Dynamic balance system for wind, light, fire and nuclear storage integrated regulation and control of power grid in cold region

Abstract

The invention provides a dynamic balance system for wind, light, fire and nuclear storage integrated regulation and control of a power grid in a cold region, which relates to the technical field of power grid regulation and control and comprises an energy data acquisition and preprocessing module, a multi-energy coordinated regulation and control module, an energy storage system optimization management module, an energy scheduling optimization module, an intelligent decision and monitoring module and a system communication and safety module. A multi-level, multi-module and intelligent comprehensive framework is constructed, collaborative management and optimal scheduling of various energy sources are achieved, stable operation, efficient energy utilization and sustainable development of a cold region power grid are ensured, global optimization is achieved through deep reinforcement learning, local precise control is conducted through fuzzy logic, and the energy utilization efficiency is improved. The system can intelligently distribute the generated power of each energy source of wind, light, fire and nuclear storage according to the real-time state and environmental conditions of the cold region power grid, gives full play to the advantages of different energy sources, realizes the complementary utilization of the energy sources, realizes the collaborative optimization of multiple energy sources, and improves the utilization efficiency and stability of the energy sources.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid regulation and control, and in particular to a dynamic balance system for integrated regulation and control of wind, solar, thermal and nuclear power storage systems in cold-region power grids. Background Art

[0002] According to a method for detecting the dynamic balance state of multiple loads in a distribution network based on distributed power sources disclosed in China with a publication number of "CN117388626B", the method includes: using multiple active loads and multiple reactive loads to obtain the multiple loads in the dynamic balance state of the load and the multiple loads in the state to be detected under a specified time series; calculating the active resonance density at the peak and valley points of the load in the two-dimensional transformation image to determine whether the multiple loads are in an unbalanced state, recorded as the first state; if unbalanced, output the result; if abnormal, calculate the lateral fitness and longitudinal fitness of the multiple reactive loads, and calculate the peak and valley fitness, to determine whether the multiple loads are in a balanced state, recorded as the second state; if so, output the result; if abnormal, calculate the lateral offset relationship and longitudinal offset relationship of the peak and valley points in the two-dimensional transformation image to calculate the load distortion rate, to determine whether the multiple loads are in a balanced state, and output the judgment result.

[0003] According to the Chinese publication number "CN117937566A", a microgrid system suitable for charging and discharging electric vehicles in high-cold areas includes: high-cold resistant charging equipment, including high-cold resistant box transformers and high-cold resistant charging terminals, the high-cold resistant charging terminals and the high-cold resistant box transformers are connected by wires; photovoltaic power generation system, the photovoltaic power generation system is connected to the high-cold resistant charging equipment; energy storage system, including distributed mobile energy storage and cascade battery energy storage system, the energy storage system is connected to the photovoltaic power generation system, the power grid and the high-cold resistant charging equipment; microgrid management and control system, including a bidirectional converter, a DC bus and an AC bus; the DC bus is used to realize power interaction between the photovoltaic power generation system, the energy storage system, and the high-cold resistant charging equipment, and the energy interaction between the AC bus and the DC bus is realized through the bidirectional converter. The present invention solves the problem of difficulty in charging electric vehicles in high-cold areas in winter due to low temperatures, as well as the problem of grid load fluctuation caused by large-scale electric vehicles entering the grid.

[0004] The above patent documents and prior art have the following technical problems when used:

[0005] Problem 1: In the above documents and existing technologies, in the existing cold-region power grid, the power generation of different energy sources is often relatively independent and lacks an effective collaborative management mechanism. For example, wind and solar energy are greatly affected by natural conditions, and power generation is intermittent and volatile. The regulation of thermal power and nuclear power is relatively lagging, and the application of energy storage systems is not smart and efficient enough, which makes it difficult to achieve a smooth transition in energy supply. The energy storage system of the traditional cold-region power grid is not accurate in predicting load demand and wind and solar power generation, and the charging and discharging strategy is not optimized enough. The unreasonable charging and discharging operation affects its service life and performance.

[0006] Problem 2: In the above documents and existing technologies, the decision-making systems of existing cold-region power grids mostly rely on simple data analysis and experience judgment, lacking a comprehensive and in-depth understanding and analysis of the power grid. When faced with complex power grid operation conditions and market changes, it is difficult to quickly and accurately make the best decision. The energy dispatching of traditional power grids mainly relies on manual experience or simple algorithms, which makes it difficult to accurately predict energy generation potential and load demand trends, resulting in unreasonable energy dispatching, low energy utilization efficiency, and poor power grid operation benefits.

[0007] Question three, in the above-mentioned documents and existing technologies, the traditional communication methods and security measures may not be perfect enough to effectively guarantee the secure transmission of data and the stable operation of the system. At the same time, there is a lack of real-time monitoring and early warning mechanisms for network security. Once a security incident occurs, it is difficult to respond and handle it quickly. The existing security protection measures for cold-region power grids may be relatively simple, mainly focusing on one aspect of security protection, such as data encryption or simple firewall settings. There is a lack of a comprehensive security protection system. There is a lack of effective response measures for complex network attacks and security threats, which can easily lead to attacks on the power grid system and failures, affecting the reliability of energy supply. Summary of the invention

[0008] In view of the shortcomings of the prior art, the present invention provides a dynamic balance system for integrated regulation of wind, solar, thermal and nuclear storage in cold-region power grids, which solves the following problems:

[0009] 1. Aiming at the problems of insufficient coordination and complementarity of multiple energy sources in cold-region power grids and low efficiency of energy storage systems;

[0010] 2. Address the issues of incomplete decision support and inaccurate energy dispatching in cold-region power grids;

[0011] 3. Address the issues of weak communication and data security and imperfect security protection system in power grid security protection.

[0012] To achieve the above objectives, the present invention is implemented through the following technical solutions: a dynamic balance system for integrated regulation of wind, solar, thermal and nuclear storage in cold-region power grids, the system comprising an energy data acquisition and preprocessing module, a multi-energy coordinated regulation module, an energy storage system optimization management module, an energy scheduling optimization module, an intelligent decision-making and monitoring module, and a system communication and security module, wherein:

[0013] The energy data acquisition and preprocessing module is used to comprehensively collect the operating data and related environmental data of various energy sources in the cold-region power grid, and clean, normalize and preprocess them. Through widely deployed sensors and data acquisition equipment, the real-time operating data of various energy sources such as wind, solar, fire and nuclear storage are obtained, including power generation power, equipment temperature, operating status, and external environmental data such as air temperature, wind speed, light intensity, humidity and operating parameters such as voltage, current and frequency of the power grid. The data cleaning algorithm is used to remove noise, outliers and erroneous data, and the data normalization technology is used to unify the scale of data of different dimensions, and the time series analysis method is used to stabilize the data, etc., to provide an accurate and reliable data basis for the subsequent analysis and decision-making of the system;

[0014] The multi-energy coordinated control module is used for the coordinated management and optimized control of multiple energy sources such as wind, solar, thermal, and nuclear energy. It is based on the global optimization of reinforcement learning, adopts a deep reinforcement learning algorithm, takes the current state of the power grid as input, and outputs macro-control instructions such as the power distribution ratio of each energy subsystem and the overall charging and discharging strategy of the energy storage system through interactive learning with the power grid environment. Based on the local control of fuzzy logic, a fuzzy logic controller is set for each energy subsystem, and fine-tuned control is performed according to their respective input variables. The global optimization of reinforcement learning provides guidance for the local control of fuzzy logic, and the local control uploads the operation feedback information for global optimization to update and optimize the strategy;

[0015] The energy storage system optimization management module optimizes the control and parameter configuration of the energy storage system. The neural network predicts and constructs a deep learning model, takes historical load data as input, and predicts the load demand and wind and solar power generation of the cold-region power grid. In terms of control, the genetic algorithm optimization uses the energy storage system charging and discharging power as the decision variable, and the grid stability and other indicators as the objective function to search for the optimal charging and discharging strategy. In terms of optimization, the parameters and layout positions of the energy storage equipment are optimized. The prediction results of the neural network provide input for the genetic algorithm optimization, and the optimization results of the genetic algorithm are fed back to the neural network to adjust the model parameters.

[0016] The energy scheduling optimization module realizes the reasonable arrangement of various energy sources in power generation distribution, energy storage charging and discharging, energy conversion and coordinated operation by optimizing the energy scheduling plan. The deep learning feature extraction and prediction adopts CNN to analyze the image data of power grid equipment, and uses the LSTM model to model and analyze the energy data to predict the power generation potential and load demand trend of different energy sources. The ant colony algorithm scheduling optimization abstracts the power grid energy nodes and transmission lines into a graph structure. The ant colony searches for the optimal energy scheduling plan in the graph based on the pheromone concentration and heuristic information. Deep learning provides prediction information as heuristic information for the ant colony algorithm. The results of the ant colony algorithm are fed back to the deep learning model for optimization and improvement, thereby improving energy utilization efficiency and power grid operation efficiency.

[0017] The intelligent decision-making and monitoring module includes a decision support unit and a monitoring and visualization unit, which are used to integrate the information of various modules of the system, provide decision support for dispatchers, and monitor and visualize the operation status of the power grid in real time. The decision support unit integrates the information of modules such as multi-energy coordinated regulation, and evaluates and predicts the operation status of the power grid in real time based on big data analysis and artificial intelligence algorithms. It has scenario simulation and risk assessment functions. The monitoring and visualization unit monitors various parameters of the power grid in real time through sensors and other acquisition equipment, and transmits them to the monitoring center through the communication network. A visualization platform is developed to display the operation data in an intuitive form, and provide historical data query and analysis functions;

[0018] The system communication and security module includes a communication network unit and a security protection unit, which are used to connect the data communication between the various modules of the system and protect the system from security threats such as network attacks and data leakage. The communication network builds a network that integrates wired and wireless communication technologies, connects various nodes and devices in the power grid, establishes communication protocols and data interface standards, and conducts seamless communication and data sharing between different systems and devices. The system security adopts encryption technology, identity authentication and access control, conducts real-time security monitoring and evaluation of key power grid equipment and systems, formulates system security emergency plans, conducts regular safety training and drills, and maintains the safe and stable operation of the cold-region power grid.

[0019] Preferably, in the energy data collection and preprocessing module, the data cleaning algorithm adopts a combination of a method based on statistical analysis and an anomaly detection method based on machine learning. Data points that are obviously deviated from the normal range are first identified through statistical analysis, and then the machine learning model is used to further detect potential outliers to ensure the accuracy of the data. The data normalization technology adopts the Z-score standardization method to convert the data into a distribution with a mean of 0 and a standard deviation of 1 to meet the requirements of different algorithms for data scales. In addition, in the process of stabilization, the time series analysis method also combines seasonal differences and trend differences in addition to the ARIMA model to more accurately remove the trend and seasonal components in the data, providing a better data basis for subsequent analysis.

[0020] Preferably, in the multi-energy coordinated control module, the deep reinforcement learning algorithm adopts a deep dual Q network DDQN, which is based on the traditional DQN, by separating the calculation of the target Q value and the selection of the action. During the training process, the experience replay buffer size is set to a specific value to make full use of historical experience data. At the same time, a small batch stochastic gradient descent method is used to update the network parameters, optimize the convergence speed and stability of the algorithm, and design a suitable reward function according to the characteristics of the cold-region power grid and the energy distribution. The energy supply stability is comprehensively considered, including the proportion of time that the voltage and frequency fluctuations are within the allowable range, cost-effectiveness, the difference between the energy production cost and the electricity sales revenue, the carbon emission reduction, and the operating life of the energy equipment, so that the algorithm can learn the optimal multi-energy coordinated control strategy.

[0021] Preferably, in the multi-energy coordinated control module, the fuzzy logic controller of each energy subsystem is designed with targeted input and output variables and a fuzzy rule base, further including the following contents:

[0022] The input variables of the fuzzy logic controller of the wind turbine include wind speed deviation, wind turbine speed deviation and power generation deviation, and the output variables are pitch angle adjustment and generator torque adjustment. The fuzzy rule base is formulated based on expert experience and a large amount of experimental data, covering control rules in different wind speed ranges and operating conditions;

[0023] The input variables of the fuzzy logic controller of the thermal power unit include grid load demand deviation, coal quality parameters, unit temperature deviation, etc. The output variables are fuel supply adjustment, water supply adjustment and steam pressure adjustment. The fuzzy rule base considers the optimal control strategy under different load changes and equipment states.

[0024] The fuzzy logic controller of the nuclear power unit focuses on input variables such as reactor power deviation and coolant flow deviation, and the output variable is the movement position adjustment of the control rod, etc., to ensure that the nuclear power unit can achieve efficient power regulation under the premise of safe operation;

[0025] The input variables of the fuzzy logic controller of the energy storage system are the energy storage power deviation, the charge and discharge power limit deviation and the peak and valley period mark of the power grid. The output variable is the charge and discharge power adjustment amount. The fuzzy rule base formulates reasonable charge and discharge strategies according to different power states and power grid requirements to achieve effective utilization and energy balance of the energy storage system.

[0026] Preferably, in the energy storage system optimization management module, the deep learning model used in the neural network prediction is a combination structure of a multi-layer long short-term memory network (LSTM) and a convolutional neural network (CNN), further comprising the following contents:

[0027] LSTM is used to process time series data, such as historical load data and time characteristics, to capture their long-term dependencies and trend changes. The LSTM layer contains multiple hidden layer units, each of which controls the flow and memory of information through a gating mechanism;

[0028] CNN is used to process meteorological data and image data of wind and solar power generation equipment, extract spatial features and local patterns. The CNN layer uses a specific convolution kernel size and number for feature extraction, and then fuses the features extracted by the two and inputs them into the fully connected layer to predict load demand and wind and solar power generation. The network training uses the Adam algorithm, an optimization algorithm with adaptive learning rate, which automatically adjusts the learning rate according to the gradient information during the training process, speeds up the convergence speed and improves the training effect.

[0029] Preferably, in the energy storage system optimization management module, when the genetic algorithm optimizes the energy storage system, the chromosome encoding adopts real number encoding to directly represent the equipment parameters and operation strategy variables of the energy storage system, including battery capacity, charge and discharge rate, SOC upper and lower limits, and charge and discharge power, etc., and further includes the following contents:

[0030] The selection operation adopts the roulette selection method, which calculates the probability of being selected according to the fitness value of the individual, so that individuals with high fitness have a greater chance of being selected and inherited to the next generation;

[0031] The crossover operation uses the real number crossover method to exchange genes at the corresponding positions of the chromosomes of the two selected parent individuals to generate new offspring individuals. The crossover probability is set to an appropriate value based on experiments and experience.

[0032] The mutation operation uses a uniform mutation method to randomly select several gene positions on the chromosome of an individual and make small random changes to them with a certain mutation probability to increase the diversity of the population;

[0033] The algorithm's fitness function comprehensively considers grid stability indicators, including voltage deviation, frequency deviation, energy cost indicators, including the investment cost, operating cost and electricity purchase cost of energy storage equipment, as well as the life loss indicator of the energy storage system, and searches for the optimal energy storage system configuration and operation strategy through multiple iterations.

[0034] Preferably, in the energy scheduling optimization module, in the deep learning feature extraction and prediction part, the CNN model adopts a specific convolution layer structure and pooling layer structure when analyzing the image data of power grid equipment. The convolution layer uses convolution kernels of different sizes and numbers to extract different features of the equipment image, such as the appearance features and temperature distribution features of the equipment. The pooling layer uses the maximum pooling or average pooling method to downsample the feature map output by the convolution layer to reduce the amount of data and extract the main features. When the LSTM model is modeling and analyzing the energy data, the time step and the number of hidden layer neurons are reasonably set to better capture the time series characteristics and dynamic change laws of the energy data. During the model training process, the batch normalization technology is used to normalize the input data to accelerate the training convergence speed of the model. At the same time, the Dropout technology is used to prevent overfitting and improve the generalization ability of the model. The training data is generated using the sliding window method, and the parameters of the model are continuously updated so that it can adapt to the real-time changing needs of energy scheduling in cold-region power grids.

[0035] Preferably, in the energy scheduling optimization module, in the scheduling optimization part of the ant colony algorithm, the number of ants is reasonably set according to the scale and complexity of the power grid to ensure that the optimal path can be fully explored in the search space, the initial concentration of pheromones is set to a small non-zero value to avoid the algorithm from converging to the local optimal solution too early, the pheromone volatility coefficient is determined according to experience and experiments, the attenuation rate of pheromones on the path is controlled, and the exploration and utilization capabilities of the algorithm are balanced. The weight of the heuristic information is adjusted according to the relative importance of factors such as energy cost, power generation efficiency, and transmission loss to guide the ants to choose a better path during the search process. In each iteration, when the ants find a new path, according to the objective function value of the energy scheduling plan corresponding to the path, such as comprehensive consideration of indicators such as energy utilization efficiency, cost and power grid stability, the pheromone concentration on the path is updated to make subsequent ants more likely to choose a better path. At the same time, the maximum number of iterations of the algorithm is set. When the maximum number of iterations is reached or a certain convergence condition is met, the algorithm stops searching and outputs the current optimal energy scheduling plan.

[0036] Preferably, in addition to evaluating and predicting the operating status of the power grid based on big data analysis and artificial intelligence algorithms, the decision support unit in the intelligent decision-making and monitoring module also introduces knowledge graph technology to construct the equipment information, energy data, operating rules, expert experience and other knowledge of the cold-region power grid into a knowledge graph, which is stored and managed through a graph database. In the decision-making process, the semantic reasoning and association analysis capabilities of the knowledge graph are used to provide dispatchers with more comprehensive and in-depth decision-making suggestions. The decision support unit is connected to an external energy market information system to obtain information such as energy prices and market demand in real time, and adjusts the power grid's operating strategy and energy dispatch plan according to market dynamics to maximize economic benefits.

[0037] Preferably, in the system communication and security module, in addition to encryption technology, identity authentication and access control, system security also introduces an intrusion detection system (IDS) and an intrusion prevention system (IPS). The IDS monitors network traffic and system activities in real time, identifies potential intrusion behaviors and abnormal activities by analyzing traffic patterns and behavioral characteristics, and issues alarms in a timely manner.

[0038] The present invention provides a dynamic balance system for integrated regulation of wind, solar, thermal and nuclear power storage in cold-region power grids. It has the following beneficial effects:

[0039] 1. The system of the present invention realizes global optimization through deep reinforcement learning and local precise control through fuzzy logic. The system can intelligently allocate the power generation of wind, solar, fire, nuclear and storage energy sources according to the real-time status and environmental conditions of the cold-region power grid, give full play to the advantages of different energy sources, realize complementary utilization of energy, and achieve coordinated optimization of multiple energy sources. At the same time, combined with neural network prediction and genetic algorithm optimization, the energy storage system can more accurately predict load demand and wind and solar power generation, so as to formulate more reasonable charging and discharging strategies, which can not only effectively reduce peak loads and fill valleys and balance the supply and demand of the power grid, but also extend the service life of energy storage equipment and reduce energy storage costs. By optimizing the parameters and layout of energy storage equipment, the performance and energy utilization efficiency of the energy storage system can be further improved, the power grid's ability to absorb renewable energy can be enhanced, the impact of energy fluctuations on the power grid can be reduced, and the stability and reliability of the power grid can be improved, thereby improving energy utilization efficiency and stability.

[0040] 2. The present invention uses the system to introduce knowledge graph technology and connect with the energy market information system. The intelligent decision-making and monitoring module can provide dispatchers with more comprehensive and in-depth decision-making support. By using the feature extraction and prediction of deep learning and the dispatch optimization of the ant colony algorithm, the system can accurately predict the power generation potential and load demand trend of different energy sources, and search for the optimal energy dispatch plan. This intelligent dispatching method can fully consider multiple factors such as energy cost, power generation efficiency, and transmission loss, and realize the reasonable arrangement of energy in power generation power distribution, energy storage charging and discharging, energy conversion and coordinated operation, improve energy utilization efficiency and grid operation efficiency. At the same time, with the continuous operation and learning of the system, it can adapt to the real-time changing needs of the cold-region power grid, such as responding to the impact of extreme weather on energy supply, further enhancing the intelligence and adaptability of the power grid.

[0041] 3. The present invention adopts the system to construct a multi-level, multi-module, intelligent integrated architecture, aiming to achieve coordinated management and optimized scheduling of multiple energy sources to ensure the stable operation, efficient energy utilization and sustainable development of the cold-region power grid. The communication and security module adopts a variety of technical means to ensure the safe and stable operation of the power grid, integrates wired and wireless communication technologies to build a reliable communication network, ensures the efficiency and stability of data transmission between modules, protects the system from the threat of data leakage and illegal access through encryption technology, identity authentication and access control, introduces intrusion detection system (IDS) and intrusion prevention system (IPS), monitors and prevents network attacks in real time, promptly discovers and handles potential safety hazards, conducts regular safety training and drills, and performs vulnerability scanning and security assessment on the system to improve the safety awareness of personnel and the overall security of the system, and ensures the safe and reliable operation of the cold-region power grid system from many aspects, reduces the risk of power grid failure caused by safety issues, and provides a solid guarantee for the energy supply in cold regions. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a system architecture diagram of the present invention;

[0043] Figure 2 It is a diagram of the system operation steps of the present invention;

[0044] Figure 3 It is a data orientation diagram of each module of the system of the present invention. DETAILED DESCRIPTION

[0045] 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. Specific embodiment one:

[0047] like Figure 1-Figure 3 As shown in the figure, the dynamic balance system of wind, solar, thermal and nuclear storage integrated control of cold-region power grid includes energy data acquisition and preprocessing module, multi-energy coordinated control module, energy storage system optimization management module, energy dispatch optimization module, intelligent decision-making and monitoring module and system communication and security module. The system builds a multi-level, multi-module, intelligent comprehensive architecture, aiming to achieve coordinated management and optimized dispatch of multiple energy sources to ensure the stable operation, efficient energy utilization and sustainable development of cold-region power grid. The communication and security module adopts a variety of technical means to ensure the safe and stable operation of the power grid, including:

[0048] The energy data acquisition and preprocessing module is used to comprehensively collect the operating data and related environmental data of various energy sources in the cold-region power grid, and to clean, normalize and preprocess them. Through widely deployed sensors and data acquisition equipment, the real-time operating data of various energy sources such as wind, solar, thermal, and nuclear storage are obtained, including power generation power, equipment temperature, and operating status. For example, the wind turbine speed, power generation power, wind direction and speed of wind farms, the light intensity and power generation power of photovoltaic power stations, the unit power, coal consumption, and steam pressure of thermal power plants, the reactor power and coolant flow of nuclear power plants, the charging and discharging power and power status of energy storage systems, as well as external environmental data such as temperature, wind speed, light intensity, and humidity, and operating parameters such as voltage, current, and frequency of the power grid, as well as geographical information, seasonal changes, and time characteristics of cold regions. These data have an important impact on the power generation efficiency of energy and the load demand of the power grid. The data cleaning algorithm is used to remove noise, outliers, and erroneous data, and the data normalization technology is used to unify the scale of data of different dimensions, and the time series analysis method is used to stabilize the data, etc., to provide an accurate and reliable data basis for the subsequent analysis and decision-making of the system.

[0049] The multi-energy coordinated control module is used for the coordinated management and optimized control of multiple energy sources such as wind, solar, thermal, and nuclear energy. A global optimization controller based on reinforcement learning is set at the top level, using a deep dual-Q network DDQN. The controller takes the current state of the power grid, including the power generation power, load demand, energy storage capacity, and environmental parameters of each energy source, as input, and outputs macro-control strategies through continuous interactive learning with the cold-region power grid environment, such as the power generation power allocation ratio of each energy subsystem, the overall charging and discharging strategy of the energy storage system, etc., to achieve the overall goals of energy supply stability, cost-effectiveness maximization, and carbon emission minimization. Based on the global optimization of reinforcement learning, a deep reinforcement learning algorithm is used, and the current state of the power grid is taken as input. Through interactive learning with the power grid environment, macro-control instructions such as the power generation power allocation ratio of each energy subsystem and the overall charging and discharging strategy of the energy storage system are output. For each energy subsystem of wind, solar, thermal and nuclear storage, local controllers based on fuzzy logic are set up in the middle layer. Fuzzy logic controllers are set up for each energy subsystem based on fuzzy logic. Refined regulation is performed according to their respective input variables. The global optimization of reinforcement learning provides guidance for the local control of fuzzy logic. The local control uploads the operation feedback information for the global optimization to update and optimize the strategy. The global optimization controller of reinforcement learning provides macro-control objectives and strategy guidance for the local controller of fuzzy logic. The local controller refines and adjusts the instructions according to the real-time operation of the local energy system, and uploads the operation feedback information to the global optimization controller in real time. The global optimization controller evaluates the effect of the current control strategy based on the feedback, updates and optimizes the strategy through the reinforcement learning algorithm, and realizes the coordinated optimization and dynamic balance of multiple energy systems.

[0050] The energy storage system optimization management module optimizes the control and parameter configuration of the energy storage system. The neural network prediction constructs a deep learning model to predict the load demand and wind and solar power generation of cold-region power grids. The input layer includes historical load data, meteorological data, time characteristics, and operating status parameters of wind and solar power generation equipment. Through training, the neural network can accurately capture the changing laws of load demand and wind and solar power generation, and provide reliable prediction information for the regulation of the energy storage system. With historical load data as input, the load demand and wind and solar power generation of cold-region power grids are predicted. In terms of control and optimization of the energy storage system, genetic algorithms are applied. For the control module, the charging and discharging power of the energy storage system is used as the decision variable, and the stability index of the power grid, energy cost index, and life loss index of the energy storage system are used. As the optimization objective function, the optimal charging and discharging strategy is found through the search operation of the genetic algorithm. In the optimization module, the genetic algorithm is used to optimize the equipment parameters of the energy storage system, such as the battery capacity, charging and discharging rate, SOC upper and lower limits, etc., while considering the layout and installation location of the energy storage system to improve the performance and life of the energy storage system and reduce costs. In terms of control, the genetic algorithm optimization uses the energy storage system charging and discharging power as the decision variable and the grid stability and other indicators as the objective function to search for the optimal charging and discharging strategy. In terms of optimization, the parameters and layout location of the energy storage equipment are optimized. The prediction results of the neural network provide input for the genetic algorithm optimization. The optimization results of the genetic algorithm are fed back to the neural network to adjust the model parameters, improve the prediction accuracy, and achieve efficient optimization and dynamic balance management of the energy storage system.

[0051] The energy dispatch optimization module optimizes the energy dispatching scheme to achieve reasonable arrangements of various energy sources in terms of power generation distribution, energy storage charging and discharging, energy conversion and coordinated operation. Deep learning feature extraction and prediction uses CNN to analyze the image data of power grid equipment to achieve automatic detection and diagnosis of equipment failures. The LSTM model is used to model and analyze energy data to capture the time-varying patterns of wind and solar power generation, load demand, etc. and their relationship with environmental factors, and predict the power generation potential and load demand trends of different energy sources in the future. The ant colony algorithm dispatch optimization abstracts the power grid energy nodes and transmission lines into a graph structure. The ant colony searches for the optimal energy dispatching scheme based on pheromone concentration and heuristic information in the graph, including the power generation distribution of various energy sources, the charging and discharging strategy of the energy storage system, and the conversion and coordinated operation mode between energy sources. Deep learning provides prediction information as heuristic information for the ant colony algorithm. The results of the ant colony algorithm are fed back to the deep learning model for optimization and improvement, thereby improving energy utilization efficiency and power grid operation efficiency.

[0052] The intelligent decision-making and monitoring module, including the decision support unit and the monitoring and visualization unit, is used to integrate the information of various modules of the system, provide decision support for dispatchers, and monitor and visualize the operation status of the power grid in real time. The decision support unit integrates information from modules such as multi-energy coordinated regulation, energy storage system optimization management, and energy dispatch optimization. Based on big data analysis and artificial intelligence algorithms, it evaluates and predicts the operation status of the power grid in real time, and has scenario simulation and risk assessment functions. For example, when the power grid experiences a load peak or energy supply shortage, the system can quickly analyze various possible response strategies and evaluate their impact on the stability, economy, and environmental protection of the power grid, recommend the optimal dispatch plan, and improve the response capability and resilience of the power grid. The monitoring and visualization unit monitors various parameters of the power grid in real time through sensors and other acquisition devices, establishes a comprehensive monitoring system, and provides real-time Monitor the operating status of each energy device in the cold-region power grid, the flow and distribution of energy, and the operating parameters of the power grid. Collect data through sensors, smart meters and other equipment, and transmit the data to the monitoring center using the communication network. Develop a visualization platform to display the operating data in an intuitive form, provide historical data query and analysis functions, and display the operating data of the power grid in the form of intuitive charts, maps, etc., so that dispatchers can grasp the operation of the power grid in real time. For example, through the visualization interface, you can view the power generation of each energy power station, the power status of the energy storage system, the power flow distribution of the power grid and other information in real time, so as to realize the visual management and monitoring of the power grid. At the same time, the visualization platform can also provide historical data query and analysis functions to help dispatchers understand the operating trends and laws of the power grid and provide reference for decision-making;

[0053] The system communication and security module, including the communication network unit and the security protection unit, is used to connect the data communication between the modules of the system and protect the system from security threats such as network attacks and data leakage. The communication network builds a network that integrates wired and wireless communication technologies. The communication network unit connects various energy nodes, energy storage equipment, control center and monitoring terminal, and adopts a combination of wired and wireless communication technologies, such as optical fiber communication, power line carrier communication, 5G, etc., to ensure rapid data transmission and real-time interaction, establish communication protocols and data interface standards, and carry out seamless communication and data sharing between different systems and equipment, ensuring that energy data collection, multi-energy coordinated regulation, energy storage system optimization management, energy scheduling optimization, and intelligent decision-making and monitoring modules can efficiently transmit information and work together. The security protection unit uses encryption technology, identity authentication and access control to conduct real-time security monitoring and evaluation of key power grid equipment and systems, formulate system safety emergency plans, conduct regular safety training and drills, and maintain the safe and stable operation of cold-region power grids.

[0054] The operation steps of the whole system are as follows:

[0055] Sp1: Data collection and preprocessing: widely collect various types of energy, environment, power grid operation and other data, use a variety of methods to remove data noise and outliers, normalize data and perform stabilization processing;

[0056] Sp2: Multi-energy collaborative control: The global optimization controller of the deep dual Q network (DDQN) obtains the current state information of the power grid, outputs macro-control instructions according to the power grid state, and the fuzzy logic local controller of each energy subsystem performs fine control based on the input variables. The local controller feeds back the operation information, and the global optimization controller updates the strategy to achieve collaborative optimization;

[0057] Sp3: Energy storage system optimization management: Neural network predicts load demand and wind and solar power generation, and genetic algorithm optimizes energy storage charging and discharging strategies and equipment parameters with multiple indicators as the target. The neural network prediction results provide input for genetic algorithm optimization, and the genetic algorithm optimization results are fed back to the neural network to adjust the model parameters to achieve efficient optimization and dynamic balance management of the energy storage system;

[0058] Sp4: Energy scheduling optimization: CNN and LSTM extract equipment image and energy data features respectively to predict power generation potential and load demand, and the ant colony algorithm searches for energy scheduling solutions based on pheromone and heuristic information. The two interact with each other to improve energy scheduling efficiency and quality;

[0059] Sp5: Intelligent decision-making and monitoring: The decision support unit integrates information to evaluate and predict the power grid status and provides dispatch solutions, while the monitoring and visualization unit monitors data and displays it in an intuitive form;

[0060] Sp6: System communication and security: The communication network unit builds a converged network to ensure data transmission and interaction, and the security protection unit uses a variety of technologies to ensure system security.

[0061] Through the contents of the above modules, the system can achieve efficient and coordinated utilization of multiple energy sources, optimized management of energy storage systems, rational scheduling of energy, and intelligent monitoring and decision-making of power grids, thereby improving the stability, reliability and economy of cold-region power grids and providing strong support for energy supply and sustainable development in cold regions. Specific embodiment 2:

[0063] like Figure 1-Figure 3 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0064] In the energy data collection and preprocessing module, the data cleaning algorithm combines a method based on statistical analysis with an anomaly detection method based on machine learning. First, statistical analysis is used to identify data points that are significantly deviated from the normal range, and then the machine learning model is used to further detect potential outliers to ensure the accuracy of the data. The data normalization technology uses the Z-score standardization method to convert the data into a distribution with a mean of 0 and a standard deviation of 1 to meet the requirements of different algorithms for data scale. In addition, in the process of stabilization, the time series analysis method also combines seasonal differences and trend differences in addition to the ARIMA model to more accurately remove the trend and seasonal components in the data, providing a better data foundation for subsequent analysis.

[0065] The statistical analysis method for data cleaning algorithm is as follows:

[0066] First, the collected data set is D = {x 1 , x 2 , ..., x n}, perform preliminary analysis and calculate the mean of the data And standard deviation, this step is to obtain the basic statistical characteristics of the data. The mean reflects the central trend of the data, and the standard deviation measures the degree of dispersion of the data. The standard deviation formula is as follows:

[0067]

[0068] Then, identify data points that deviate significantly from the normal range and set a range |x i -μ|>kσ, where k is a constant set according to experience, for example, k=3. In many cases, data points that meet this condition can be considered outliers, but the specific value can be adjusted according to the characteristics of the data and actual needs. Data points that meet this condition are considered outliers. For example, if the mean of a set of data is 50 and the standard deviation is 10, when k=3, then |x i Data points with -50|>3×10=30 will be initially identified as outliers;

[0069] Machine learning-based anomaly detection methods: Isolation forest algorithm can be used;

[0070] For a data point x, randomly select a feature q and randomly select a split value p within the value range of feature q to divide the data into two parts, one part is less than or equal to p, and the other part is greater than p;

[0071] Repeat this process until the data point x is isolated or reaches a certain tree depth;

[0072] Calculate outlier scores: Outliers usually have shorter path lengths, that is, the path from the root node to the data point passes through fewer nodes. The path length is calculated to determine whether a data point is an outlier.

[0073] Data normalization technique: Z-score standardization method:

[0074] For the data set D = {x 1 , x 2 , …, x n}, normalized data Where μ is the mean and σ is the standard deviation. First, the mean and standard deviation are calculated, which is the same as the steps of calculating the mean and standard deviation in the data cleaning algorithm.

[0075] Then, for each data point x i Perform normalization to obtain normalized data

[0076] For example, there is a set of data {10, 20, 30, 40, 50}. First, calculate the mean μ=30 and the standard deviation:

[0077]

[0078] Then the normalized data are:

[0079] The energy data acquisition and preprocessing module obtains a large amount of multi-source data through widely deployed sensors and data acquisition equipment, including real-time operation data of various energy sources such as wind, solar, fire, and nuclear storage, as well as environmental data. These data have different dimensions and numerical ranges. For example, the power generation power may be between a few kilowatts and megawatts, the temperature may be between tens of degrees Celsius and hundreds of degrees Celsius, and the light intensity also has a specific range. Directly using these raw data for analysis and calculation will cause the algorithm to have large differences in sensitivity to different features, which may cause certain features to dominate the algorithm while ignoring other important features.

[0080] After removing noise, outliers and erroneous data through the data cleaning algorithm, the Z-score standardization method is used to convert the data into a distribution with a mean of and a standard deviation of . After this processing, data with different characteristics are at the same scale, so that subsequent data analysis and algorithm models (such as neural networks, reinforcement learning, etc.) can treat each feature more fairly and improve the accuracy and stability of the model. For example, in the multi-energy coordinated control module, the deep reinforcement learning algorithm needs to make decisions based on the current state of the power grid (including various characteristics such as the power generation power of each energy source, load demand, energy storage capacity, environmental parameters, etc.). After data normalization, these features are at the same scale, and the algorithm can more accurately learn the relationship and rules between them, thereby outputting a more reasonable macro-control strategy. Specific embodiment three:

[0082] like Figure 1-Figure 3 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0083] In the multi-energy coordinated control module, the deep reinforcement learning algorithm adopts the deep dual Q network DDQN. Based on the traditional DQN, it separates the calculation of the target Q value and the selection of actions. During the training process, it sets the experience replay buffer size to a specific value to make full use of historical experience data. At the same time, it uses a small batch stochastic gradient descent method to update the network parameters and optimize the convergence speed and stability of the algorithm. In addition, according to the characteristics of the cold-region power grid and the energy distribution, a suitable reward function is designed, and the energy supply stability is comprehensively considered, including the proportion of time that voltage and frequency fluctuations are within the allowable range, cost-effectiveness, the difference between energy production costs and electricity sales revenue, carbon emission reductions, and the operating life of energy equipment. Factors such as these enable the algorithm to learn the optimal multi-energy coordinated control strategy.

[0084] The deep double-Q network further includes the following:

[0085] Sp1: Estimate of the Q-value function: The Q-value function Q(s, a; θ) represents the estimated value of taking an action in state s, with parameter θ;

[0086] Sp2: Target Q value calculation: Target Q value y = r + γmax in traditional DQN a′ Q(s′, a′; θ), where r is the current reward, γ is a discount factor that weighs the importance of current rewards and future rewards, usually between 0 and 1, s′ is the next state, and a′ is the action in the next state;

[0087] DDQN separates the calculation of the target Q value and the selection of actions. First, the action a′=argmax is selected through the current network. a′Q(s′, a′; θ), that is, the optimal action is selected based on the current network’s estimate of the Q value of the next state, and then the target network is used to calculate the target Q value y=r+γQ(s′, a′; θ - ), where θ - The parameters of the target network are usually copied from the current network at regular intervals, such as every 100 iterations, to increase the stability of the algorithm.

[0088] Sp3: Calculation of loss function: Loss function: L(θ) = (yQ(s, a; θ)) 2 , update the network parameters θ by the stochastic gradient descent method. In each iteration, the target Q value y and the current network estimated Q value Q(s, a; θ) are calculated according to the current state s, the action a taken, the reward r obtained, and the next state s′. Then, the gradient is calculated according to the loss function, and the network parameters are updated so that the network estimated Q value gradually approaches the target Q value.

[0089] Sp4: Reward function design:

[0090] Energy Supply Stability Rewards Among them, α is the weight coefficient, T within is the time that voltage and frequency fluctuations are within the allowable range, T total is the total time. By monitoring the voltage and frequency of the power grid, the proportion of time that the voltage and frequency fluctuations are within the allowable range within a certain period of time is counted. This is used as a quantitative indicator of energy supply stability and a corresponding reward is given. The higher the stability of the power grid, the greater the reward value, which encourages the algorithm to learn a control strategy that can maintain the stable operation of the power grid.

[0091] Cost-effectiveness reward R cost =β(cost production -cost revenue ), where β is the weight coefficient, cost production is the energy production cost, cost revenue The revenue from electricity sales is measured by calculating the difference between the cost of energy production and the revenue from electricity sales. The larger the reward value, the better the cost-effectiveness. This encourages the algorithm to consider economic factors in the regulation process, minimize costs and increase revenue.

[0092] Carbon emission reduction reward R carbon =γ×reduction carbon , where γ is the weight coefficient, reduction carbonThe reduction in carbon emissions. For the links that generate carbon emissions in the energy production process, such as thermal power plants, the reduction in carbon emissions under different control strategies is calculated and corresponding rewards are given to encourage the algorithm to choose a more environmentally friendly control strategy, reduce carbon emissions, and meet the requirements of sustainable development.

[0093] Energy equipment operating life bonus R life =δ×life remaining , where δ is the weight coefficient, life remaining The remaining life of the equipment is estimated by monitoring the operating status and historical data of energy equipment, and rewards are given. The larger the reward value, the longer the remaining life of the equipment. This encourages the algorithm to avoid overuse of equipment during the regulation process, extend the service life of the equipment, and reduce the cost of equipment replacement.

[0094] Total reward: R = R stability +R cost +R carbon +R life ,In each iteration, the rewards of each part are calculated according to the actual ,operation of the power grid, and then they are added to get the total reward, which is used to update the ,parameters of the deep double-Q network, so that the algorithm can learn the ,optimal multi-energy coordinated regulation strategy that comprehensively ,considers energy supply stability, cost-effectiveness, carbon emission reduction, and equipment life.

[0095] In the multi-energy coordinated control module, the fuzzy logic controller of each energy subsystem is designed with targeted input and output variables and fuzzy rule base, which further includes the following contents:

[0096] The input variables of the fuzzy logic controller of the wind turbine include wind speed deviation, wind turbine speed deviation and power generation deviation, and the output variables are pitch angle adjustment and generator torque adjustment. The fuzzy rule base is formulated based on expert experience and a large amount of experimental data, covering control rules in different wind speed ranges and operating conditions;

[0097] The input variables of the fuzzy logic controller of the thermal power unit include grid load demand deviation, coal quality parameters, unit temperature deviation, etc. The output variables are fuel supply adjustment, water supply adjustment and steam pressure adjustment. The fuzzy rule base considers the optimal control strategy under different load changes and equipment states.

[0098] The fuzzy logic controller of the nuclear power unit focuses on input variables such as reactor power deviation and coolant flow deviation, and the output variable is the movement position adjustment of the control rod, etc., to ensure that the nuclear power unit can achieve efficient power regulation under the premise of safe operation;

[0099] The input variables of the fuzzy logic controller of the energy storage system are the energy storage power deviation, the charge and discharge power limit deviation and the peak and valley period mark of the power grid. The output variable is the charge and discharge power adjustment amount. The fuzzy rule base formulates reasonable charge and discharge strategies according to different power states and power grid requirements to achieve effective utilization and energy balance of the energy storage system.

[0100] The fuzzy logic controller, taking a wind turbine as an example, further includes the following contents:

[0101] Sp1: Determine input and output variables: The input variables of the fuzzy logic controller of the wind turbine include wind speed deviation Δv, wind turbine speed deviation Δn and power generation deviation ΔP. These input variables reflect the difference between the current operating state and the expected state of the wind turbine. The output variables are the pitch angle adjustment Δβ and the generator torque adjustment ΔT, which are used to adjust the operating state of the wind turbine to make it closer to the optimal operating state.

[0102] Sp2: Define fuzzy sets: define fuzzy sets for each input and output variable. For example, for the wind speed deviation Δv, the fuzzy sets can be defined as "low", "medium", "high", etc. For the pitch angle adjustment Δβ, the fuzzy sets can be defined as "small adjustment", "medium adjustment", "large adjustment", etc. Each fuzzy set has a corresponding membership function, which is used to describe the degree to which the input and output variables belong to the fuzzy set;

[0103] Sp3: Develop a fuzzy rule base: Develop a fuzzy rule base based on expert experience and a large amount of experimental data. For example, a rule can be "If the wind speed deviation is high and the wind turbine speed deviation is low, then the pitch angle adjustment is a large adjustment and the generator torque adjustment is a medium increase." The rule base covers control rules in different wind speed ranges and operating states. These rules are used to achieve fine control of wind turbines.

[0104] Sp4: Fuzzy reasoning: After obtaining the current input variable values ​​(wind speed deviation, wind turbine speed deviation and power generation deviation), the membership of each input variable to each fuzzy set is calculated according to the membership function. Then, fuzzy reasoning is performed according to the fuzzy rule base to obtain the membership of the output variables (pitch angle adjustment and generator torque adjustment) on each fuzzy set.

[0105] Sp5: Defuzzification: Use appropriate defuzzification methods (such as the centroid method, maximum membership method, etc.) to convert the fuzzy values ​​of the output variables obtained by fuzzy reasoning into precise values, and obtain the final pitch angle adjustment and generator torque adjustment, which are used to control the operation of the wind turbine.

[0106] In the multi-energy coordinated control module, the global optimization controller of the deep dual-Q network performs macro-control at the top level and outputs macro-control instructions such as the power distribution ratio of each energy subsystem and the overall charging and discharging strategy of the energy storage system. However, the operating characteristics and control requirements of each energy subsystem of wind, solar, thermal and nuclear storage are different, which require more refined control. Therefore, local controllers based on fuzzy logic are set up in the middle layer.

[0107] Taking wind turbines as an example, the fuzzy logic controller obtains accurate pitch angle adjustment and generator torque adjustment according to the real-time operating status of the wind turbine, input variables such as wind speed deviation, wind turbine speed deviation and power generation deviation through fuzzy reasoning and defuzzification process. In this way, real-time and accurate adjustments can be made according to the actual situation of the wind turbine, so that it can operate as efficiently as possible under different wind speeds and operating conditions. At the same time, it can also better respond to the macro-control instructions of the global optimization controller to achieve coordinated optimization and dynamic balance of multi-energy systems. For example, when the global optimization controller requires the wind turbine to increase the power generation according to the overall load demand and energy supply of the power grid, the fuzzy logic controller of the wind turbine will reasonably adjust the pitch angle and generator torque according to the actual operating status such as the current wind speed deviation to achieve an increase in power generation, and ensure the safe and stable operation of the wind turbine during the adjustment process. Specific embodiment four:

[0109] like Figure 1-Figure 3 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0110] In the energy storage system optimization management module, the deep learning model used in neural network prediction is a combination of multi-layer long short-term memory network (LSTM) and convolutional neural network (CNN), which further includes the following contents:

[0111] LSTM is used to process time series data, such as historical load data and time characteristics, to capture their long-term dependencies and trend changes. The LSTM layer contains multiple hidden layer units, each of which controls the flow and memory of information through a gating mechanism;

[0112] CNN is used to process meteorological data and image data of wind and solar power generation equipment, extract spatial features and local patterns. The CNN layer uses a specific convolution kernel size and number for feature extraction, and then fuses the features extracted by the two and inputs them into the fully connected layer to predict load demand and wind and solar power generation. The network training uses the Adam algorithm, an optimization algorithm with adaptive learning rate, which automatically adjusts the learning rate according to the gradient information during the training process, speeds up the convergence speed and improves the training effect.

[0113] LSTM further includes the following:

[0114] LSTM part:

[0115] Forget Gate t =σ(W f ·[h t-1 , x t ]+b f ), where σ is the sigmoid function, which sets the input value between 0 and 1 to determine the cell state C at the previous moment t-1 How much information needs to be forgotten? f is the weight matrix of the forget gate, b f is the bias term of the forget gate, h t-1 is the hidden state at the previous moment, x t is the input at the current moment, for example, if f t The value of is close to 0, which means that most of the cell state information at the previous moment is forgotten; if f t If the value of is close to 1, it means that most of the cell state information at the previous moment is retained;

[0116] Input Gate i t =σ(W i ·[h t-1 , x t ]+b i ), which is used to determine how much new information needs to be stored in the cell state at the current moment, where W i is the weight matrix of the input gate, b i is the bias term of the input gate;

[0117] Candidate Status Among them, W C is the weight matrix of the candidate state, b C It is the bias term of the candidate state, which maps the value to between -1 and 1 through the tanh function to generate a candidate cell state;

[0118] Current state Among them, ⊙ represents element-by-element multiplication. Through the forget gate and the input gate, the cell state at the previous moment and the candidate cell state at the current moment are weighted summed to obtain the cell state at the current moment, thus realizing the update of the cell state and the transmission of information.

[0119] Output gate o t =σ(W o ·[h t-1 , x t ]+b o ), which is used to determine how much information of the current cell state needs to be output to the hidden state, where W o is the weight matrix of the output gate, b ωis the bias term of the output gate;

[0120] The current hidden state h t =o t ⊙tanh(C t ), multiply the output of the output gate by the cell state after function processing element by element, and obtain the hidden state at the current moment, which is passed to the next time step or other layer as the output of the LSTM unit;

[0121] The CNN part further includes the following:

[0122] Convolutional layer output O conv =W conv* X+b conv , where W conv is the convolution kernel, * indicates the convolution operation, X is the input data, b conv is the bias term;

[0123] Pooling layer output O pool =poling(O conv ), where poling represents the pooling operation;

[0124] After feature fusion, it is input into the fully connected layer for prediction. Among them, W fc is the weight matrix of the fully connected layer, b fc is the bias term of the fully connected layer;

[0125] The network training adopts the Adam algorithm, an optimization algorithm with adaptive learning rate. The formula for updating parameters is:

[0126]

[0127] Among them, β 1 is the exponential decay rate of the first-order moment estimate, m t is the first-order moment estimate, is the gradient of the loss function at the current moment, θ t is the parameter at the current moment;

[0128] Among them, β 2 is the exponential decay rate of the second-order moment estimate, v t is the second-order moment estimate;

[0129] Where t is the current iteration number;

[0130] Here, α is the learning rate and ∈ is a small constant to prevent the denominator from being zero.

[0131] The pooling layer uses the maximum pooling or average pooling method to downsample the feature map output by the convolution layer. For example, the maximum pooling method selects the maximum value in the local area of ​​the feature map as the output of the area, while the average pooling method calculates the average value of the local area as the output. This can reduce the amount of data while extracting the main features, reducing the computational complexity, and improving the robustness of the model and its adaptability to features of different scales.

[0132] Feature fusion and prediction: The time series features extracted by LSTM and the spatial features extracted by CNN are fused. For example, the outputs of the two can be spliced ​​together and then input into the fully connected layer to predict the load demand and wind and solar power generation. The prediction results are as follows:

[0133]

[0134] Among them, W fc is the weight matrix of the fully connected layer, b fc is the bias term of the fully connected layer. The fully connected layer processes the fused features comprehensively to obtain the final prediction value, i.e., the prediction value of load demand and wind and solar power generation power;

[0135] First, initialize the network parameters, including the weight matrix and bias term of LSTM, CNN and fully connected layers, as well as the parameters of the Adam algorithm, such as the learning rate α, which can be set to 0.001 initially and can be adjusted according to the training situation, and the exponential decay rate β of the first-order moment estimate 1 , usually set to 0.9, the exponential decay rate of the second-order moment estimate β 2 , usually set to 0.999, a small constant ∈, to prevent the denominator from being zero;

[0136] In each training iteration: calculate the predicted value of the current batch of data Calculate the loss function based on the actual value Where θ represents all parameters of the network;

[0137] Calculate the gradient of the loss function with respect to each parameter

[0138] Update first moment estimate and the second moment estimate

[0139] Compute the corrected first moment estimate and the second moment estimate

[0140] Update network parameters

[0141] Energy storage systems play a key role in balancing energy supply and demand and improving energy efficiency in cold-region power grids. Accurately predicting the load demand and wind and solar power generation in cold-region power grids is crucial for optimal management of energy storage systems.

[0142] LSTM can process time series data, such as historical load data and time characteristics, and capture their long-term dependencies and trend changes. For example, load demand often has daily, weekly, monthly and other periodic changes. LSTM can effectively learn and memorize these time series characteristics through its gating mechanism, and better predict future load change trends.

[0143] CNN is good at processing meteorological data and image data of wind and solar power generation equipment, extracting spatial features and local patterns. For example, the light intensity and temperature in meteorological data have certain distribution characteristics in space, and CNN can extract these features through convolution operations. The image data of wind and solar power generation equipment also contains information such as the status and appearance of the equipment. CNN can extract useful features from them, which are of great significance for the prediction of wind and solar power generation.

[0144] The fusion of features extracted by LSTM and CNN can comprehensively consider information in time and space dimensions, more comprehensively capture the changing patterns of load demand and wind and solar power generation, and provide more reliable prediction information for the regulation of energy storage systems. In terms of energy storage system control, genetic algorithms are used to optimize the charging and discharging strategies of energy storage systems based on the predicted load demand and wind and solar power generation, so as to achieve goals such as grid stability, reduce energy costs, and extend the life of energy storage systems. At the same time, the results of genetic algorithm optimization are fed back to the neural network to adjust model parameters and improve prediction accuracy, forming a closed-loop optimization process and continuously improving the optimization management level of energy storage systems.

[0145] In the energy storage system optimization management module, when the genetic algorithm optimizes the energy storage system, the chromosome encoding adopts real number encoding to directly represent the equipment parameters and operation strategy variables of the energy storage system, including battery capacity, charge and discharge rate, SOC upper and lower limits, and charge and discharge power, etc., and further includes the following contents:

[0146] The selection operation adopts the roulette selection method, which calculates the probability of being selected according to the fitness value of the individual, so that individuals with high fitness have a greater chance of being selected and inherited to the next generation;

[0147] The crossover operation uses the real number crossover method to exchange genes at the corresponding positions of the chromosomes of the two selected parent individuals to generate new offspring individuals. The crossover probability is set to an appropriate value based on experiments and experience.

[0148] The mutation operation uses a uniform mutation method to randomly select several gene positions on the chromosome of an individual and make small random changes to them with a certain mutation probability to increase the diversity of the population;

[0149] The algorithm's fitness function comprehensively considers grid stability indicators, including voltage deviation, frequency deviation, energy cost indicators, including the investment cost, operating cost and electricity purchase cost of energy storage equipment, as well as the life loss indicator of the energy storage system, and searches for the optimal energy storage system configuration and operation strategy through multiple iterations.

[0150] The legacy algorithm further includes the following:

[0151] Sp1: Chromosome encoding: using real number encoding to directly represent the equipment parameters and operation strategy variables of the energy storage system, X = (x 1 , x 2 , …, x m ), where x i It can be battery capacity, charge and discharge rate, SOC upper and lower limits, and charge and discharge power, etc. This encoding method allows each chromosome to directly correspond to the configuration and operation strategy of an energy storage system, which is convenient for operation and optimization in the genetic algorithm;

[0152] Sp2: Initialize the population: randomly generate a certain number of chromosomes as the initial population. The population size is determined according to the complexity of the problem and the computing resources.

[0153] Sp3: Fitness function calculation:

[0154] Grid stability indicators:

[0155] F stability =λ 1 (voltage_deviation+frequency_deviation)

[0156] Among them, λ 1 is the weight coefficient, voltage_deviation is the voltage deviation, which is calculated by monitoring the difference between the voltage of the power grid and the rated voltage, and frequency_deviation is the frequency deviation. Smaller voltage and frequency deviations indicate a more stable power grid, so the smaller the value of this part in the fitness function, the better;

[0157] Energy cost indicators:

[0158] F cost =λ 2 (investment_cost+operation_cost+purchase_cost)

[0159] Among them, λ 2 is the weight coefficient, investment_cost is the investment cost of the energy storage equipment, operation_cost is the operating cost, including equipment maintenance costs, energy consumption costs, etc., purchase_cost is the cost of purchasing electricity. This part of the indicator hopes to minimize the total cost, so the smaller the value of this part in the fitness function, the better;

[0160] Energy storage system life loss indicators:

[0161] F life =λ 3 ×life_loss

[0162] Among them, λ 3 is the weight coefficient, life_loss is the life loss of the energy storage system. The smaller the life loss, the higher the fitness. Therefore, the smaller the value of this part in the fitness function, the better.

[0163] Fitness function:

[0164]

[0165] For each chromosome (i.e., energy storage system solution), calculate its corresponding grid stability index, energy cost index, and energy storage system life loss index, and then substitute them into the fitness function to calculate the fitness value. The higher the fitness value, the better the solution.

[0166] Sp4: The selection operation adopts the roulette selection method:

[0167] Calculate the probability of individual i being selected:

[0168]

[0169] Among them, f(i) is the fitness value of the individual, and N is the population size;

[0170] For example: suppose there are 3 individuals, whose fitness values ​​are f 1 =20, f 2 =30, f 3 =50, then the total fitness The probability of individual 1 being selected The probability of individual 2 being selected The probability of individual 3 being selected According to the calculated probability, individuals are selected using a roulette wheel. For example, a random number between 0 and 1 can be generated. If the random number is less than or equal to 0.2, individual 1 is selected; if the random number is greater than and less than or equal to 0.5 (0.2+0.3), individual 2 is selected; if the random number is greater than 0.5, individual 3 is selected. In this way, individuals with high fitness have a greater chance of being selected and inherited to the next generation, thereby allowing the population to evolve in a better direction.

[0171] Sp5: Crossover operation (real number crossover method):

[0172] The two selected parent individuals X 1 =(x 11 , x 12 , ..., x 1m ) and X 2 =(x 21 , x 22 , ..., x 2m ), gene exchange is performed at the corresponding position of the chromosome to generate a new offspring individual X′ 1 =(x′ 11 , x′ 12 , …, x′ 1n ) and X′ 2 =(x′ 21 , x′ 22 , …, x′ 2m ), for example x′ 1j =αx 1j +(1-α)x 2j , x′ 2j =αx 2j +(1-α)x 2j , where α is a random number between [0, 1];

[0173] First, according to a certain crossover probability (set according to experiments and experience, for example, it can be set to 0.8), select the parent individual pairs that need to be crossover. Suppose two parent individuals X are selected. 1 =(x 11 , x 12 , …, x 1m ) and X 2 =(x 21 , x 22 , …, x 2m );

[0174] For each gene position (i.e. each element in the chromosome), a random number α between 0 and 1 is randomly generated, and then a gene exchange is performed to generate a new offspring individual X′ 1 =(x′ 11 , x′ 12 , …, x′1m ) and X′ 2 =(x′ 21 , x′ 22 , …, x′ 2m ), where x′ 1j =αx 1j +(1-α)x 2j , x′ 2j =αx 2j +(1-α)x 1j , for example, α = 0.6, x 11 =10,x 22 =20, then x′ 11 =0.6×10+(1-0.6)×20=14,x′ 21 =0.6×20+(1-0.6)×10=16

[0175] Through the crossover operation, new individuals can be generated, combining the characteristics of the parent individuals, increasing the diversity of the population, and facilitating the search for a better solution;

[0176] Sp6: Mutation operation (uniform mutation method):

[0177] Randomly select several gene positions on the individual's chromosome, set the gene position j, and change it with a certain mutation probability p m Make a small random change to it, x′ ij =x ij +Δx, where Δx is a random number between [-δ, δ] and δ is the variation range set according to the problem;

[0178] Sp7: Iterative optimization: Repeat the selection, crossover and mutation operations to generate a new population, then calculate the fitness value of each individual in the new population, and iterate this process until certain termination conditions are met, such as reaching the maximum number of iterations (set to 100 times, which can be adjusted according to actual conditions) or the fitness value no longer improves significantly. In each iteration, the individuals in the population continue to evolve and gradually approach the optimal energy storage system configuration and operation strategy;

[0179] In the energy storage system optimization management module, the genetic algorithm is used to optimize the charging and discharging strategies and equipment parameters of the energy storage system. By defining the fitness function, the grid stability index, energy cost index and energy storage system life loss index are comprehensively considered, so that the algorithm can balance multiple objectives in the process of searching for the optimal solution;

[0180] The selection operation is performed according to the fitness value of the individual. Individuals with high fitness have a greater chance of being selected and inherited to the next generation, which ensures that excellent energy storage system solutions can be continued and spread in the population. The crossover operation generates new offspring individuals with different feature combinations by exchanging the genetic information of the parent individuals, increasing the diversity of solutions, and helping to explore a broader solution space and find better energy storage system configurations and strategies. The mutation operation avoids the algorithm from falling into the local optimal solution to a certain extent, introduces new genetic changes to the population, and enables the algorithm to jump out of the local optimal solution and continue to search for the global optimal solution.

[0181] The load demand and wind and solar power generation predicted by the neural network provide input information for the genetic algorithm. Based on these prediction results, the genetic algorithm can optimize the charging and discharging strategies of the energy storage system in a more targeted manner to adapt to future energy supply and demand conditions. At the same time, the results of the genetic algorithm optimization are fed back to the neural network to adjust the model parameters and improve the prediction accuracy, so that the neural network can more accurately predict the load demand and wind and solar power generation, and further provide more reliable input for the genetic algorithm, forming a mutually reinforcing and continuously optimized closed-loop system, realizing efficient optimization and dynamic balance management of the energy storage system, and improving the overall performance and economic benefits of the cold-region power grid. Specific embodiment five:

[0183] like Figure 1-Figure 3 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0184] In the energy dispatch optimization module, in the deep learning feature extraction and prediction part, the CNN model uses a specific convolution layer structure and pooling layer structure when analyzing the image data of power grid equipment. The convolution layer uses convolution kernels of different sizes and numbers to extract different features of the equipment image, such as the appearance features and temperature distribution features of the equipment. The pooling layer uses the maximum pooling or average pooling method to downsample the feature map output by the convolution layer to reduce the amount of data and extract the main features. When modeling and analyzing energy data, the LSTM model reasonably sets the time step and the number of hidden layer neurons to better capture the time series characteristics and dynamic change laws of energy data. During the model training process, batch normalization technology is used to normalize the input data to accelerate the training convergence speed of the model. At the same time, Dropout technology is used to prevent overfitting and improve the generalization ability of the model. The training data is generated using the sliding window method, and the parameters of the model are continuously updated to enable it to adapt to the real-time changing needs of energy dispatching in cold-region power grids.

[0185] CNN model analyzes power grid equipment image data:

[0186] Convolutional layer output Among them, l represents the number of convolutional layers, is the convolution kernel of the lth layer is the output of the previous convolutional layer, is the bias term of the lth layer;

[0187] Pooling layer output Use the maximum pooling or average pooling method to downsample the feature map output by the convolutional layer;

[0188] LSTM model modeling and analysis of energy data:

[0189] Forget Gate t =σ(W f ·[h t-1 , x t ]+b f ), input gate i t =σ(W i ·[h t-1 , x t ]+b i ), candidate status Current state Output gate o t =σ(W o ·[h t-1 , x t ]+b o ), the current hidden state h t =o t ⊙tanh(C t ), where x t is the energy data input at the current moment, h t-1 is the hidden state of the previous moment, and the parameter meaning is the same as in the previous LSTM part;

[0190] Batch normalization technique: For input data x, calculate the mean and variance Among them, m is the batch size, the normalized data Then zoom and pan Among them, γ and β are learnable parameters;

[0191] Dropout technology: During the training process, the output of the neuron is randomly set to zero with probability p, that is, r~Bernoulli(p), then the output of the neuron becomes y=r×x, where x is the input of the neuron. The Dropout technology randomly discards some neurons during the training process so that the model will not be overly dependent on certain specific neurons during training, thereby increasing the generalization ability of the model and preventing overfitting. In the testing phase, Dropout is usually not used, but the output of all neurons is multiplied by to adjust to approximately simulate the effect during training.

[0192] Training data generation and model update: The training data is generated using the sliding window method. During the training process, the model parameters are continuously updated, and the optimization algorithm (such as the Adam algorithm) is used to adjust the model's weights and biases according to the gradient of the loss function (such as the mean square error, etc.), so that it can adapt to the real-time changes in the energy dispatch of the cold-region power grid and improve the model's ability to predict energy data.

[0193] In the energy scheduling optimization module, accurately predicting the power generation potential and load demand trend of different energy sources in the future is the key to achieving reasonable energy scheduling. By analyzing the image data of power grid equipment, the CNN model can extract the appearance characteristics, temperature distribution and other information of the equipment. This information is of great significance for evaluating the operating status and potential failure risks of the equipment, which in turn can affect the power generation capacity and reliability of the energy. For example, by analyzing the image data of the transformer, if its temperature distribution is found to be abnormal, it may indicate that the transformer is at risk of overload or failure, and the energy scheduling strategy needs to be adjusted accordingly to avoid energy supply interruption or instability due to equipment failure.

[0194] The LSTM model focuses on processing the time series characteristics of energy data, such as the time-varying patterns of wind and solar power generation, load demand, and their relationship with environmental factors. By reasonably setting the time step and the number of hidden layer neurons, LSTM can capture the long-term dependencies and dynamic change trends of energy data. For example, load demand often has seasonal and cyclical changes. LSTM can learn these patterns and thus more accurately predict future load demand.

[0195] In the process of model training, batch normalization technology and Dropout technology are used to improve the training effect and generalization ability of the model. Batch normalization makes data distribution more stable and accelerates convergence. Dropout prevents overfitting, so that the model can better adapt to different data sets and actual application scenarios. Training data is generated by the sliding window method, and model parameters are continuously updated, so that the model can track changes in energy data in real time and provide accurate prediction information for energy scheduling. These prediction information are provided to the ant colony algorithm as heuristic information, helping the ant colony algorithm to make more reasonable decisions when searching for energy scheduling solutions, improve the efficiency and quality of energy scheduling, realize the reasonable allocation and coordinated operation of energy, and improve the operating efficiency and stability of cold-region power grids.

[0196] In the energy scheduling optimization module, in the scheduling optimization part of the ant colony algorithm, the number of ants is reasonably set according to the scale and complexity of the power grid to ensure that the optimal path can be fully explored in the search space. The initial concentration of pheromones is set to a small non-zero value to prevent the algorithm from converging to the local optimal solution too early. The pheromone volatility coefficient is determined based on experience and experiments to control the decay rate of pheromones on the path and balance the exploration and utilization capabilities of the algorithm. The weight of the heuristic information is adjusted according to the relative importance of factors such as energy cost, power generation efficiency, and transmission loss to guide the ants to choose a better path during the search process. In each iteration, when the ants find a new path, according to the objective function value of the energy scheduling plan corresponding to the path, such as comprehensive consideration of indicators such as energy utilization efficiency, cost and power grid stability, the pheromone concentration on the path is updated to make subsequent ants more likely to choose a better path. At the same time, the maximum number of iterations of the algorithm is set. When the maximum number of iterations is reached or a certain convergence condition is met, the algorithm stops searching and outputs the current optimal energy scheduling plan.

[0197] Ant Colony Algorithm Scheduling Optimization

[0198] Pheromone update formula:

[0199] When ant k finds a path L k Then, according to the objective function value f(L k ) Update the pheromone concentration on the path;

[0200] The pheromone update formula is:

[0201]

[0202] Among them, τ ij (t) is the pheromone concentration on the path (i, j) at time t, ρ is the pheromone volatility coefficient, m is the number of ants, If ant k passes through path (i, j) in this iteration, otherwise Δτ i j k =0, Q is a constant;

[0203] Heuristic information weight adjustment, heuristic information,

[0204]

[0205] Among them, ω 1 ,ω 2 ,ω 3 They are the weights of factors such as energy cost, power generation efficiency, and transmission loss, which are adjusted according to their relative importance. Cost is the energy cost, efficiency is the power generation efficiency, and loss is the transmission loss.

[0206] The operation steps are as follows:

[0207] Sp1: Initialization parameters:

[0208] Determine the number of ants m, and set it reasonably according to the scale and complexity of the power grid. For example, for a medium-sized cold-region power grid, the number of ants can be set to 50 (adjustable according to actual conditions). Too many ants may lead to waste of computing resources and too long search time, while too few ants may not be able to fully explore the search space.

[0209] Set the initial pheromone concentration τ ij (0) is a small non-zero value, for example, it can be set to 0.1 (it can be determined based on experience and experiments to prevent the algorithm from converging to the local optimal solution too early and ensure that the ants have a certain degree of randomness to explore different paths in the early stage of the search;

[0210] Determine the pheromone volatility coefficient ρ, which is determined based on experience and experiments. For example, it is set to 0.5 (the value range is generally between 0 and 1. The pheromone volatility coefficient controls the decay speed of pheromones on the path and balances the exploration and utilization capabilities of the algorithm. If ρ is too large, the pheromone evaporates too quickly, which may cause ants to give up some potential good paths too early; if ρ is too small, the pheromone accumulates too much, and the algorithm may fall into a local optimal solution;

[0211] Determine the weight of heuristic information according to the relative importance of factors such as energy cost, power generation efficiency, and transmission loss 1 ,ω 2 ,ω 3 For example, if energy cost is relatively more important, you can set ω 1 =0.5,ω 2 =0.3,ω 3 =0.2 (the specific value can be adjusted according to the actual situation);

[0212] Sp2: Constructing a graph structure: abstracting the grid energy nodes and transmission lines into a graph structure G = (V, E), where V is a node set representing energy nodes (such as power plants, substations, energy storage stations, etc.), E is an edge set representing transmission lines, and each edge (i, j) corresponds to a pheromone concentration and heuristic information;

[0213] Sp3: Ant search path:

[0214] Each ant starts from the starting node (such as a power plant) and calculates the probability of selecting the next node based on the pheromone concentration and heuristic information from the current node to other adjacent nodes. The calculation formula for the selection probability is

[0215]

[0216] in, is the probability that ant k chooses node j from node i, τ ij is the pheromone concentration on edge (i, j), η ij is the heuristic information, α and β are the weight parameters of pheromone and heuristic information respectively (used to adjust the relative importance of the two in the selection decision), N i is the set of adjacent nodes of a node;

[0217] The ant selects the next node based on the calculated probability and then moves to that node, repeating the process until it reaches the target node (e.g., load center), thus forming a complete path;

[0218] Sp4: Pheromone Update:

[0219] When all ants have completed a path search, they will update the pheromone. Through the pheromone update, the pheromone concentration on the paths corresponding to the better energy scheduling solutions will increase, so that in subsequent iterations, the ants are more likely to choose these paths, guiding the algorithm to search for better solutions.

[0220] Sp5: Iterative Optimization:

[0221] Repeat the process of ant path search and pheromone update, set the maximum number of iterations of the algorithm (for example, set it to 100 times, which can be adjusted according to actual conditions). When the maximum number of iterations is reached or certain convergence conditions are met (such as the optimal solution is no longer significantly improved after several consecutive iterations), the algorithm stops searching and outputs the current optimal energy scheduling plan, that is, the energy scheduling plan corresponding to the path with the highest pheromone concentration, including the power generation distribution of each energy source, the charging and discharging strategy of the energy storage system, and the conversion and coordinated operation mode between energy sources.

[0222] The ant colony algorithm is used to search for the optimal energy scheduling solution in the energy scheduling optimization module. After abstracting the grid energy nodes and transmission lines into a graph structure, the ants continuously search for paths in the graph to find the optimal energy distribution and transmission method. The pheromone concentration and heuristic information jointly determine the probability of ants choosing a path. The pheromone concentration reflects the past search experience of ants on the path. The higher the concentration, the greater the possibility of the path being selected, which reflects the algorithm's use of historical experience. The heuristic information is calculated based on factors such as energy cost, power generation efficiency, and transmission loss, guiding ants to search in a better direction. For example, for paths with lower energy cost, higher power generation efficiency, and lower transmission loss, the value of the heuristic information is larger, and ants are more likely to choose these paths.

[0223] In each iteration, ants select paths based on probability and form an energy scheduling plan, and then update the pheromone concentration on the path according to the objective function value of the plan. Through continuous iteration, the algorithm gradually converges to the optimal or near-optimal energy scheduling plan, which can achieve the reasonable arrangement of various energy sources in power generation power distribution, energy storage charging and discharging, energy conversion and coordinated operation, improve energy utilization efficiency and grid operation benefits, and meet the energy demand and stability requirements of cold-region power grids. At the same time, deep learning provides prediction information as heuristic information for the ant colony algorithm, so that the ant colony algorithm can select paths more specifically during the search process, improve search efficiency and accuracy, and the results of the ant colony algorithm are fed back to the deep learning model for optimization and improvement of the model, further improving the prediction accuracy of the deep learning model. The two work together to achieve the optimization of energy scheduling. Specific embodiment six:

[0225] like Figure 1-Figure 3 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0226] In addition to evaluating and predicting the operating status of the power grid based on big data analysis and artificial intelligence algorithms, the decision support unit in the intelligent decision-making and monitoring module also introduces knowledge graph technology to construct the equipment information, energy data, operating rules, expert experience and other knowledge of the cold-region power grid into a knowledge graph, which is stored and managed through a graph database. In the decision-making process, the semantic reasoning and association analysis capabilities of the knowledge graph are used to provide dispatchers with more comprehensive and in-depth decision-making suggestions. The decision support unit is connected to the external energy market information system to obtain information such as energy prices and market demand in real time, and adjusts the power grid's operating strategy and energy dispatch plan according to market dynamics to maximize economic benefits.

[0227] The knowledge graph can be expressed as G = (V, E), where V is a node set representing entities such as equipment, data, and knowledge in the cold region power grid, and E is an edge set representing the relationship between entities.

[0228] In the decision-making process, we can use the query language of the graph database, such as Cypher, to perform semantic reasoning and association analysis. For example, to query the expert experience related to a certain device, we can express it as:

[0229] MATCH(d:Device)-[:RELATED T O]-

[0230] >(e:ExpertExperience)WHEREd.name=′c

[0231] Where d represents the device node, e represents the expert experience node, and device_name is the name of the device to be queried;

[0232] In the intelligent decision-making and monitoring module, knowledge graph technology provides dispatchers with more comprehensive and in-depth decision-making suggestions. It integrates multi-source knowledge of cold-region power grids, organizes equipment information, energy data, operating rules and expert experience in a structured manner, and forms an interconnected knowledge network.

[0233] Through the semantic reasoning and association analysis capabilities of the knowledge graph, dispatchers can quickly obtain various information and knowledge related to the current operation of the power grid. For example, when a certain device fails in the power grid, the system can quickly find relevant information about the device through the knowledge graph, such as the equipment model, installation time, historical maintenance records, etc. It can also find other related equipment and areas that may be affected. In addition, the knowledge graph can be associated with the corresponding expert experience and processing suggestions to help dispatchers quickly develop reasonable fault handling plans, shorten fault recovery time, and reduce the impact on power grid operations.

[0234] In daily operations, the knowledge graph can be continuously updated and improved based on real-time monitoring data so that it always reflects the latest status and knowledge of the power grid. For example, when new equipment is put into operation or energy data changes, the corresponding knowledge graph nodes and relationships can be updated in a timely manner. In this way, the intelligent decision-making and monitoring module can provide dispatchers with more accurate and timely decision-making support based on the latest knowledge graph, improve the response capability and resilience of the power grid, and ensure the stable operation and efficient management of the cold-region power grid.

[0235] Steps for connecting the decision support unit with the external system:

[0236] Sp1: Establish interface: Establish a data interface with the external energy market information system, determine the format and protocol of data transmission, for example, data interaction can be carried out by using RESTful API or message queue, define the request and response format of the interface, including the fields and data types of the energy price, market demand and other information to be obtained;

[0237] Sp2: Real-time data acquisition: through the established interface, regularly send requests to the energy market information system to obtain real-time energy prices, market demand and other information. For example, send a request to the energy market information system every minute (the time interval can be adjusted according to actual needs) to obtain the latest data;

[0238] Sp3: Data fusion and analysis: Integrate the acquired external energy market information with the grid operation data within the system. For example, conduct correlation analysis between energy prices and grid power generation costs, electricity sales revenue and other data, and conduct comparative analysis between market demand and grid load forecast data.

[0239] Sp4: Decision adjustment: Adjust the operation strategy of the power grid and the energy dispatch plan according to market dynamics. If energy prices rise and market demand is large, consider increasing the power generation of energy units with higher power generation and relatively lower cost, and optimize the charging and discharging strategy of the energy storage system to meet market demand while improving economic benefits. If market demand decreases, the power generation can be appropriately reduced to avoid energy waste, and the energy procurement plan can be adjusted at the same time;

[0240] The connection between the decision support unit and the external energy market information system enables the cold-region power grid to respond to market changes in real time and maximize economic benefits. By obtaining information such as energy prices and market demand, the power grid can flexibly adjust its operation strategy and energy dispatch plan according to market conditions. When energy prices are low, it can appropriately increase energy procurement and storage, use the energy storage system to charge during off-peak electricity price periods, and discharge during peak electricity price periods, to achieve arbitrage and cost savings. At the same time, adjust the power distribution according to market demand to avoid over-generation or under-generation, and improve energy utilization efficiency and the economic benefits of the power grid.

[0241] In addition, understanding the changing trends in market demand can also help the power grid plan equipment maintenance and upgrade plans in advance. For example, if it is predicted that market demand will increase significantly in the future, equipment inspection and maintenance can be arranged in advance to ensure the reliability of the equipment when operating at high loads. At the same time, consider whether it is necessary to add new power generation equipment or upgrade existing equipment to meet the growth in market demand. In this way, through docking and data interaction with external systems, the intelligent decision-making and monitoring module can better coordinate the internal operation of the power grid and the external market environment, and achieve sustainable development of the cold-region power grid and improved economic benefits. Specific embodiment seven:

[0243] like Figure 1-Figure 3 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0244] In the system communication and security module, in addition to encryption technology, identity authentication and access control, system security also introduces intrusion detection system (IDS) and intrusion prevention system (IPS). IDS monitors network traffic and system activities in real time, identifies potential intrusion behaviors and abnormal activities by analyzing traffic patterns and behavioral characteristics, and issues alarms in a timely manner.

[0245] The traditional communication and security module includes the following steps:

[0246] Sp1: Data collection: Real-time monitoring of network traffic and system activities, by deploying sensors at key network nodes or using network monitoring tools to collect information such as network data packets and system logs. For example, port mirroring can be set up on firewalls, switches and other devices to copy network traffic to IDS devices for analysis.

[0247] Sp2: Feature extraction: Analyze the collected data and extract features. These features may include the source address, destination address, port number, protocol type, packet size, packet content pattern, etc. For example, for common network attacks such as port scanning, IDS will focus on a large number of continuous connection requests to different ports; for DDoS attacks, it will focus on a large number of identical or similar packets from multiple source addresses, etc.

[0248] Sp3: Anomaly detection: The extracted features are compared with known attack patterns and normal behavior models. A signature-based detection method can be used, that is, matching detection is performed based on a pre-defined attack feature library (signature library). If the data feature matches an attack feature in the signature library, it is determined that a corresponding attack may exist. At the same time, an anomaly-based detection method can also be used. By establishing a baseline model of normal behavior, when the monitored behavior deviates greatly from the baseline, it is determined to be abnormal behavior, which may be a potential intrusion. For example, under normal circumstances, the traffic of a network node fluctuates within a certain range. If there is a sudden and significant increase in traffic and the duration is long, which exceeds the normal fluctuation range, it may be determined as an anomaly;

[0249] Sp4: Alarm generation: When potential intrusion or abnormal activity is detected, the IDS will issue an alarm in a timely manner. The alarm information may include detailed information such as the type of attack, source address, target address, time, etc., so that security managers can take timely measures to respond. For example, alarm notifications are sent to security managers via email, SMS, or system pop-up windows.

[0250] IPS working steps:

[0251] Sp1: Real-time monitoring: Similar to IDS, IPS also monitors network traffic and system activities in real time, but its monitoring is more in-depth and real-time. IPS is usually connected in series in the network and can directly block or modify network traffic in real time;

[0252] Sp2: Intrusion detection: It uses detection methods similar to IDS, including feature extraction and anomaly detection, to analyze network traffic and determine whether there is intrusion behavior, for example, detecting malicious data packets or traffic patterns that meet attack characteristics;

[0253] Sp3: Defense response: When an intrusion is detected, IPS immediately takes defensive measures. It can prevent the further spread and impact of the attack by discarding malicious data packets, blocking the connection to the attack source, modifying the data packet content, etc. For example, if DDoS attack traffic from a certain IP address is detected, IPS can directly discard these data packets and blacklist the IP address to temporarily block its connection to the network. At the same time, IPS can also protect the attacked system or service, for example, through traffic cleaning and other technologies to ensure that normal network services are not affected;

[0254] Sp4: Logging and reporting: IPS will record all detection and defense operation information and generate detailed log records. These logs include details of the attack, defense measures taken, time and other information for subsequent analysis and auditing. At the same time, IPS can also generate reports regularly to provide security managers with a summary and analysis of network security status, helping managers understand the security situation of the network and adjust security policies in a timely manner.

[0255] In the system communication and security module, IDS and IPS play a vital role in protecting the system from security threats such as network attacks and data leakage. As a network security monitor, IDS can detect potential intrusions and abnormal activities in a timely manner by monitoring network traffic and system activities in real time. Its feature extraction and anomaly detection functions can identify various known and unknown attack patterns and provide early warnings for security managers. For example, when hackers try to find system vulnerabilities through port scanning, IDS can detect and issue an alarm in time, giving security managers enough time to take countermeasures, such as strengthening the protection of related ports and initiating emergency response processes.

[0256] IPS takes direct defensive measures when it detects intrusion. It is connected in series in the network and can block attack traffic in real time to prevent the attack from causing substantial damage to the system. For example, when facing a DDoS attack, IPS can quickly discard attack data packets to protect network resources from being exhausted and ensure that the normal operation of the cold-region power grid is not affected. At the same time, the logging and reporting functions of IPS provide security managers with detailed security event records and analysis basis, which helps to conduct security audits and summarize lessons learned afterwards, and further enhance the system's security protection capabilities.

[0257] The two work together, with IDS responsible for detection and early warning and IPS responsible for real-time defense and response, forming a multi-level security protection system to comprehensively ensure the communication security and data security of the integrated control system of wind, solar, thermal and nuclear storage in cold-region power grids, ensure the stable operation and reliability of the system, and provide solid security guarantees for energy supply and sustainable development in cold-region areas. Specific embodiment eight:

[0259] like Figure 1-Figure 3 As shown, according to the contents in the above specific embodiments, the following contents are further disclosed:

[0260] The hardware structure of each module in the system is as follows:

[0261] The hardware of the energy data acquisition and preprocessing module is as follows:

[0262] Sensor Networks:

[0263] For wind farms, wind speed sensors and wind direction sensors are installed. Wind speed sensors can be ultrasonic wind speed sensors or mechanical anemometers. Ultrasonic wind speed sensors have high accuracy and fast response speed and can measure wind speed in real time. Wind direction sensors determine wind direction through mechanical structures or electronic compass technology. At the same time, speed sensors (such as photoelectric speed sensors) and power sensors are installed on wind turbines to obtain wind turbine speed and power generation data.

[0264] In photovoltaic power plants, light intensity sensors (such as silicon photocell sensors) are installed to accurately measure light intensity, and power sensors are installed to monitor the power generation;

[0265] In thermal power plants, pressure sensors are installed to measure steam pressure, flow sensors are installed in coal pipelines to monitor coal consumption, and power sensors are installed on generator sets to obtain unit power;

[0266] Coolant flow sensors are installed around the reactors of nuclear power plants to monitor coolant flow, and power sensors are installed near the core area of ​​the reactor to detect reactor power;

[0267] The energy storage system is equipped with a power sensor and a charge and discharge power sensor. The power sensor (such as a sensor based on the coulomb meter principle) is used to accurately measure the battery state of charge, and the charge and discharge power sensor can obtain the charge and discharge power of the energy storage system in real time.

[0268] Data acquisition equipment:

[0269] A data acquisition terminal (DAT) is used, which has multiple analog and digital input interfaces. It can connect various sensors, convert the analog signals obtained by the sensors into digital signals, and transmit the data to the background processing system through a communication interface (such as RS-485, Ethernet, etc.). These data acquisition terminals usually have a certain data caching function to cope with emergencies such as network interruptions to ensure that data is not lost.

[0270] The hardware of the multi-energy coordinated control module is as follows:

[0271] Compute Server:

[0272] As the hardware platform for running the global optimization controller and fuzzy logic local controller of the deep double Q network (DDQN), the computing server needs to have a high-performance CPU (central processing unit) and GPU (graphics processing unit). The GPU plays an important role in the training and calculation process of the deep reinforcement learning algorithm, which can accelerate the operation of the neural network and improve the computing efficiency. For example, the high-performance GPU server of NVIDIA has a large number of CUDA cores, which can process a large amount of data in parallel and meet the high requirements of the deep reinforcement learning algorithm for computing resources;

[0273] The server should also have large-capacity memory and high-speed storage devices. The memory is used to store intermediate data and model parameters during the algorithm operation process. High-speed storage devices (such as solid-state drives) can quickly read and store training data and model files to ensure rapid convergence and stable operation of the algorithm.

[0274] Communication interface equipment:

[0275] Equipped with communication interface devices such as Ethernet network cards, it is used to interact with the energy data acquisition and preprocessing module to obtain the current status information of the power grid, and send control instructions to the control devices of each energy subsystem. The communication interface device needs to support high-speed and stable data transmission to ensure that the control instructions with high real-time requirements can be transmitted accurately.

[0276] The hardware of the energy storage system optimization management module is as follows:

[0277] Hardware monitoring equipment:

[0278] For energy storage systems, such as battery packs, a battery management system (BMS) is installed. BMS is a hardware system that integrates multiple sensors and controllers. It can monitor the voltage, current, temperature and other parameters of the battery in real time, and analyze and process these parameters through the internal microcontroller unit (MCU). For example, when the battery temperature is too high, the BMS can control the cooling system to dissipate heat to ensure the safe and stable operation of the battery.

[0279] Computing Devices:

[0280] High-performance computing devices are also needed to run neural network predictions and genetic algorithms. The computing devices can be dedicated workstations or servers, equipped with high-performance CPUs and GPUs similar to those in the multi-energy collaborative control module. When training neural networks, these hardware resources can accelerate the model training process and improve prediction accuracy. For the operation of genetic algorithms, sufficient computing resources can speed up the algorithm's search speed and find the optimal energy storage system configuration and operation strategy more quickly.

[0281] The hardware of the energy scheduling optimization module is as follows:

[0282] Image acquisition equipment (for CNN analysis):

[0283] Install high-definition cameras around power grid equipment to collect image data of power grid equipment. These cameras need to have high resolution, good light adaptability, and protection functions such as dust and water resistance to ensure stable operation in complex industrial environments. For example, industrial-grade IP67 protection grade network cameras can normally collect images in harsh weather conditions and dusty environments, providing clear image data such as equipment appearance and temperature distribution for CNN analysis;

[0284] Data storage and computing equipment (for LSTM and ant colony algorithms):

[0285] Equipped with large-capacity data storage devices, such as disk arrays, to store historical energy data, equipment image data, etc., to provide data support for the training of the LSTM model and the iteration of the ant colony algorithm. At the same time, high-performance computing servers are required to run the LSTM model and the ant colony algorithm. The configuration requirements of the server are similar to those of the previous module. Powerful CPUs and GPUs are required to accelerate the model training and algorithm search process to achieve effective modeling of energy data and optimization of energy scheduling solutions.

[0286] The hardware of the intelligent decision-making and monitoring module is as follows;

[0287] Monitoring center hardware facilities:

[0288] Establish a monitoring center and configure servers to receive and process data from various sensors and acquisition devices. These servers need to have multi-threaded processing capabilities to handle a large number of data access requests at the same time. Set up a large display screen in the monitoring center to visualize the grid operation data. For example, a spliced ​​LED display screen can be used to display the power flow distribution of the power grid, the power generation capacity of each energy power station, the power status of the energy storage system and other information in the form of intuitive charts and maps, which is convenient for dispatchers to view in real time.

[0289] Data storage server (for historical data query):

[0290] Equipped with large-capacity data storage servers to store the historical operation data of the power grid. These data storage servers can adopt distributed storage architecture, such as Ceph, to improve the reliability and scalability of data storage. Through the data storage server, dispatchers can query the historical operation trends and rules of the power grid to provide reference for decision-making.

[0291] The system communication and security module hardware is as follows:

[0292] Communication network hardware:

[0293] Wired communication equipment: Optical fiber communication equipment is used, including optical fiber transceivers, optical switches, etc. Optical fiber communication has the characteristics of fast transmission speed and strong anti-interference ability, which is suitable for long-distance and large-capacity data transmission between the control center and the main energy nodes. At the same time, for some short-distance data transmission scenarios, power line carrier communication equipment can be used to use existing power lines for data transmission, reducing wiring costs;

[0294] Wireless communication equipment: Deploy 5G base stations or other wireless communication base stations to connect mobile monitoring equipment and some energy equipment that is difficult to connect via wired means. 5G communication has the characteristics of low latency and high bandwidth, which can meet the needs of energy data transmission with high real-time requirements;

[0295] Security hardware:

[0296] Install firewall equipment. The firewall can filter data packets entering and leaving the network based on preset rules to prevent unauthorized access. At the same time, it is equipped with intrusion detection system (IDS) and intrusion prevention system (IPS) hardware equipment. IDS equipment monitors network traffic and system activities in real time by deploying sensors at key nodes in the network. IPS equipment can take active measures when intrusion behavior is detected, such as blocking attack traffic, to ensure the network security of the system. Encryption equipment, such as hardware encryption machines, can also be used to encrypt sensitive data to ensure the security of data during transmission and storage.

[0297] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprising a reference structure" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0298] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dynamic balance system for integrated regulation of wind, solar, thermal, and nuclear power storage in cold-region power grids, characterized by: The system includes an energy data acquisition and preprocessing module, a multi-energy coordinated control module, an energy storage system optimization management module, an energy scheduling optimization module, an intelligent decision-making and monitoring module, and a system communication and security module, wherein: The energy data acquisition and preprocessing module is used to comprehensively collect the operating data and related environmental data of various energy sources in the cold-region power grid, and clean, normalize and preprocess them. Through widely deployed sensors and data acquisition equipment, the real-time operating data of various energy sources such as wind, solar, fire and nuclear storage are obtained, including power generation power, equipment temperature, operating status, and external environmental data such as air temperature, wind speed, light intensity, humidity and operating parameters such as voltage, current and frequency of the power grid. The data cleaning algorithm is used to remove noise, outliers and erroneous data, and the data normalization technology is used to unify the scale of data of different dimensions, and the time series analysis method is used to stabilize the data, etc., to provide an accurate and reliable data basis for the subsequent analysis and decision-making of the system; The multi-energy coordinated control module is used for the coordinated management and optimized control of multiple energy sources such as wind, solar, thermal, and nuclear energy. It is based on the global optimization of reinforcement learning, adopts a deep reinforcement learning algorithm, takes the current state of the power grid as input, and outputs macro-control instructions such as the power distribution ratio of each energy subsystem and the overall charging and discharging strategy of the energy storage system through interactive learning with the power grid environment. Based on the local control of fuzzy logic, a fuzzy logic controller is set for each energy subsystem, and fine-tuned control is performed according to their respective input variables. The global optimization of reinforcement learning provides guidance for the local control of fuzzy logic, and the local control uploads the operation feedback information for global optimization to update and optimize the strategy; The energy storage system optimization management module optimizes the control and parameter configuration of the energy storage system. The neural network predicts and constructs a deep learning model, takes historical load data as input, and predicts the load demand and wind and solar power generation of the cold-region power grid. In terms of control, the genetic algorithm optimization uses the energy storage system charging and discharging power as the decision variable, and the grid stability and other indicators as the objective function to search for the optimal charging and discharging strategy. In terms of optimization, the parameters and layout positions of the energy storage equipment are optimized. The prediction results of the neural network provide input for the genetic algorithm optimization, and the optimization results of the genetic algorithm are fed back to the neural network to adjust the model parameters. The energy scheduling optimization module realizes the reasonable arrangement of various energy sources in power generation distribution, energy storage charging and discharging, energy conversion and coordinated operation by optimizing the energy scheduling plan. The deep learning feature extraction and prediction adopts CNN to analyze the image data of power grid equipment, and uses the LSTM model to model and analyze the energy data to predict the power generation potential and load demand trend of different energy sources. The ant colony algorithm scheduling optimization abstracts the power grid energy nodes and transmission lines into a graph structure. The ant colony searches for the optimal energy scheduling plan in the graph based on the pheromone concentration and heuristic information. Deep learning provides prediction information as heuristic information for the ant colony algorithm. The results of the ant colony algorithm are fed back to the deep learning model for optimization and improvement, thereby improving energy utilization efficiency and power grid operation efficiency. The intelligent decision-making and monitoring module includes a decision support unit and a monitoring and visualization unit, which are used to integrate the information of various modules of the system, provide decision support for dispatchers, and monitor and visualize the operation status of the power grid in real time. The decision support unit integrates the information of modules such as multi-energy coordinated regulation, and evaluates and predicts the operation status of the power grid in real time based on big data analysis and artificial intelligence algorithms. It has scenario simulation and risk assessment functions. The monitoring and visualization unit monitors various parameters of the power grid in real time through sensors and other acquisition equipment, and transmits them to the monitoring center through the communication network. A visualization platform is developed to display the operation data in an intuitive form, and provide historical data query and analysis functions; The system communication and security module includes a communication network unit and a security protection unit, which are used to connect the data communication between the various modules of the system and protect the system from security threats such as network attacks and data leakage. The communication network builds a network that integrates wired and wireless communication technologies, connects various nodes and devices in the power grid, establishes communication protocols and data interface standards, and conducts seamless communication and data sharing between different systems and devices. The system security adopts encryption technology, identity authentication and access control, conducts real-time security monitoring and evaluation of key power grid equipment and systems, formulates system security emergency plans, conducts regular safety training and drills, and maintains the safe and stable operation of the cold-region power grid.

2. The dynamic balance system for integrated regulation of wind, solar, thermal and nuclear power storage in cold regions power grid according to claim 1 is characterized by: In the energy data collection and preprocessing module, the data cleaning algorithm combines a statistical analysis-based method with a machine learning-based anomaly detection method. First, statistical analysis is used to identify data points that are significantly deviated from the normal range, and then a machine learning model is used to further detect potential outliers to ensure data accuracy. The data normalization technology uses the Z-score standardization method to convert the data into a distribution with a mean of 0 and a standard deviation of 1 to meet the requirements of different algorithms for data scales. In addition, in the process of stabilization, the time series analysis method also combines seasonal differences and trend differences in addition to the ARIMA model to more accurately remove trend and seasonal components in the data, providing a better data foundation for subsequent analysis.

3. The dynamic balance system for integrated control of wind, solar, thermal and nuclear power storage in cold-region power grid according to claim 1 is characterized by: In the multi-energy collaborative control module, the deep reinforcement learning algorithm adopts a deep dual Q network DDQN. On the basis of the traditional DQN, it separates the calculation of the target Q value and the selection of the action. During the training process, the experience replay buffer size is set to a specific value to make full use of historical experience data. At the same time, the small batch stochastic gradient descent method is used to update the network parameters, optimize the convergence speed and stability of the algorithm, and design a suitable reward function according to the characteristics of the cold-region power grid and the energy distribution. The energy supply stability is comprehensively considered, including the time proportion of voltage and frequency fluctuations within the allowable range, cost-effectiveness, the difference between energy production cost and electricity sales revenue, carbon emission reduction, and the operating life of energy equipment, so that the algorithm can learn the optimal multi-energy collaborative control strategy.

4. The dynamic balance system for integrated regulation of wind, solar, thermal and nuclear power storage in cold-region power grid according to claim 1 is characterized by: In the multi-energy coordinated control module, the fuzzy logic controller of each energy subsystem is designed with targeted input and output variables and fuzzy rule base, further including the following contents: The input variables of the fuzzy logic controller of the wind turbine include wind speed deviation, wind turbine speed deviation and power generation deviation, and the output variables are pitch angle adjustment and generator torque adjustment. The fuzzy rule base is formulated based on expert experience and a large amount of experimental data, covering control rules in different wind speed ranges and operating conditions; The input variables of the fuzzy logic controller of the thermal power unit include grid load demand deviation, coal quality parameters, unit temperature deviation, etc. The output variables are fuel supply adjustment, water supply adjustment and steam pressure adjustment. The fuzzy rule base considers the optimal control strategy under different load changes and equipment states. The fuzzy logic controller of the nuclear power unit focuses on input variables such as reactor power deviation and coolant flow deviation, and the output variable is the movement position adjustment of the control rod, etc., to ensure that the nuclear power unit can achieve efficient power regulation under the premise of safe operation; The input variables of the fuzzy logic controller of the energy storage system are the energy storage power deviation, the charge and discharge power limit deviation and the peak and valley period mark of the power grid. The output variable is the charge and discharge power adjustment amount. The fuzzy rule base formulates reasonable charge and discharge strategies according to different power states and power grid requirements to achieve effective utilization and energy balance of the energy storage system.

5. The dynamic balance system for integrated regulation of wind, solar, thermal and nuclear power storage in cold-region power grid according to claim 1 is characterized by: In the energy storage system optimization management module, the deep learning model used in the neural network prediction is a combination structure of a multi-layer long short-term memory network (LSTM) and a convolutional neural network (CNN), further including the following contents: LSTM is used to process time series data, such as historical load data and time characteristics, to capture their long-term dependencies and trend changes. The LSTM layer contains multiple hidden layer units, each of which controls the flow and memory of information through a gating mechanism; CNN is used to process meteorological data and image data of wind and solar power generation equipment, extract spatial features and local patterns. The CNN layer uses a specific convolution kernel size and number for feature extraction, and then fuses the features extracted by the two and inputs them into the fully connected layer to predict load demand and wind and solar power generation. The network training uses the Adam algorithm, an optimization algorithm with adaptive learning rate, which automatically adjusts the learning rate according to the gradient information during the training process, speeds up the convergence speed and improves the training effect.

6. The dynamic balance system for integrated regulation of wind, solar, thermal and nuclear power storage in cold regions power grid according to claim 1 is characterized by: In the energy storage system optimization management module, when the genetic algorithm optimizes the energy storage system, the chromosome encoding adopts real number encoding to directly represent the equipment parameters and operation strategy variables of the energy storage system, including battery capacity, charge and discharge rate, SOC upper and lower limits, and charge and discharge power, and further includes the following contents: The selection operation adopts the roulette wheel selection method, which calculates the probability of being selected according to the fitness value of the individual, so that individuals with high fitness have a greater chance of being selected and inherited to the next generation; The crossover operation uses the real number crossover method to exchange genes at the corresponding positions of the chromosomes of the two selected parent individuals to generate new offspring individuals. The crossover probability is set to an appropriate value based on experiments and experience. The mutation operation uses a uniform mutation method to randomly select several gene positions on the chromosome of an individual and make small random changes to them with a certain mutation probability to increase the diversity of the population. The algorithm's fitness function comprehensively considers grid stability indicators, including voltage deviation, frequency deviation, energy cost indicators, including the investment cost, operating cost and electricity purchase cost of energy storage equipment, as well as the life loss indicator of the energy storage system, and searches for the optimal energy storage system configuration and operation strategy through multiple iterations.

7. The dynamic balance system for integrated control of wind, solar, thermal and nuclear storage in cold-region power grid according to claim 1 is characterized by: In the energy scheduling optimization module, in the deep learning feature extraction and prediction part, the CNN model adopts a specific convolution layer structure and pooling layer structure when analyzing the image data of power grid equipment. The convolution layer uses convolution kernels of different sizes and numbers to extract different features of the equipment image, the appearance features of the equipment, the temperature distribution features, etc. The pooling layer uses the maximum pooling or average pooling method to downsample the feature map output by the convolution layer to reduce the amount of data and extract the main features. When the LSTM model is modeling and analyzing the energy data, the time step and the number of hidden layer neurons are reasonably set to better capture the time series characteristics and dynamic change laws of the energy data. During the model training process, the batch normalization technology is used to normalize the input data to accelerate the training convergence speed of the model. At the same time, the Dropout technology is used to prevent overfitting and improve the generalization ability of the model. The training data is generated by the sliding window method, and the parameters of the model are continuously updated so that it can adapt to the real-time changing needs of the energy scheduling of the cold-region power grid.

8. The dynamic balance system for integrated regulation of wind, solar, thermal and nuclear power storage in cold-region power grid according to claim 1 is characterized by: In the energy scheduling optimization module, in the scheduling optimization part of the ant colony algorithm, the number of ants is reasonably set according to the scale and complexity of the power grid to ensure that the optimal path can be fully explored in the search space. The initial concentration of pheromones is set to a small non-zero value to prevent the algorithm from converging to the local optimal solution too early. The pheromone volatility coefficient is determined based on experience and experiments to control the decay rate of pheromones on the path and balance the exploration and utilization capabilities of the algorithm. The weight of the heuristic information is adjusted according to the relative importance of factors such as energy cost, power generation efficiency, and transmission loss to guide the ants to choose a better path during the search process. In each iteration, when the ants find a new path, according to the objective function value of the energy scheduling plan corresponding to the path, such as comprehensive consideration of indicators such as energy utilization efficiency, cost and power grid stability, the pheromone concentration on the path is updated to make subsequent ants more likely to choose a better path. At the same time, the maximum number of iterations of the algorithm is set. When the maximum number of iterations is reached or a certain convergence condition is met, the algorithm stops searching and outputs the current optimal energy scheduling plan.

9. The dynamic balance system for integrated regulation of wind, solar, thermal and nuclear power storage in cold regions power grid according to claim 1 is characterized by: In addition to evaluating and predicting the operating status of the power grid based on big data analysis and artificial intelligence algorithms, the decision support unit in the intelligent decision-making and monitoring module also introduces knowledge graph technology to construct the equipment information, energy data, operating rules, expert experience and other knowledge of the cold-region power grid into a knowledge graph, which is stored and managed through a graph database. In the decision-making process, the semantic reasoning and association analysis capabilities of the knowledge graph are used to provide dispatchers with more comprehensive and in-depth decision-making suggestions. The decision support unit is connected to the external energy market information system to obtain information such as energy prices and market demand in real time, and adjusts the power grid's operating strategy and energy dispatch plan according to market dynamics to maximize economic benefits.

10. The dynamic balance system for integrated regulation of wind, solar, thermal and nuclear power storage in cold regions power grid according to claim 1 is characterized by: In the system communication and security module, in addition to encryption technology, identity authentication and access control, system security also introduces intrusion detection system (IDS) and intrusion prevention system (IPS). IDS monitors network traffic and system activities in real time, identifies potential intrusion behaviors and abnormal activities by analyzing traffic patterns and behavioral characteristics, and issues alarms in a timely manner.

Images