a wind power generation efficiency optimization system based on big data according to the present invention;
[0089] Figure 2 A diagram showing the working principle and steps of the spatiotemporal graph deep learning algorithm of the present invention;
[0090] Figure 3 It is a working principle step diagram of the fuzzy logic controller of the present invention;
[0091] Figure 4 This is a principle framework diagram of the energy storage and management module of the present invention;
[0092] Figure 5 It is a working step diagram of the genetic algorithm of the present invention. DETAILED DESCRIPTION
[0093] 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.
[0094] like Figure 1-Figure 5 As shown, a wind power generation efficiency optimization system based on big data includes:
[0095] A data acquisition and management module is used to collect and manage real-time and historical data of the wind farm; the data acquisition and management module collects environmental data and equipment status data in real time through a sensor network, and transmits the collected environmental data and equipment status data to a central data management platform through a wireless communication network; the central data management platform sends the data to a data lake for storage; wherein the environmental data at least includes wind speed, wind direction, temperature and humidity; the equipment status data at least includes generator speed, power output, blade angle, bearing temperature, vibration data, as well as voltage, current and frequency;
[0096] A data analysis and prediction module, which is used to train a prediction model based on the historical data stored in the data lake, and analyze and predict the real-time data collected by the data acquisition and management module; the prediction model predicts wind power generation trends and equipment failures through a deep learning algorithm of a spatiotemporal graph; the wind power generation trends include future wind speed, wind direction and power generation;
[0097] A dynamic control and optimization module, used to dynamically adjust the wind power generation operating parameters according to the prediction results of the data analysis and prediction module; based on the prediction model output, the dynamic control and optimization module dynamically adjusts the wind power generation operating parameters through a fuzzy logic controller, and dynamically adjusts the power output to match the grid demand through a mixed integer linear programming method; the wind power generation operating parameters include but are not limited to pitch angle and generator speed;
[0098] The energy storage and management module is used to smooth the power output through the energy storage system according to the prediction results of the data analysis and prediction module; the energy storage and management module integrates the lithium-ion battery energy storage system with the wind farm through the energy storage system integration method, and monitors the battery status through the battery management system BMS; based on the power generation and grid demand predicted by the data analysis and prediction module, the energy storage and management module also optimizes the charging and discharging strategy of the energy storage system through a genetic algorithm.
[0099] In a specific embodiment, the core of the data acquisition and management module is to establish a comprehensive sensor network, covering anemometers, wind vanes, temperature sensors, humidity sensors, and sensors for monitoring generator speed, power output, blade angle, bearing temperature, vibration data, and electrical parameters (voltage, current, frequency). These sensors continuously collect real-time data and transmit the data to the central data management platform through wireless communication networks (such as LoRa, NB-IoT or 5G). The platform uses advanced data cleaning, preprocessing and storage technologies to ensure the accuracy and completeness of the data, and then stores the data in the data lake, laying the foundation for subsequent deep learning and data analysis. The data analysis and prediction module uses the spatiotemporal graph convolutional neural network (ST-GCN) in deep learning to train the historical data stored in the data lake and build a prediction model. ST-GCN can simultaneously capture the spatial dependencies between the wind turbines in the wind power station and the dynamic characteristics of time series data. Specifically, by constructing the spatiotemporal graph structure of the wind farm, treating the wind turbines as nodes in the graph, and treating the spatial distance between wind turbines and the correlation between wind direction and wind speed as the weight of the edge, ST-GCN can learn the power generation efficiency of each wind turbine at different wind speeds and wind directions, as well as the mode of equipment status change. The prediction model can not only predict future wind speed, wind direction and power generation, but also predict potential signs of equipment failure and provide early warning for maintenance plans.
[0100] Based on the output of the data analysis and prediction module, the dynamic control and optimization module optimizes the operating parameters of wind power generation and adjusts the power output through the fuzzy logic controller (FLC) and the mixed integer linear programming (MILP) algorithm. FLC uses fuzzy logic rules to dynamically adjust the pitch angle and generator speed according to the predicted wind speed and power generation to achieve maximum wind energy capture and minimum mechanical stress. MILP considers the grid demand, power market rules and energy storage status, and dynamically adjusts the power output by solving the optimization problem to ensure the balance of power supply and demand while maximizing economic benefits. The energy storage and management module integrates the lithium-ion battery energy storage system (BESS) and optimizes the charging and discharging strategy using the genetic algorithm (GA) to achieve smooth power output. The integration of BESS and the wind farm monitors the battery status in real time through the battery management system (BMS), including SOC (state of charge), temperature and voltage, to ensure the safe operation of the battery. The GA algorithm optimizes the charging and discharging strategy of the energy storage system according to the predicted power generation and grid demand to minimize power fluctuations while maximizing the service life and economic benefits of the energy storage system. The specific implementation process of this system is as follows:
[0101] Deploy sensor networks: Install sensors at key locations in wind farms to ensure comprehensive data coverage.
[0102] Data transmission and storage: Sensor data is transmitted to the central data management platform in real time via wireless networks and stored in the data lake after preprocessing.
[0103] Training prediction models: Use ST-GCN to train historical data and build prediction models for wind power generation trends and equipment failures.
[0104] Dynamic control and optimization: Based on the prediction results, the operating parameters are adjusted through FLC and the power output is optimized through MILP.
[0105] Energy storage strategy optimization: Combined with the predicted power generation and grid demand, the GA algorithm is used to optimize the charging and discharging strategy of BESS to achieve smooth power output.
[0106] The implementation of this system has significantly improved the efficiency and reliability of wind power generation, solving the problems of unstable power output, low conversion efficiency and high maintenance costs. Through deep learning prediction models, wind farms are able to plan operation strategies in advance, reduce power fluctuations, and improve power quality and conversion efficiency. The dynamic control and optimization module ensures the optimal performance of wind power generation, while the integration of the energy storage system provides additional flexibility to adapt to changes in grid demand. In addition, the implementation of the intelligent maintenance plan significantly reduces equipment failure rate and maintenance costs, and extends equipment life. Overall, this system has promoted the development of the wind power industry towards a more intelligent, efficient and sustainable direction through technological innovation.
[0107] In the above embodiment, the central data management platform includes a data preprocessing module, a data management module and a security compliance module; the data preprocessing module includes a data receiving unit, a data cleaning unit and a data formatting unit; the data receiving unit receives real-time data streams from the wind farm sensor network through the message publishing / subscribing MQTT protocol; the data cleaning unit identifies and removes outliers through the interquartile range IQR method; the data formatting unit ensures the consistency and standardization of data structure through data mapping; the data management module stores and queries data with timestamps through a time series database, and ensures the transparency and traceability of data assets through a metadata registry and metadata management APIs; the security compliance module automatically evaluates the legality of the data processing process through log auditing and compliance checking algorithms; the compliance checking algorithm interprets contracts, policy documents and regulatory requirements through natural language processing to ensure that business processes and documents comply with regulations, and performs logical judgments based on input data through a rule engine; the rule engine identifies potential violations or operations through logical reasoning and condition matching.
[0108] The security compliance module also ensures the security of data during transmission and storage through symmetric encryption and asymmetric encryption methods, and ensures the security of data access through attribute access control.
[0109] In a specific embodiment, the data preprocessing module is responsible for ensuring the quality and consistency of the data. The module uses advanced data cleaning techniques, including deduplication, filling missing values, correcting erroneous readings, and outlier detection. In addition, the module also performs data standardization and format conversion to meet the requirements of subsequent data analysis and machine learning models. The algorithms used in the preprocessing process include, but are not limited to, K nearest neighbor (KNN), principal component analysis (PCA), and time series interpolation to ensure the accuracy and integrity of the data. The data management module is the manager of the data lake and is responsible for the storage, retrieval, and backup of data. It uses a distributed file system, such as Hadoop Distributed File System (HDFS), to ensure high availability and fault tolerance. The module also uses metadata management tools, such as Apache Hive or Apache Drill, to provide data indexing and query functions, making data retrieval fast and efficient. The data management module also supports data lifecycle management, including data classification, archiving, and deletion policies to optimize the use of storage resources. The security and compliance module ensures the security and compliance of data to prevent data leakage, tampering, and unauthorized access. It uses encryption technology and access control lists (ACLs) to protect data, while complying with relevant regulatory requirements such as GDPR and HIPAA. The module also implements audit and monitoring mechanisms to record the history of data access and modification for easy tracking and accountability. In addition, the module also integrates a disaster recovery plan, including regular data backup and redundant storage to ensure data recoverability in the event of a failure.
[0110] In the wind power generation efficiency optimization system based on big data, the specific implementation process of the central data management platform is as follows: the wind farm sensor network collects environmental and equipment data in real time and transmits it to the central data management platform via the wireless communication network. The data preprocessing module is immediately started to perform operations such as data cleaning, format conversion and standardization to ensure data quality. The preprocessed data is sent to the data lake, and the data management module classifies and stores it according to data types and attributes, and creates metadata indexes to facilitate subsequent retrieval and analysis. At the same time, the module implements data lifecycle management strategies to ensure effective storage and reasonable use of data. The security and compliance module starts data encryption and access control to ensure the safe storage of data. The module also regularly performs data backup and disaster recovery drills to deal with possible data loss or system failures. In addition, the module also conducts compliance reviews to ensure that the data processing process complies with relevant laws and regulations.
[0111] In the above embodiment, the workflow of the data analysis and prediction module is as follows:
[0112] Step 401: data integration and preprocessing;
[0113] Acquire historical data from the data lake and receive real-time data streams at the same time, and perform data cleaning, format unification, and missing value processing through a distributed data processing system; the distributed data processing system improves data quality through data cleaning algorithms and data conversion methods;
[0114] Step 402: feature engineering;
[0115] Features were extracted from the preprocessed data through principal component analysis, recursive feature elimination, and feature importance scoring for training the prediction model;
[0116] Step 403: model training;
[0117] A prediction model is trained using historical data through a spatiotemporal graph deep learning algorithm to predict wind power generation trends and equipment failures; the spatiotemporal graph deep learning algorithm captures the spatial characteristics of wind field distribution through a convolutional neural network (CNN) and captures time series dependencies through a long short-term memory (LSTM) network, and trains a model to predict future wind speed, wind direction, and power generation;
[0118] Step 404: real-time data analysis;
[0119] Push real-time data to the prediction model through the real-time data stream processing framework, and use the model to output the prediction results;
[0120] Step 405: output and apply the prediction results;
[0121] Feedback the prediction results to the dynamic control and optimization module through the message queue for adjusting the wind turbine operating parameters; the prediction results at least include future wind speed, wind direction and power generation; the operating parameters at least include blade angle and generator speed;
[0122] Step 406: Model verification and update;
[0123] The accuracy and stability of the prediction model are continuously verified through cross-validation and rolling window validation methods. Based on new historical data, the model is periodically updated through online learning methods to adapt to changes in the environment and equipment.
[0124] In the above embodiment, the working method of the spatiotemporal graph deep learning algorithm in the wind power generation efficiency optimization system is:
[0125] Step 501, data preprocessing and feature engineering;
[0126] The missing values of wind speed data are filled by time series interpolation method, and the wind speed data is standardized by wind speed standardization to construct the wind power data feature set;
[0127] Step 502: construct a space-time graph structure;
[0128] Based on the wind power data feature set, a weighted matrix of spatial distances between nodes is defined by a spatial distance weighted matrix construction formula, and a graph structure is constructed to reflect the spatial relationship between the wind power station and its surrounding environment. The spatial distance weighted matrix construction formula expresses spatial proximity through an exponential decay function; the formula expression of the spatial distance weighted matrix construction formula is:
[0129]
[0130] In formula (1), W ig is the spatial distance weight between node i and node g, d ig Represents the geographical distance between node i and node g, which is used to calculate the spatial weight. Represents the attenuation coefficient of the control space influence range, which is used to determine the influence of spatial distance on weight;
[0131] Step 503: spatial feature learning;
[0132] The spatial features between nodes are extracted from the constructed graph structure through the graph convolution network update formula to reflect the correlation between wind speed and wind direction at different locations in the wind power station; the formula expression of the graph convolution network update formula is:
[0133]
[0134] In formula (2), B i (l+1) is the updated feature representation of node i at the (i+1) layer, which is used to capture spatial dependencies; B i (l) represents the feature representation of node j in the lth layer, which is used for aggregation calculation; C(i) represents the neighbor set of node i, which is used to aggregate neighborhood information; N(l) represents the weight matrix of the lth layer, which is used to learn feature transformation; α represents the activation function, which is used to introduce nonlinearity and enhance the expression ability of the model; C ig Represents the normalization factor, which is used to ensure the smoothness of the aggregation operation;
[0135] Step 504: time series modeling;
[0136] The time dynamic characteristics of wind speed changes are captured by the long short-term memory network update function to predict future wind speed and wind direction; the formula expression of the long short-term memory network update function is:
[0137]
[0138] In formula (3), C t represents the unit state, which is used to capture long-term dependencies in the sequence; x is the input vector of the current time step, which carries immediate information; bc Represents the bias term, which is used to adjust the degree of door opening; f t 、i t and t Respectively represent the gating mechanisms of forgetting, input and output information flows, which are used to determine the retention, update and output status of information; W c represents the weight matrix, which is used to learn the mapping between input and state; * represents element-wise multiplication; h t Represents the hidden state, which is passed to the next time step as the output of the model;
[0139] Step 505: integrating spatiotemporal features;
[0140] The final prediction result is generated by combining the spatial features and the time series features through the spatiotemporal feature fusion formula; the spatiotemporal feature fusion formula fuses the H output of the graph convolutional network through weighted summation. (L) and h output by the LSTM network t , generate the comprehensive spatiotemporal feature vector Z t ; The formula expression of the spatiotemporal feature fusion formula is:
[0141]
[0142] In formula (4), Z t represents the fused spatiotemporal feature vector, w s and w t are the weights of spatial features and temporal features, respectively, used to balance the contributions of the two; H (L) represents the output of the last layer of the graph convolutional network, including spatial features; h t Represents the output of the last layer of the long short-term memory network, including time features; W fision represents the weight matrix of the fusion network, which is used to learn feature weights; b fusior Represents the bias term of the fusion network, which is used to adjust the benchmark for weight calculation;
[0143] Step 506: prediction and post-processing;
[0144] Based on the fusion of spatiotemporal features Z t , power generation prediction will be performed through the multi-layer perceptron network MLP.
[0145] In a specific embodiment of a wind power generation efficiency optimization system based on big data, a spatiotemporal graph deep learning algorithm integrates the theories of deep learning, spatiotemporal sequence analysis, and graph theory, aiming to process and predict the complex dynamic behavior of wind power generation. The algorithm captures the dynamic characteristics of the mutual influence between wind turbines in space and the change of wind force in time by constructing a spatiotemporal graph model of a wind farm, thereby realizing accurate prediction of wind power generation trends and equipment failures. Specifically, a wind farm is regarded as a graph, in which wind turbines are nodes, and the spatial distance between nodes and the correlation between wind direction and wind speed are used as edge weights. By defining appropriate similarity or distance metrics, a weighted adjacency matrix is constructed to reflect the spatial dependency between wind turbines. A graph convolutional network (GCN) is a neural network that can perform convolution operations on a graph structure to capture the spatial dependency between nodes in a graph. By defining a graph convolution layer, GCN can learn the spatial correlation pattern between wind turbines from the spatiotemporal graph of a wind farm. Then, a recurrent neural network (RNN) or its variants, such as a long short-term memory network (LSTM) and a gated recurrent unit (GRU), is used to model the time series data of the wind farm. These networks can capture long-term dependencies in time series data and predict future wind speed, wind direction, and power generation. By fusing the outputs of GCN and RNN / GRU, a spatiotemporal feature vector is formed to fully reflect the dynamic behavior of the wind farm. The fusion strategy can be a simple weighted sum, splicing, or a more complex attention mechanism to adaptively adjust the weights of spatial and temporal features. In addition to wind power generation trend forecasting, spatiotemporal graph deep learning algorithms can also predict equipment failures and achieve preventive maintenance by analyzing abnormal patterns in equipment status data.
[0146] In the wind power generation efficiency optimization system based on big data, the specific implementation process of the spatiotemporal graph deep learning algorithm is as follows:
[0147] Data preparation: Extract historical wind power generation data and equipment status data from the data lake, including wind speed, wind direction, power generation, generator speed, temperature, vibration and other indicators.
[0148] Graph construction and feature extraction: A spatiotemporal graph of the wind farm is constructed based on the geographical location of the wind turbines and the wind direction and speed data, and then GCN is used to extract spatial features.
[0149] Time Series Modeling: Use LSTM or GRU to model time series data and capture temporal dependencies.
[0150] Model training and optimization: Input spatiotemporal features into the prediction model, train the model through supervised learning, use historical data as the training set, use the difference between the predicted value and the actual value as the loss function, and optimize the model parameters through the back propagation algorithm.
[0151] Prediction and evaluation: Use the trained model to predict real-time data, evaluate the accuracy and stability of the prediction results, and adjust the model parameters if necessary.
[0152] Specifically, Experiment 1 is designed to verify the effectiveness of the spatiotemporal graph deep learning algorithm in predicting wind power generation and equipment status; and to evaluate the prediction accuracy of the algorithm and the impact of model parameters on the prediction results;
[0153] Specific experimental method: First, extract data such as wind speed, wind direction, power generation, and equipment status from the historical data of the wind farm. Then fill in the missing values, standardize the wind speed data, and construct a wind power data feature set. Then, use formula (1) to calculate the spatial distance weight between nodes and construct a graph structure to reflect the spatial relationship of the wind farm. Use formula (2) to extract the spatial characteristics between wind turbines in the wind farm. Apply formula (3) to capture the temporal dynamic characteristics of wind speed and wind direction. Use formula (4) to combine spatial and temporal features to generate a comprehensive spatiotemporal feature vector. Then, adjust the learning rate, optimize the model parameters, and record the key indicators of each iteration. Based on the fused spatiotemporal features, use a multi-layer perceptron (MLP) to predict power generation.
[0154] The experimental environment includes: Hardware: high-performance server, equipped with GPU accelerated computing. Software: Python programming language, PyTorch deep learning framework. The data record table is shown in Table 1:
[0155] Table 1 Algorithm running process record
[0156]
[0157]
[0158] In data table 1, iteration rounds refer to the number of rounds of model training, and each round will traverse the entire training data set once. The learning rate refers to the speed of controlling the update of model parameters. It is usually high in the early stage of iteration and gradually decreases in the later stage to stabilize the model. The GCN weight matrix and LSTM weight matrix represent the weight matrices of the graph convolutional network and the long short-term memory network, respectively, and are updated as the iteration proceeds. The spatial feature weights and temporal feature weights represent the weights used when fusing spatiotemporal features. Their values will be adjusted according to the performance during the training process to achieve the best prediction effect. The training loss and validation loss represent the loss function values on the training set and validation set, respectively, reflecting the degree of fit of the model on the data. The prediction accuracy represents the prediction accuracy of the model on the test set, which is an important indicator for measuring the generalization ability of the model.
[0159] In addition, in order to evaluate the performance of the spatiotemporal graph deep learning algorithm (STGDL) and the traditional machine learning algorithm in wind power generation efficiency prediction, Experiment 2 was designed to compare the differences between STGDL and random forest (RF), support vector machine (SVM), and traditional LSTM in terms of prediction accuracy, model complexity, and running time.
[0160] Among them, the specific implementation process of using Random Forest (RF), Support Vector Machine (SVM) and traditional Long Short-Term Memory (LSTM) is as follows:
[0161] Data preprocessing: U1. Data cleaning: remove missing values, outliers and duplicate data. U2. Feature scaling: standardize or normalize the data to ensure that different features are at the same level. U3. Time series processing: If the data is a time series, perform time series preprocessing, such as sliding window method or sequence segmentation.
[0162] Feature selection: M1. Correlation analysis: By calculating the correlation between features, exclude highly correlated features and reduce multicollinearity problems. M2. Feature importance: Use methods such as recursive feature elimination (RFE) or model-based feature selection to determine the most relevant features.
[0163] Model training: Random Forest (RF), first set parameters such as the number of trees, maximum number of features, maximum depth, etc. Then divide the data into training set and validation set. Then use the training set data to train the RF model, and each decision tree is trained on a randomly selected feature subset. Support Vector Machine (SVM), first select a suitable kernel function, such as linear, polynomial, Gaussian radial basis (RBF), etc. Then adjust the regularization parameter C and kernel function parameters (such as γ). Then use the training data to fit the SVM model. Long Short-Term Memory Network (LSTM), first define the LSTM network structure, including the input layer, LSTM layer, output layer, etc. Then set the network parameters, such as the number of hidden units, batch size, learning rate, etc. Then compile the model, define the loss function and optimizer. Finally, use the training data set to train the model, and perform multiple iterations until the model converges.
[0164] Model validation: T1. Cross validation: Use k-fold cross validation to evaluate the generalization ability of the model. T2. Hyperparameter tuning: Use grid search or random search to find the optimal hyperparameter combination. T3. Performance evaluation: Based on the validation set data, calculate the performance indicators of the model, such as mean square error (MSE), mean absolute error (MAE), R 2 Scores, etc.
[0165] Finally, the unseen test set data is input into the model for prediction, and the predicted results are compared with the true values to evaluate the predictive ability of the model.
[0166] The data set uses a wind power generation data set containing multi-dimensional historical data such as wind speed, wind direction, temperature, humidity, air pressure, and power generation. The model training and verification adopts a cross-validation strategy to ensure the generalization ability of the model. The evaluation indicators use root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), R 2 The scores are combined with the model’s complexity (number of parameters) and running time.
[0167] The experimental method is: data preprocessing; feature engineering; model training and parameter adjustment; setting hyperparameters such as the number of GCN layers, the number of LSTM layers, and the learning rate. Traditional algorithms, including RF, SVM, and LSTM, adjust the number of trees, kernel functions, hidden layer sizes, etc. Finally, a performance evaluation is conducted to predict the wind speed, wind direction, and power generation in the next 24 hours. The evaluation indicators of each algorithm in the prediction task are recorded. The data recording table is shown in Table 2:
[0168] Table 2 Algorithm comparison data record table
[0169]
[0170] According to data table 2, STGDL showed significant advantages in all prediction indicators, especially in R 2 The scores and MAPE show that STGDL can more accurately capture and predict the complex spatiotemporal dynamics in wind power data. Although the number of parameters and running time of STGDL are slightly higher than other algorithms, the improvement in its prediction accuracy makes this resource consumption reasonable. It can be seen that the spatiotemporal graph deep learning algorithm can provide more accurate and stable prediction results when dealing with wind power performance prediction problems compared with traditional algorithms, and is suitable for use in wind power performance optimization systems to improve the efficiency and reliability of wind power generation.
[0171] In specific implementation, by predicting wind power generation trends, wind power generation strategies can be adjusted in advance to improve power generation and efficiency. By predicting equipment failures, preventive maintenance can be achieved to reduce maintenance costs and downtime. Based on the prediction results, the resource allocation of wind farms, such as energy storage systems and grid dispatching, can be optimized to improve the overall economy and stability of the system. At the same time, data-driven decision support is provided for the operation and management of wind farms, promoting the intelligent and sustainable development of the wind power industry.
[0172] In the above embodiment, the workflow steps of the fuzzy logic controller in the wind power generation efficiency optimization system are as follows:
[0173] Step 601, fuzzification of input variables;
[0174] The input signal is converted into a fuzzy set through a membership function; the input signal includes at least the actual wind speed, wind direction and predicted power generation; the fuzzy set includes three sets of low wind speed, medium wind speed and high wind speed; the membership function adopts the shape of triangle, trapezoid and Gaussian function;
[0175] Step 602: Establishment of fuzzy rules;
[0176] Define the decision logic of the fuzzy logic controller through expert systems and historical data;
[0177] Step 603: fuzzy reasoning;
[0178] According to the current input variables and fuzzy rule base, fuzzy reasoning is performed through the fuzzy reasoning method Sugeno to calculate the output fuzzy set; the fuzzy reasoning method obtains the fuzzy set of the output variable through the activation, aggregation and defuzzification of fuzzy rules;
[0179] Step 604, defuzzification of the output;
[0180] The fuzzy set obtained by fuzzy reasoning is converted into a numerical value as an actual control signal through the center of gravity method, the maximum membership method or the weighted average method;
[0181] Step 605: Dynamic adjustment and optimization;
[0182] The control signal obtained by defuzzification is used as input to adjust the pitch angle and generator speed through the wind turbine control system to optimize the power generation efficiency in real time. In addition, the power output is dynamically adjusted through the mixed integer linear programming method to achieve the balance of the power system.
[0183] Step 606: real-time feedback and correction;
[0184] The fuzzy rules and membership functions are learned and adjusted online based on real-time monitoring system performance and feedback through minimum mean square error and recursive least squares method.
[0185] In a specific embodiment, the real-time feedback and correction mechanism uses real-time data streams to learn and adjust the fuzzy rules and membership functions of the fuzzy logic controller online in a wind power generation efficiency optimization system based on big data to adapt to environmental changes and fluctuations in equipment status, wherein adaptive fuzzy logic control allows the fuzzy logic controller to adjust its rules and parameters according to real-time data to optimize control performance. This usually involves adaptive learning algorithms such as minimum mean square error (LMS), recursive least squares (RLS) or more complex neural network learning algorithms such as back propagation (BP) algorithms. By monitoring the difference between the system output and the expected output in real time, the adaptive algorithm can automatically adjust the parameters of the fuzzy rules (such as the center point and width of the membership function) to reduce this difference, thereby improving the accuracy and robustness of the controller. Recursive algorithms such as recursive least squares (RLS) can estimate system parameters online and update the estimated values through recursive formulas to reflect the latest data information. In real-time data streams, recursive algorithms can quickly respond to new data points and continuously adjust parameter estimates to ensure that fuzzy rules and membership functions always match the current operating status. Reinforcement learning methods learn the optimal strategy through a trial-and-error process to maximize a certain cumulative reward. In wind power control scenarios, reinforcement learning can be used to adjust the parameters of fuzzy controllers to maximize power generation or minimize equipment wear while meeting grid demand. Through interaction with the environment, reinforcement learning algorithms can continuously optimize their decisions and maintain good control effects even in uncertain or changing environments. Combined with fuzzy logic control, sliding mode control can provide a fast response mechanism that maintains the stability and accuracy of the controller even when system parameters change or external disturbances exist. Sliding mode control is a nonlinear control method that achieves fast and robust control by designing a sliding surface and the corresponding switching control law so that the system state can reach and remain on the sliding surface within a limited time. By analyzing a large amount of operating data, data-driven adaptive algorithms can identify the nonlinear characteristics and dynamic change laws of the system, thereby more accurately adjusting the controller parameters and improving the system's adaptability and control effect; using big data analysis and machine learning techniques, the dynamic characteristics of the system are learned from historical data, and then this knowledge is applied to online adjustment of fuzzy rules and membership functions.
[0186] In the above embodiment, the mixed integer linear programming method first constructs an efficiency optimization model including integer variables and continuous variables through an objective function; then, the mixed integer linear programming method explores the solution space through branch and bound method, cutting plane method and branch and price method, and uses linear relaxation to approximate the optimal solution, gradually narrowing the search range of integer solutions until the global optimal solution is found or the preset termination criteria are met; wherein, the branch and bound method improves the solution efficiency through solution space segmentation and boundary condition pruning; the cutting plane method approximates the integer solution through constraints to avoid redundant search paths; the constraints include but are not limited to power grid power balance, wind turbine operating limits, equipment maintenance cycle and electricity market price fluctuations; according to the global optimal solution obtained through exploration, the mixed integer linear programming method obtains real-time wind speed, wind direction, power generation forecast and power grid demand data from the data analysis and prediction module through a data stream processing framework, and dynamically updates the efficiency optimization model parameters.
[0187] In a specific embodiment, mixed integer linear programming (MILP) plays a key role in the wind power generation efficiency optimization system. Through its unique mathematical modeling and solution technology, it realizes the dynamic adjustment of wind power generation operating parameters to ensure that the power output can accurately match the grid demand.
[0188] Specifically, the core of the MILP method is to build a linear programming model containing integer variables and continuous variables, where integer variables correspond to discrete decisions, such as the start and stop status of wind turbines, and continuous variables represent parameters that can be continuously adjusted, such as pitch angle and generator speed. The construction of the model first relies on establishing an objective function, which usually aims to maximize power generation or minimize operating costs while ensuring system stability and safety. The setting of constraints covers physical limitations, economic factors, and grid requirements, such as the upper limit of wind turbine power output, the instantaneous power balance of the grid, and long-term energy scheduling plans, to ensure that the optimization scheme is feasible in the real world.
[0189] In order to solve the constructed MILP model, a series of advanced optimization algorithms are applied, including Branch and Bound, Cutting Plane Method and Branch and Price, as well as more modern heuristic algorithms and meta-heuristic algorithms. These algorithms gradually explore the solution space and use linear relaxation to approach the optimal solution, gradually narrowing the search range of integer solutions until the global optimal solution is found or the preset termination criteria are met. Among them, the branch and bound method greatly improves the solution efficiency by systematically dividing the solution space and combining boundary conditions for pruning; the cutting plane method gradually approaches the integer solution by adding additional constraints, avoiding redundant search paths.
[0190] The flexibility of the MILP method is reflected in its ability to seamlessly connect with real-time data streams to achieve dynamic adjustment of model parameters. This function is achieved through data stream processing frameworks (such as Apache Storm or Apache Kafka), which are responsible for obtaining the latest wind speed forecast, power generation forecast and grid demand data from the data analysis and prediction module, and timely updating the relevant parameters in the MILP model to ensure that the optimization results can reflect the current operating status and market conditions in real time.
[0191] In order to ensure the effectiveness of the MILP optimization results, a feedback mechanism is also embedded in the system. Through a closed-loop control strategy, the actual performance of the wind power system is continuously monitored, operating data is collected, and compared with the expected results. Based on deviation analysis, the system can automatically adjust the prediction parameters in the optimization model and even re-run the MILP solution process when necessary to adapt to environmental changes and fluctuations in equipment status, thereby ensuring the continuous optimization of wind power generation efficiency and accurate matching of grid demand.
[0192] In the above embodiment, the energy storage and management module includes an energy storage system integrated control submodule, a battery management submodule and an energy storage strategy optimization submodule; the energy storage system integrated control submodule includes an energy storage system interface unit and a wind farm integration unit; the energy storage system interface unit exchanges data with the lithium-ion battery energy storage system through a standardized communication protocol; the wind farm integration unit connects the energy storage system to the AC power grid of the wind farm through a power electronic converter; the battery management submodule includes a battery status monitoring unit, a battery balancing control unit and a fault diagnosis and protection unit; the battery status monitoring unit continuously monitors at least the voltage, current and temperature parameters of the battery through a sensor array; the battery balancing control unit adopts a passive balancing strategy, and adjusts the voltage difference between the cells in the battery pack through a balancing circuit to extend the battery life; the fault diagnosis and protection unit identifies battery faults through Kalman filtering, and prevents battery overcharging, over-discharging and overheating through a protection circuit; the energy storage strategy optimization submodule generates a charging and discharging strategy of the energy storage system through a genetic algorithm based on the wind farm power generation and grid demand predicted by the data analysis and prediction module, and dynamically adjusts the charging and discharging plan of the energy storage system through a fuzzy logic control strategy.
[0193] In a specific embodiment, the energy storage system integrated control submodule is responsible for monitoring and controlling the charging and discharging process of the energy storage system (such as battery packs, flywheels, compressed air energy storage, etc.) to ensure that the energy storage unit can respond quickly when needed. Through integrated control, the module can coordinate different types of energy storage resources and optimize the performance of the entire energy storage system. The battery management submodule focuses on the health status monitoring and maintenance of the battery pack, including real-time monitoring of voltage, current, temperature, and management of battery balancing and cooling systems. In addition, the module is also responsible for the battery's SOC (State of Charge) and SOH (State of Health) estimation to prevent overcharging, over-discharging and other behaviors that may damage the battery life. Energy storage strategy optimization submodule: Based on big data analysis, it can formulate the optimal energy storage strategy. It takes into account factors such as the volatility of wind power generation, power demand forecasts, market price fluctuations, physical limitations of energy storage equipment, and maintenance cycles, and predicts the best charging and discharging time through machine learning algorithms to maximize economic benefits and maintain grid stability.
[0194] In the wind power generation efficiency optimization system based on big data, the system will continuously monitor wind power generation, power demand, energy storage status and market prices, and automatically adjust the charging and discharging strategy. Based on big data analysis, the system predicts wind power generation and power demand in the future, and evaluates the health of energy storage equipment. Then, the optimized energy storage strategy is executed, and the execution results are collected for subsequent strategy adjustment and optimization. Finally, the system performance is regularly evaluated, including energy storage efficiency, economic benefits and equipment life, and the system parameters and strategies are adjusted according to the evaluation results.
[0195] Through the energy storage system, the wind power generation system can better cope with wind fluctuations, reduce wind abandonment, and increase the stability of power supply. In addition, the optimization of energy storage strategy can effectively utilize the peak-valley electricity price difference, and increase revenue by charging during low-price periods and discharging during high-price periods. Secondly, the battery management submodule avoids overcharging and discharging and extends battery life through fine-grained state of charge and health status monitoring. At the same time, by optimizing the energy storage strategy, the system can more effectively integrate renewable energy, reduce dependence on traditional fossil fuels, and promote the green transformation of the energy structure. In addition, the energy storage system can respond quickly during peak electricity demand, provide additional power, help balance the grid load, and enhance the overall stability of the grid.
[0196] In the above embodiment, the working steps of the genetic algorithm are:
[0197] Step 901: Convert decision variables into genes on chromosomes through binary coding; the decision variables include but are not limited to charging and discharging power and time parameters;
[0198] Step 902: randomly generate an initial working strategy through a random number generator to form a population;
[0199] Step 903: Evaluate the effectiveness of each strategy based on the goal of minimizing cost or maximizing benefit through a fitness function;
[0200] Step 904: select individuals according to fitness values by proportional selection, tournament selection and roulette selection methods to enter the mating pool and prepare for genetic operations;
[0201] Step 905: select some or all features from the parent strategy through the crossover operator and combine them to generate a new offspring strategy;
[0202] Step 906: Increase the diversity of strategies through mutation operators to prevent the algorithm from converging to a local optimal solution too early;
[0203] Step 907: Replace part of the old population with the newly generated offspring strategy by overlapping generations;
[0204] Step 908, checking the algorithm stopping condition through the maximum number of iterations, fitness threshold and population diversity level;
[0205] Step 909, repeating steps 903 to 907 until the convergence condition is met;
[0206] Step 910: When the algorithm ends, the individual with the highest fitness value is identified and output from the current population.
[0207] In a specific embodiment, a genetic algorithm is a global optimization search algorithm based on the principle of genetic evolution in nature. It imitates genetic operations such as selection, crossover (hybridization) and mutation in the process of biological evolution, and is used to solve optimization and search problems. In the wind power generation efficiency optimization system, the genetic algorithm can be used to optimize the charging and discharging strategy of the energy storage system to match the grid demand and maximize economic benefits. Specifically, the genetic algorithm uses a coding mechanism to represent the solution of the problem as a "chromosome" or "genotype", usually in the form of binary coding, real number coding or symbol coding. The coding converts the solution in the problem domain into a form that can be operated by the genetic algorithm, which is convenient for genetic operations. Next, a set of initial solutions (individuals) are randomly generated to form an initial population, which represents a random sample in the problem solution space. Create a diverse initial population as the starting point of the genetic algorithm search process. Then, define the fitness function, and calculate the quality or fitness of each individual (solution) according to the specific requirements of the problem. This evaluates the pros and cons of individuals in the problem solution space, and individuals with high fitness have a greater chance of being inherited to the next generation. Then, based on the fitness ratio or other selection strategies (such as roulette selection, tournament selection), "parent" individuals are selected from the current population. This ensures that individuals with higher fitness have a greater chance to participate in genetic operations, thereby increasing the probability of excellent characteristics in the offspring. Then, the hybridization process in biological genetics is simulated, and some gene fragments of the two parent individuals are exchanged to generate new individuals (offspring). By recombining the characteristics of the parent individuals, new areas of the solution space are explored to promote genetic diversity. Then, certain bits in the individual genes are randomly changed with a certain probability to introduce random perturbations. This is to increase the genetic diversity of the population, avoid the trap of local optimal solutions, and enhance the global search ability of the algorithm. Then decide how to add the newly generated offspring individuals to the population, possibly replacing some old individuals. Update the population, keep the population size unchanged, and introduce better or more diverse individuals. Finally, set the stopping criteria, such as reaching the predetermined number of iterations, the fitness value is no longer significantly improved, etc. Determine whether the algorithm should continue to iterate or has found a satisfactory solution.
[0208] In the above embodiment, the wind power generation efficiency optimization system based on big data also includes a decision support and optimization strategy module, which is used to provide operation strategy recommendations, optimize asset allocation, maintenance plans and market trading strategies; the decision support and optimization strategy module is based on data analysis results, comprehensively considers power generation, equipment health and energy storage status through a particle swarm optimization algorithm, optimizes asset allocation and maintenance plans, and combines predicted power generation and market prices to formulate market trading strategies using a reinforcement learning algorithm.
[0209] In the specific embodiment, the decision support and optimization strategy module plays a core role. It provides scientific suggestions for the operation strategy of the wind farm through advanced analysis and optimization algorithms, and optimizes asset allocation, maintenance plan and market transaction strategy. This module combines the particle swarm optimization algorithm (PSO) and the reinforcement learning algorithm (RL), and uses the powerful functions of these two algorithms to handle complex decision problems.
[0210] Particle Swarm Optimization (PSO) is a heuristic global optimization algorithm inspired by the behavior of bird and fish flocks. In the decision support and optimization strategy module, PSO is used to find the optimal balance between power generation, equipment health, and energy storage status. Each "particle" represents a potential solution, and its position and velocity are updated iteratively to approach the global optimal solution. PSO is capable of handling multi-objective optimization problems and is very suitable for optimizing asset configuration and maintenance plans in wind farms. Reinforcement Learning (RL) is a trial-and-error based learning method that learns how to take actions to maximize cumulative rewards by interacting with the environment. In the market trading strategy of wind farms, RL can learn the best electricity sales strategy under different market conditions, including when to sell, at what price, and how to respond to market changes. Through interaction with the market, the RL algorithm can gradually optimize the strategy to maximize profits.
[0211] The specific implementation process is:
[0212] Data collection and preprocessing: Extract historical data on wind power generation, equipment health status, energy storage status, market transaction data, etc. from the data lake, perform data cleaning and preprocessing to ensure data quality.
[0213] Feature Engineering and Model Building:
[0214] -Based on the preprocessed data, construct a feature set, including wind speed, wind direction, power generation, equipment operating status, etc.
[0215] -Develop the PSO model and set the initialization parameters, including the number of particles, search space boundaries, inertia weight, etc.
[0216] -Design RL environments, define state space, action space, reward function, and policy network.
[0217] Algorithm training and optimization:
[0218] -Use the PSO algorithm to find the optimal solution for asset configuration and maintenance plan. Through multiple iterations, update the position and speed of particles to approach the global optimum.
[0219] - Train the RL model to learn market trading strategies through interaction with the simulated market environment, and gradually adjust the policy network parameters to maximize the cumulative reward.
[0220] Strategy Execution and Evaluation:
[0221] -Execute asset allocation and maintenance plans and implement market trading strategies based on the optimal solutions obtained by PSO and RL algorithms.
[0222] -Monitor execution results, collect feedback data, evaluate the actual effect of the strategy, and adjust algorithm parameters when necessary.
[0223] Continuous optimization and iteration: Based on the feedback of execution results, PSO and RL models are continuously optimized to achieve continuous improvement of strategies.
[0224] In the specific implementation, by optimizing asset allocation and maintenance plans, equipment failures and unplanned downtime can be reduced, and the power generation efficiency and availability of wind farms can be improved. In addition, optimized maintenance plans can reduce maintenance costs while reducing production losses caused by equipment failures. Secondly, the optimization of market trading strategies can seize market opportunities and increase sales revenue by selling electricity at favorable times. At the same time, decision support and optimization strategy modules based on big data and AI enable wind farms to quickly adapt to market changes and technological advances and maintain their competitiveness. In addition, through efficient operations and market strategies, wind farms can better integrate into the electricity market, promote the popularization of renewable energy and the construction of a sustainable energy system.
[0225] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that these specific embodiments are only illustrative, and those skilled in the art may omit, replace, and change the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, merging the above method steps so as to perform substantially the same functions in substantially the same manner to achieve substantially the same results is within the scope of the present invention. Therefore, the scope of the present invention is limited only by the appended claims.
Claims
1. A wind power generation efficiency optimization system based on big data, characterized by: The system comprises: A data acquisition and management module is used to collect and manage real-time and historical data of the wind farm; the data acquisition and management module collects environmental data and equipment status data in real time through a sensor network, and transmits the collected environmental data and equipment status data to a central data management platform through a wireless communication network; the central data management platform sends the data to a data lake for storage; wherein the environmental data at least includes wind speed, wind direction, temperature and humidity; the equipment status data at least includes generator speed, power output, blade angle, bearing temperature, vibration data, as well as voltage, current and frequency; A data analysis and prediction module, which is used to train a prediction model based on the historical data stored in the data lake, and analyze and predict the real-time data collected by the data acquisition and management module; the prediction model predicts wind power generation trends and equipment failures through a deep learning algorithm of a spatiotemporal graph; the wind power generation trends include future wind speed, wind direction and power generation; A dynamic control and optimization module, used to dynamically adjust the wind power generation operating parameters according to the prediction results of the data analysis and prediction module; based on the prediction model output, the dynamic control and optimization module dynamically adjusts the wind power generation operating parameters through a fuzzy logic controller, and dynamically adjusts the power output to match the grid demand through a mixed integer linear programming method; the wind power generation operating parameters include but are not limited to pitch angle and generator speed; The energy storage and management module is used to smooth the power output through the energy storage system according to the prediction results of the data analysis and prediction module; the energy storage and management module integrates the lithium-ion battery energy storage system with the wind farm through the energy storage system integration method, and monitors the battery status through the battery management system BMS; based on the power generation and grid demand predicted by the data analysis and prediction module, the energy storage and management module also optimizes the charging and discharging strategy of the energy storage system through a genetic algorithm.
2. The wind power generation efficiency optimization system based on big data according to claim 1, characterized in that: The central data management platform includes a data preprocessing module, a data management module and a security compliance module; the data preprocessing module includes a data receiving unit, a data cleaning unit and a data formatting unit; the data receiving unit receives real-time data streams from the wind farm sensor network through the message publishing / subscribing MQTT protocol; the data cleaning unit identifies and removes outliers through the interquartile range IQR method; the data formatting unit ensures the consistency and standardization of the data structure through data mapping; The data management module stores and queries data with timestamps through a time series database, and ensures data asset transparency and traceability through a metadata registry and metadata management APIs; The security compliance module automatically evaluates the legality of the data processing process through log auditing and compliance checking algorithms; the compliance checking algorithm interprets contracts, policy documents and regulatory requirements through natural language processing to ensure that business processes and documents comply with regulations, and performs logical judgments based on input data through the rule engine; the rule engine identifies potential violations or operations through logical reasoning and condition matching.
3. The wind power generation efficiency optimization system based on big data according to claim 2, characterized in that: The security compliance module also ensures the security of data during transmission and storage through symmetric encryption and asymmetric encryption methods, and ensures the security of data access through attribute access control.
4. The wind power generation efficiency optimization system based on big data according to claim 1, characterized in that: The workflow of the data analysis and prediction module is as follows: Step 401: data integration and preprocessing; Acquire historical data from the data lake and receive real-time data streams at the same time, and perform data cleaning, format unification, and missing value processing through a distributed data processing system; the distributed data processing system improves data quality through data cleaning algorithms and data conversion methods; Step 402: feature engineering; Features were extracted from the preprocessed data through principal component analysis, recursive feature elimination, and feature importance scoring for training the prediction model; Step 403: model training; A prediction model is trained using historical data through a spatiotemporal graph deep learning algorithm to predict wind power generation trends and equipment failures; the spatiotemporal graph deep learning algorithm captures the spatial characteristics of wind field distribution through a convolutional neural network (CNN) and captures time series dependencies through a long short-term memory (LSTM) network, and trains a model to predict future wind speed, wind direction, and power generation; Step 404: real-time data analysis; Push real-time data to the prediction model through the real-time data stream processing framework, and use the model to output the prediction results; Step 405: output and apply the prediction results; Feedback the prediction results to the dynamic control and optimization module through the message queue for adjusting the wind turbine operating parameters; the prediction results at least include future wind speed, wind direction and power generation; the operating parameters at least include blade angle and generator speed; Step 406: Model verification and update; The accuracy and stability of the prediction model are continuously verified through cross-validation and rolling window validation methods. Based on new historical data, the model is periodically updated through online learning methods to adapt to changes in the environment and equipment.
5. The wind power generation efficiency optimization system based on big data according to claim 1, characterized in that: The working method of the spatiotemporal graph deep learning algorithm in the wind power generation efficiency optimization system is as follows: Step 501, data preprocessing and feature engineering; The missing values of wind speed data are filled by time series interpolation method, and the wind speed data is standardized by wind speed standardization to construct the wind power data feature set; Step 502: construct a space-time graph structure; Based on the wind power data feature set, a weighted matrix of spatial distances between nodes is defined by a spatial distance weighted matrix construction formula, and a graph structure is constructed to reflect the spatial relationship between the wind power station and its surrounding environment. The spatial distance weighted matrix construction formula expresses spatial proximity through an exponential decay function; the formula expression of the spatial distance weighted matrix construction formula is: In formula (1), W ig is the spatial distance weight between node i and node g, d ig Represents the geographical distance between node i and node g, which is used to calculate the spatial weight. Represents the attenuation coefficient of the control space influence range, which is used to determine the influence of spatial distance on weight; Step 503: spatial feature learning; The spatial features between nodes are extracted from the constructed graph structure through the graph convolution network update formula to reflect the correlation between wind speed and wind direction at different locations in the wind power station; the formula expression of the graph convolution network update formula is: In formula (2), B i (l+1) is the updated feature representation of node i at the (l+1) layer, which is used to capture spatial dependencies; B i (l) represents the feature representation of node j in the lth layer, which is used for aggregation calculation; C(i) represents the neighbor set of node i, which is used to aggregate neighborhood information; N(l) represents the weight matrix of the lth layer, which is used to learn feature transformation; α represents the activation function, which is used to introduce nonlinearity and enhance the expression ability of the model; C ig Represents the normalization factor, which is used to ensure the smoothness of the aggregation operation; Step 504: time series modeling; The time dynamic characteristics of wind speed changes are captured by the long short-term memory network update function to predict future wind speed and wind direction; the formula expression of the long short-term memory network update function is: In formula (3), C t represents the unit state, which is used to capture long-term dependencies in the sequence; x is the input vector of the current time step, carrying instant information; b c Represents the bias term, which is used to adjust the degree of door opening; f t 、i t and t Respectively represent the gating mechanisms of forgetting, input and output information flows, which are used to determine the retention, update and output status of information; W c represents the weight matrix, which is used to learn the mapping between input and state; * represents element-wise multiplication; h t Represents the hidden state, which is passed to the next time step as the output of the model; Step 505: integrating spatiotemporal features; The final prediction result is generated by combining the spatial features and the time series features through the spatiotemporal feature fusion formula; the spatiotemporal feature fusion formula fuses the H output of the graph convolutional network through weighted summation. (L) and h output by the LSTM network t , generate the comprehensive spatiotemporal feature vector Z t ; The formula expression of the spatiotemporal feature fusion formula is: In formula (4), Z t represents the fused spatiotemporal feature vector, w s and w t are the weights of spatial features and temporal features, respectively, used to balance the contributions of the two; H (L) Represents the output of the last layer of the graph convolutional network, including spatial features; h t Represents the output of the last layer of the long short-term memory network, which contains time features; W fision Represents the weight matrix of the fusion network, which is used to learn feature weights; b fusior Represents the bias term of the fusion network, which is used to adjust the benchmark for weight calculation; Step 506: prediction and post-processing; Based on the fusion of spatiotemporal features Z t , power generation prediction will be performed through the multi-layer perceptron network MLP.
6. The wind power generation efficiency optimization system based on big data according to claim 1, characterized in that: The workflow steps of the fuzzy logic controller in the wind power generation efficiency optimization system are as follows: Step 601, fuzzification of input variables; The input signal is converted into a fuzzy set through a membership function; the input signal includes at least the actual wind speed, wind direction and predicted power generation; the fuzzy set includes three sets of low wind speed, medium wind speed and high wind speed; the membership function adopts the shape of triangle, trapezoid and Gaussian function; Step 602: Establishment of fuzzy rules; Define the decision logic of the fuzzy logic controller through expert systems and historical data; Step 603: fuzzy reasoning; According to the current input variables and fuzzy rule base, fuzzy reasoning is performed through the fuzzy reasoning method Sugeno to calculate the output fuzzy set; the fuzzy reasoning method obtains the fuzzy set of the output variable through the activation, aggregation and defuzzification of fuzzy rules; Step 604, defuzzification of the output; The fuzzy set obtained by fuzzy reasoning is converted into a numerical value as an actual control signal through the center of gravity method, the maximum membership method or the weighted average method; Step 605: Dynamic adjustment and optimization; The control signal obtained by defuzzification is used as input to adjust the pitch angle and generator speed through the wind turbine control system to optimize the power generation efficiency in real time. In addition, the power output is dynamically adjusted through the mixed integer linear programming method to achieve the balance of the power system; Step 606: real-time feedback and correction; The fuzzy rules and membership functions are learned and adjusted online based on real-time monitoring system performance and feedback through minimum mean square error and recursive least squares method.
7. The wind power generation efficiency optimization system based on big data according to claim 1, characterized in that: The mixed integer linear programming method first constructs an efficiency optimization model containing integer variables and continuous variables through an objective function; then, the mixed integer linear programming method explores the solution space through branch and bound method, cutting plane method and branch and price method, and uses linear relaxation to approximate the optimal solution, gradually narrowing the search range of integer solutions until the global optimal solution is found or the preset termination criteria are met; wherein, the branch and bound method improves the solution efficiency through solution space segmentation and boundary condition pruning; the cutting plane method approximates the integer solution through constraints to avoid redundant search paths; the constraints include but are not limited to power grid power balance, wind turbine operating limits, equipment maintenance cycle and electricity market price fluctuations; according to the global optimal solution obtained through exploration, the mixed integer linear programming method obtains real-time wind speed, wind direction, power generation forecast and power grid demand data from the data analysis and prediction module through a data stream processing framework, and dynamically updates the efficiency optimization model parameters.
8. The wind power generation efficiency optimization system based on big data according to claim 1, characterized in that: The energy storage and management module includes an energy storage system integrated control submodule, a battery management submodule and an energy storage strategy optimization submodule; the energy storage system integrated control submodule includes an energy storage system interface unit and a wind farm integration unit; the energy storage system interface unit exchanges data with the lithium-ion battery energy storage system through a standardized communication protocol; the wind farm integration unit connects the energy storage system to the AC power grid of the wind farm through a power electronic converter; the battery management submodule includes a battery status monitoring unit, a battery balancing control unit and a fault diagnosis and protection unit; the battery status monitoring unit continuously monitors at least the voltage, current and temperature parameters of the battery through a sensor array; the battery balancing control unit adopts a passive balancing strategy, adjusts the voltage difference between the cells in the battery pack through a balancing circuit, and prolongs the battery life; the fault diagnosis and protection unit identifies battery faults through Kalman filtering, and prevents battery overcharging, over-discharging and overheating through a protection circuit; the energy storage strategy optimization submodule generates a charging and discharging strategy of the energy storage system through a genetic algorithm based on the wind farm power generation and grid demand predicted by the data analysis and prediction module, and dynamically adjusts the charging and discharging plan of the energy storage system through a fuzzy logic control strategy.
9. The wind power generation efficiency optimization system based on big data according to claim 8, characterized in that: The working steps of the genetic algorithm are: Step 901: Convert decision variables into genes on chromosomes through binary coding; the decision variables include but are not limited to charging and discharging power and time parameters; Step 902: randomly generate an initial working strategy through a random number generator to form a population; Step 903: Evaluate the effectiveness of each strategy based on the goal of minimizing cost or maximizing benefit through a fitness function; Step 904: select individuals according to fitness values by proportional selection, tournament selection and roulette selection methods to enter the mating pool and prepare for genetic operations; Step 905: select some or all features from the parent strategy through the crossover operator and combine them to generate a new offspring strategy; Step 906: Increase the diversity of strategies through mutation operators to prevent the algorithm from converging to a local optimal solution too early; Step 907: Replace part of the old population with the newly generated offspring strategy by overlapping generations; Step 908, checking the algorithm stopping condition through the maximum number of iterations, fitness threshold and population diversity level; Step 909, repeating steps 903 to 907 until the convergence condition is met; Step 910: When the algorithm ends, the individual with the highest fitness value is identified and output from the current population.
10. The wind power generation efficiency optimization system based on big data according to claim 1, characterized in that: The wind power generation efficiency optimization system based on big data also includes a decision support and optimization strategy module, which is used to provide operation strategy suggestions, optimize asset allocation, maintenance plans and market trading strategies; based on the data analysis results, the decision support and optimization strategy module comprehensively considers power generation, equipment health and energy storage status through a particle swarm optimization algorithm, optimizes asset allocation and maintenance plans, and combines the predicted power generation and market prices to formulate market trading strategies using a reinforcement learning algorithm.
Citation Information
Patent Citations
Wind power generation grid connection stored energy adjusting device
CN102856921A
Wind turbine generator fault early warning method based on graph neural network
CN114372504A
Power grid demand side load balance analysis method and system
CN115587833A
Charging control method and device for new energy charging station
CN117410988A
Intelligent wind power prediction and scheduling method and system
CN117424293A
Cited By
Series battery pack voltage detection device and detection method for dynamic equalization
CN119230988A
Multi-type capacity application method and system of virtual power plant
CN119358974A
Unattended hydrogen ion battery charging station
CN119428295A
Breeze power generation system and method and related components thereof
CN119616767A
Space-time joint rotating speed optimization method for hybrid propulsion system of hovercar
CN119620605A
