Distributed energy storage cluster coordinated scheduling method based on climate change

Abstract

The invention discloses a distributed energy storage cluster coordinated scheduling method based on climate change, and belongs to the technical field of power distribution network energy storage scheduling. The method aims at solving the technical problems that in the prior art, due to the fact that climate factors are neglected, energy storage equipment performance prediction deviation is large, and a scheduling strategy is mismatched with an actual operation environment. Photovoltaic / wind power output and load characteristics are dynamically corrected by establishing a multi-dimensional climate impact factor model, a climate perception scheduling model fusing LSTM time sequence feature extraction and a PPO reinforcement learning algorithm is constructed, and an optimized charging and discharging instruction of climate sensitive energy storage is generated; meanwhile, the mode switching of the non-climate-sensitive energy storage cluster is triggered through the photovoltaic power threshold after climate correction, and the dynamic balance of the storage of the photovoltaic surplus electric energy and the compensation of the power supply gap is realized. According to the method, a dynamic closed-loop response mechanism is formed by coupling the multi-dimensional reward function with climate dynamic adaptation and instruction tracking optimization, and the response precision of energy storage cluster scheduling is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage scheduling, and particularly relates to a distributed energy storage cluster coordination scheduling method based on climate change. BACKGROUND

[0002] In the traditional field of energy storage scheduling, a scheduling strategy based on fixed rules is generally used to regulate the power of the energy storage cluster. This method is difficult to adapt to real-time changes in power demand, especially as the scale of the energy storage cluster continues to expand, its scheduling flexibility and refinement are significantly insufficient, resulting in low system operating power. Existing technologies attempt to improve this problem by constructing an optimization model with the goal of minimizing load fluctuations, such as by regulating the charging and discharging power in each time period to improve response capability, however, this method often ignores the dynamic influence of climate conditions on energy storage performance and power grid load, resulting in limited applicability of the scheduling results in actual complex environments.

[0003] Chinese patent document CN112670999A discloses a real-time voltage control method for low-voltage distribution networks based on user-side flexible resources, but this patent focuses on local voltage stability control, adjusting the reactive or active power of user-side flexible resources (such as photovoltaic, electric vehicles) through cloud platform and edge node coordination to achieve voltage adjustment. Its technical means does not involve the coupling analysis of climate factors and global operation of the power grid, and lacks consideration of distributed energy storage in multi-objective coordinated scheduling (such as overall planning of electricity market and auxiliary service market), climate adaptive optimization, etc. Therefore, when dealing with the global coordinated scheduling problem of large-scale distributed energy storage clusters, there are still technical limitations such as insufficient adaptability and poor economic efficiency. SUMMARY

[0004] The present application aims to provide a distribution network distributed energy storage cluster coordination scheduling method based on historical climate data, solving the technical problems of large performance prediction deviation of energy storage devices, mismatch between scheduling strategy and actual operating environment caused by ignoring climate factors in existing technologies.

[0005] The present application also aims to realize adaptive optimization scheduling of energy storage clusters under variable climate conditions by combining climate impact factor modeling with reinforcement learning algorithms, improving the stability of distribution network operation and energy utilization power.

[0006] The present application also aims to establish a multi-time scale high-precision scheduling framework to overcome the response lag and insufficient global optimization capability of traditional methods in dealing with large-scale distributed energy storage cluster coordinated control.

[0007] The application provides a distributed energy storage cluster coordination scheduling method based on climate change, which comprises the following steps: a multi-dimensional climate influence factor model is established according to historical climate data, and the upper limit of photovoltaic / wind power output and the load characteristics of a power distribution network are dynamically corrected; a climate-aware scheduling model is constructed, a state space is constructed by fusing climate factors and SOC time sequence characteristics through LSTM, and a PPO algorithm is used to generate charge-discharge instructions of climate-sensitive energy storage in the climate-corrected power interval; the charge-discharge mode switching of non-climate-sensitive energy storage clusters is triggered by judging the photovoltaic power threshold corrected by the climate, and the dynamic balance of photovoltaic excess power storage and power supply gap compensation is assisted in scheduling; and the climate is dynamically adapted through a multi-dimensional reward function and is subjected to instruction tracking optimization in a dynamic closed loop. Through the establishment of the multi-dimensional climate influence factor model and the construction of the climate-aware scheduling model, the dynamic and accurate correction of the photovoltaic / wind power output and the load characteristics of the power distribution network is realized, and the adaptability and response accuracy of the energy storage cluster scheduling to the climate conditions are significantly improved.

[0008] Preferably, the multi-dimensional climate influence factor model is constructed in the following manner: the charge-discharge power of the energy storage equipment and the load curve are corrected according to the temperature factor, the upper limit of the photovoltaic unit output is corrected based on the sunshine factor, and the upper limit of the wind power unit output is corrected by using the wind speed factor; and the weight coefficients of the temperature factor, the sunshine factor and the wind speed factor are obtained by training the historical climate data. Through the cooperative correction mechanism of the temperature factor, the sunshine factor and the wind speed factor, the charge-discharge power of the energy storage equipment, the load curve and the upper limit of the new energy unit output are comprehensively optimized, and the economy and reliability of the system operation are improved.

[0009] Preferably, the method adopts a time series analysis method to process the historical climate data, establishes a correlation model of the climate parameters and the load characteristics of the power distribution network, determines the initial weights of the temperature factor, the sunshine factor and the wind speed factor through multiple regression analysis, and dynamically optimizes the weight coefficients by using a neural network, so that the climate influence factor can reflect the correction demand of the current environment on the charge-discharge power in real time. The use of the time series analysis combined with the dynamic optimization of the weight coefficients by the neural network enables the climate influence factor to accurately reflect the correction demand of the environment change on the charge-discharge power in real time, thereby enhancing the adaptability and accuracy of the model.

[0010] Preferably, when the PPO algorithm generates the charge-discharge instructions, the action space is normalized to the interval [-1, 1] and mapped to the actual charge-discharge power range corrected by the climate; and the total charge-discharge power needs to meet the total power demand of the power distribution network coordination scheduling, and the target consistency is obtained through penalty constraints. Through the action space normalization and the climate correction power interval mapping mechanism, the physical feasibility of the charge-discharge instructions is ensured, and the consistency of the total power of the cluster and the scheduling target is ensured through the penalty constraints.

[0011] As preferred, the method adds a scheduling matching degree penalty term in the reinforcement learning reward function, calculates the absolute deviation of the actual total charging and discharging power from the scheduling target of the power distribution network; and drives the PPO algorithm to adjust the action output through the back propagation mechanism, so that the cluster total power converges to the target power demand range. By adding the scheduling matching degree penalty term and the back propagation mechanism, the deviation of the actual charging and discharging power from the scheduling target is effectively reduced, and the precision and convergence speed of the cluster power control are improved.

[0012] As preferred, the time sequence features extracted by the LSTM include: historical fluctuation trend of climate factors, historical charging and discharging behavior and SOC evolution trajectory of each energy storage unit, and current period climate prediction value; and the state space includes power frequency modulation capacity demand and power energy technical parameters of the power distribution network. By using the LSTM to extract the time sequence features of climate factors and SOC, and combining the power frequency modulation capacity demand and power energy technical parameters of the power distribution network, the information integrity of the state space is enhanced, and comprehensive data support is provided for intelligent decision-making.

[0013] As preferred, the method corrects the maximum / minimum capacity limit of the energy storage device based on the climate factor; calculates the optimal SOC working interval in real time in combination with the power frequency modulation capacity demand of the power distribution network; and automatically compresses the SOC adjustment interval to prevent battery overcharging / overdischarging when an extreme temperature environment is detected. Based on the dynamic correction of the capacity limit of the energy storage device and the SOC working interval based on the climate factor, and the automatic compression of the adjustment interval in the extreme temperature environment, the battery overcharging / overdischarging is effectively prevented, and the system safety and device life are improved.

[0014] As preferred, the multi-dimensional reward function includes: a climate adaptability reward term to improve the charging and discharging power regulation accuracy in an extreme temperature environment; an SOC balance penalty term to maintain the state of charge of the cluster within a set adjustment interval; and a scheduling matching degree penalty term to take the deviation of the actual charging and discharging curve from the target value as the optimization basis. Through the design of the multi-dimensional reward function, the climate adaptability, SOC balance and scheduling matching degree are simultaneously optimized, and the comprehensive performance and stability of the energy storage cluster scheduling are comprehensively improved.

[0015] As preferred, the method extracts power energy technical parameters and regulation service technical parameters as reference values; calculates the deviation amount of the actual charging and discharging power of each energy storage unit from the declared parameters; and multiplies the deviation amount by the distributed energy storage unit charging and discharging technical parameters as the core coefficient of the penalty term. Through the deviation amount penalty mechanism based on technical parameters, the actual operation of each energy storage unit is close to the declared parameters, and the predictability and economy of market participation are improved.

[0016] As preferred, the objective function of the climate-aware scheduling model is to minimize the climate-corrected total power loss, including the traditional unit start-stop loss, power generation output loss and energy storage charging / discharging power loss; the marginal node power parameter is obtained by solving the load balance constraint dual multiplier, and the frequency modulation service technical parameter is determined in combination with the branch transfer distribution factor. With the objective function of minimizing the climate-corrected total power loss, and by solving the dual multiplier and the branch transfer distribution factor to determine the technical parameter, the collaborative optimization of system loss minimization and frequency modulation service is realized.

[0017] The present application has the following advantages: 1. The scheduling method of the present application can effectively control the load fluctuation of the power grid, and the utilization power of the power resource is higher when the energy storage units in the energy storage cluster meet the energy supply demand and the frequency modulation demand at the same time.

[0018] 2. The present application introduces a reinforcement learning algorithm to adaptively optimize the charging and discharging decision-making problem in a complex dynamic environment. Through continuous iteration and updating of the state-action space, the present application realizes efficient response and optimal control of the energy storage cluster in uncertain scenarios such as changes in power grid demand, and has good learning ability and dynamic adaptability.

[0019] 3. The present application integrates energy management and auxiliary service demand into a unified optimization framework, taking into account power tracking and energy capacity constraints, and considering system regulation capability and economic targets, solving the problem of inconsistent strategies when energy and power indicators conflict, and improving the executability and service coverage of the scheduling scheme. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a method flowchart of the present application.

[0021] Figure 2 is a distributed energy storage cluster coordinated scheduling time sequence power curve diagram of the present application.

[0022] Figure 3 is a distributed energy storage scheduling result diagram of the present application. DETAILED DESCRIPTION

[0023] According to Figure 1 As shown in the figure, relying on a certain comprehensive energy management platform in Jiashan, a climate-aware operation + cluster scheduling module is designed in the platform test environment, and the case verification of the proposed distributed energy storage cluster coordinated scheduling method for distribution network based on historical climate data is carried out. The platform accesses 224 energy storage projects (including 55 user-side energy storage), totaling 37.38 megawatts (including 36.08 megawatts of user-side energy storage and 1.3 megawatts of base station energy storage), which together form a distributed energy storage cluster, and access the Jiaxing distribution network scheduling platform through communication port modification.

[0024] The integrated energy management platform first obtains long-term historical climate data from the weather station, including solar intensity, wind speed, temperature, humidity, and other multi-dimensional weather parameters. These data cover complete weather records over the past few years. The platform uses advanced time series analysis methods to clean, normalize, and extract features from these massive climate data, and establishes a correlation model between climate parameters and power grid operation characteristics through machine learning algorithms. Based on multivariate regression analysis and neural network dynamic optimization technology, the platform determines the weight coefficients of key climate impact factors such as solar factor and wind speed factor, and forms a set of climate impact factor calculation model that can reflect the impact of environmental changes on power grid operation in real time. This model can dynamically correct the photovoltaic power generation prediction output curve and load characteristics according to the current weather conditions, providing a solid climate perception foundation for subsequent high-precision scheduling.

[0025] After the calculation of climate correction factors, the platform applies them to the photovoltaic power generation prediction of Tianhong Foundry Energy Storage Project and Zhongxin Industrial Park Energy Storage Project. The photovoltaic installed capacity of Tianhong Foundry Energy Storage Project is 1260kW, and its annual average power generation data is included in the correction model for analysis; the photovoltaic installed capacity of Zhongxin Industrial Park is 2.55MW, with an annual average power generation of about 2371306kWh, and the platform dynamically adjusts its output limit according to the real-time solar factor. Through the photovoltaic power threshold judgment mechanism after climate correction, the platform can accurately identify the surplus or shortage of photovoltaic power generation in the current period and generate corresponding energy storage scheduling instructions accordingly. This process fully considers the variation characteristics of photovoltaic output under different seasons and weather conditions, ensuring the accuracy and adaptability of the scheduling strategy.

[0026] For non-climate-sensitive energy storage clusters, including 174 lithium battery energy storage units of the China Telecom base station and the container energy storage project of Changsheng Bearings, the platform uses reinforcement learning scheduling method to assist in scheduling based on power grid scheduling targets and market bidding results (such as Figure 2 The total capacity of the China Telecom energy storage project is 1.3MW, distributed in various communication base stations, forming a distributed energy storage network; the total capacity of the Changsheng Bearings energy storage project is 1MW / 2.15MWh, completed in two phases, and designed with standard container type. The scheduling strategy of these energy storages generates charge and discharge instructions based on real-time electricity price signals, frequency modulation requirements, and SOC state, and realizes intelligent decision-making in complex operating environments through the deep reinforcement learning framework combining PPO algorithm and LSTM network.

[0027] When the climate correction model shows that the photovoltaic power generation is high, the platform instructs the non-climate sensitive energy storage clusters to perform charging operations, effectively storing the excess power generated by photovoltaic power generation; on the contrary, when the photovoltaic output is insufficient or encounters rainy weather, these energy storage clusters immediately switch to discharging mode to compensate for the power supply gap in time. During the entire scheduling process, the platform optimizes the instruction tracking accuracy through a multi-dimensional reward function, ensuring that the total power of the clusters Figure 3 The scheduling results shown remain consistent. The scheduling system monitors the charging and discharging states, SOC changes, and power output of each energy storage unit in real time, continuously adjusts the control strategy through a closed-loop feedback mechanism, and realizes the dynamic balance and optimized cooperation of photovoltaic power generation and energy storage response.

[0028] The entire scheduling process relies on the actual operating environment of Jiashan platform, and realizes data sharing and collaborative control with Jiaxing power distribution network scheduling platform through a dedicated communication port. The platform dynamically adjusts the operating behavior of each energy storage cluster in combination with the market bid curve of Figure 2 and the scheduling results of Figure 3 The system displays the operating states of four energy storage projects, including the operating states of the new industry park, the communication base station, the Changsheng bearing, and the Tianhong casting, including charging and discharging power, SOC changes, operating power, and other key parameters. Through the comparison and analysis of actual operating data, it is proved that the method of combining climate-aware scheduling with non-climate sensitive energy storage auxiliary scheduling can effectively improve the operating power of the power distribution network and the renewable energy consumption capacity, and verifies the practicality and reliability of the invention under complex climate conditions.

[0029] In the climate data acquisition and processing stage, Jiashan platform adopts multi-level data fusion technology to perform spatio-temporal alignment processing on the original climate data provided by the meteorological station and the historical operating data of the power distribution network. The platform uses time series analysis method to deeply mine the climate data of the past years, and establishes the correlation model of temperature, humidity, wind speed, sunshine, and other parameters with the load characteristics of the power distribution network. This processing process ensures that the climate influence factors can accurately reflect the actual influence of local climate characteristics on the operation of the power grid, providing a reliable data basis for subsequent scheduling.

[0030] In the climate influence factor model construction process, the platform uses machine learning algorithm to dynamically optimize the weight coefficients of each climate parameter. After determining the initial weight through multivariate regression analysis, the neural network is used for further optimization and adjustment, so that the climate influence factor can respond to environmental changes in real time. Especially for the sunshine factor which has a significant impact on photovoltaic clusters, the platform analyzes the relationship between historical irradiance data and photovoltaic output curve, and establishes an accurate sunshine-output correction model, which provides an important basis for photovoltaic power generation prediction.

[0031] In terms of scheduling model construction, the platform establishes a multi-objective optimization model based on climate-corrected parameters. The model aims to minimize the total loss of the distribution network, while considering multiple aspects such as climate-corrected power balance constraints, branch power flow constraints, and frequency regulation requirements. For the two photovoltaic energy storage projects in Tianhong Foundry and Zhongxin Industrial Park, the platform specially sets up a photovoltaic power threshold triggering mechanism. When the actual power generation reaches the set threshold, the auxiliary scheduling program is automatically started to ensure the timely storage and utilization of excess photovoltaic power.

[0032] In the implementation of the reinforcement learning algorithm, the platform uses a deep reinforcement learning framework combining PPO algorithm and LSTM network. Through the LSTM network, the time series features of climate factors and energy storage states are extracted, and a state space is constructed containing multiple dimensions of information such as climate information, distribution network operating state, and energy storage cluster state. The action space is defined as the charging and discharging power range of each energy storage unit after climate correction, and normalization processing is performed to ensure the feasibility of the instructions.

[0033] In terms of reward function design, the platform sets up multiple evaluation indicators, including climate adaptability reward, scheduling matching degree penalty, and operating technical performance reward. Through the synergistic effect of these reward mechanisms, the reinforcement learning model is guided to achieve optimal scheduling while meeting climate constraints. In particular, in terms of scheduling matching degree, the platform takes the market bid result as the benchmark and ensures the consistency of the actual scheduling result with the market target through the penalty mechanism. Figure 2

[0034] In the actual scheduling instruction generation process, the platform first generates a pre-scheduling scheme based on climate prediction data, and then performs real-time optimization and adjustment through the reinforcement learning algorithm. For photovoltaic clusters, the platform generates charging and discharging instructions based on climate-corrected output prediction; for non-climate-sensitive energy storage, the platform mainly schedules based on market signals and system demand. Throughout the process, the platform continuously optimizes the scheduling strategy through a closed-loop feedback mechanism to ensure the stability and economy of system operation.

[0035] Through the actual operation of the Jiashan platform, the method effectively solves the adverse effects of climate factors on energy storage scheduling, improves the consumption capacity of photovoltaic power generation, and enhances the stability and reliability of distribution network operation. This case fully demonstrates the application value of climate-aware scheduling methods in practical engineering, providing an important reference for subsequent promotion.

[0036] Embodiment Two ​The application proposes a power distribution network distributed energy storage cluster coordination scheduling method based on historical climate data. The application is explained in detail by constructing a historical climate data analysis model, establishing a climate impact factor, explaining the declaration parameters of the distributed energy storage cluster, constructing a power distribution network distributed energy storage cluster scheduling model considering the impact of climate factors, and constructing a distributed energy storage cluster scheduling algorithm based on improved reinforcement learning.

[0037] (1) Historical climate data analysis and climate impact factor establishment The application first collects and analyzes historical climate data to establish a climate impact factor model. Historical climate data includes temperature, humidity, wind speed, precipitation, sunshine duration and other multi-dimensional meteorological parameters, which directly affect the load characteristics of the power distribution network and the operation performance of the energy storage equipment.

[0038] The climate impact factor is defined as the key climate characteristic parameter affecting the operation of the power distribution network by statistical analysis and machine learning modeling of historical climate data. Specifically, the temperature factor affects the power load and the charge and discharge power of the energy storage equipment, the humidity factor affects the insulation performance of electrical equipment, the wind speed factor affects the wind power output prediction, and the sunshine factor affects the photovoltaic power output prediction.

[0039] The historical climate data analysis model uses time series analysis method to process the climate data of the past N years, and establishes the correlation between the climate parameters and the operation parameters of the power distribution network. Through multivariate regression analysis, neural network and other methods, a climate impact factor calculation model is constructed, which can predict the influence degree of current climate conditions on the operation of the power distribution network.

[0040] The climate impact factor calculation formula is defined as the comprehensive climate impact factor of the current period equal to the temperature impact weight multiplied by the temperature standardized value plus the humidity impact weight multiplied by the humidity standardized value plus the wind speed impact weight multiplied by the wind speed standardized value plus the sunshine impact weight multiplied by the sunshine standardized value, wherein each impact weight is obtained by training historical data, and the standardized value is the ratio of the current climate parameter to the historical average value.

[0041] (2) Power distribution network distributed energy storage cluster scheduling under the influence of climate factors When the distributed energy storage cluster participates in the coordinated scheduling of the power distribution network, the impact of climate factors on the performance of energy storage equipment and the load of the power distribution network needs to be considered. Under the action of climate factors, the charge and discharge power, capacity attenuation and power output of energy storage equipment will change, and the load distribution and peak-valley characteristics of the power distribution network will also be adjusted accordingly.

[0042] Distinguish from traditional power distribution network energy storage scheduling method, the scheduling strategy designed in the application fully considers the influence of climate factor, and dynamically adjusts the technical parameters of the energy storage equipment through the climate correction coefficient. When the climate factor indicates a high temperature environment, the charging and discharging power of the energy storage equipment will be reduced, and the scheduling algorithm will be adjusted accordingly; when the climate factor indicates a low temperature environment, the energy storage capacity will be affected, and the scheduling algorithm will optimize the capacity allocation strategy.

[0043] (3) Distributed energy storage cluster scheduling model of power distribution network considering the influence of climate factor In the construction of the distributed energy storage cluster scheduling model of the power distribution network considering the influence of climate factor, unlike existing methods, the model realizes the coupled optimization of climate factors and power distribution network operation, avoiding the shortcomings of ignoring the influence of climate in traditional methods.

[0044] In the selection of the objective function, the power balance of the energy storage cluster in the power distribution network in each time period under the influence of the climate factor is taken as the optimization target to establish an optimization target model. The objective function is defined as the minimization of the power loss of the power distribution network under the influence of the climate factor, and the function is summed for all time periods t. The loss of each time period includes two parts: the first part is the operation loss of the traditional generator set under the influence of the climate factor, including the start-up loss, shutdown loss and power generation output loss (output power multiplied by the loss coefficient after climate correction) of all units i; the second part is the operation loss of the distributed energy storage system under the influence of the climate factor, including the climate corrected power loss during discharging and charging of the energy storage, and the capacity loss of the energy storage participating in the regulation of the power distribution network. The whole objective function aims to optimize the coordinated operation strategy of traditional units and distributed energy storage under the influence of the climate factor, to minimize the total operation loss of the power distribution network under the constraints of meeting the power balance and regulation demand of the power distribution network.

[0045] The constraint conditions include the climate corrected power distribution network load balance constraint, branch power flow constraint, frequency regulation supply and demand constraint and various types of unit constraint, etc.

[0046] 1) The climate corrected power distribution network load balance constraint is defined as the sum of the power generation power of all generator sets, the power generation power of new energy and the discharging power of distributed energy storage, minus the charging power of distributed energy storage, which must be equal to the total load demand of the power distribution network after the climate factor correction in any time period t, and the total load of the power distribution network is equal to the sum of the bus load values of all nodes after the climate factor correction in the time period, wherein the Lagrange dual multiplier of the power distribution network load balance constraint represents the marginal power distribution network power parameter in the non-network blocking state, and the constraint ensures that the power distribution network realizes the instantaneous power balance between the power generation side and the power consumption side under the influence of the climate factor in each scheduling time period.

[0047] 2) The weather-corrected branch power flow constraint is defined as the power flow of the distribution network branch must be within its transmission capacity limit after being corrected by the weather factor, i.e. the branch power flow should be greater than or equal to the negative maximum power flow transmission limit and less than or equal to the positive maximum power flow transmission limit, where the branch power flow is equal to the sum of the product of the generator output power transfer distribution factor of all generators, renewable energy units and distributed energy storage at the nodes to the corresponding power, minus the sum of the product of the generator output power transfer distribution factor of all nodes to the load, the constraint ensures that the power transmission of each line in the distribution network does not exceed its physical carrying capacity under the influence of the weather factor.

[0048] 3) The weather-corrected frequency regulation supply and demand constraint is defined as the distribution network frequency regulation service is scheduled in order according to the comprehensive technical parameters, where the sum of the frequency regulation capacity of all generators and the frequency regulation capacity of distributed energy storage must be greater than or equal to the weather factor corrected distribution network frequency regulation service capacity demand, the sum of the product of the frequency regulation capacity of all generators and its mileage power factor plus the product of the frequency regulation capacity of distributed energy storage and its mileage power factor must be greater than or equal to the weather factor corrected distribution network frequency regulation service mileage demand, the distribution network frequency regulation service capacity demand is equal to the frequency regulation capacity proportion of the distribution network load multiplied by the weather factor corrected total load of the distribution network, and the distribution network frequency regulation service mileage demand is equal to the frequency regulation mileage proportion of the distribution network load multiplied by the weather factor corrected total load of the distribution network.

[0049] 4) The weather-corrected generator constraint: a. Output and ramping limit, the output of the generator should be within its maximum / minimum output range under the influence of the weather factor, where the output constraint requires the output of the generator at time period t to be between the product of the start-stop state variable and the weather-corrected minimum output and the product of the start-stop state variable and the weather-corrected maximum output, the start-stop state variable is a Boolean variable, 0 represents shutdown and 1 represents startup; the ramping constraint requires that the change in output of the generator between adjacent time periods cannot exceed its ramping rate limit under the influence of the weather factor.

[0050] b. Unit start-up and shutdown loss constraint is defined as the start-up loss of the generator is equal to the start-up state variable multiplied by the single start-up loss of the generator under the influence of the weather factor, and the shutdown loss of the generator is equal to the shutdown state variable multiplied by the single shutdown loss of the generator under the influence of the weather factor, the constraint ensures that the loss generated by the unit start-stop operation under the influence of the weather factor can be correctly included in the total operating loss of the distribution network.

[0051] c. The generator set power factor constraint is defined as the sum of the product of the generator set i's climatic correction power factor at time period t and the declared power of all technical segments k of the generator set i at time period t, which ensures that the declared power of each segment is within its declared range and the total declared power is equal to the sum of the declared power of each segment when the generator set participates in the power grid operation according to the segmented technical parameters under the influence of the climate factor.

[0052] 5) The climatic correction new energy generator set constraint is defined as the output constraint of photovoltaic generator sets and wind power generator sets under the influence of the climate factor, wherein the actual output of the photovoltaic generator set at time period t should be greater than or equal to 0 and less than or equal to the predicted output of the photovoltaic generator set at time period t after being corrected by the climate factor, the actual output of the wind power generator set at time period t should be greater than or equal to 0 and less than or equal to the predicted output of the wind power generator set at time period t after being corrected by the climate factor, the predicted output of the photovoltaic generator set at time period t is equal to the standardized output value at time period t multiplied by the rated capacity of the photovoltaic generator set according to the irradiance condition at the location of the generator set and the climate factor, and the predicted output of the wind power generator set at time period t is equal to the standardized output value at time period t multiplied by the rated capacity of the wind power generator set according to the wind condition at the location of the generator set and the climate factor, which ensures that the actual output of the new energy generator set does not exceed the maximum power generation capacity predicted based on the weather condition and the climate factor.

[0053] 6) The climatic correction distributed energy storage cluster constraint: The distributed energy storage cluster constraint is defined as the operating constraint under the influence of the climate factor that needs to be met when the distributed energy storage participates in the coordinated scheduling of the power grid, wherein the capacity of the distributed energy storage at time period t is equal to the capacity of the previous time period plus the charging power of the current time period multiplied by the climate correction charging power coefficient minus the discharging power of the current time period divided by the climate correction discharging power coefficient, the charging power of the distributed energy storage at time period t should be greater than or equal to 0 and less than or equal to the climate correction maximum charging power limit multiplied by the Boolean variable representing the charging state, the discharging power of the distributed energy storage at time period t should be greater than or equal to 0 and less than or equal to the climate correction maximum discharging power limit multiplied by the Boolean variable representing the discharging state, the sum of the Boolean variables representing the charging state and the discharging state should be less than or equal to 1 to ensure that the charging and discharging states are mutually exclusive, and the capacity of the distributed energy storage at time period t should be greater than or equal to the climate correction minimum capacity limit and less than or equal to the climate correction maximum capacity limit, which ensures that the distributed energy storage can safely and reliably participate in the operation of the power grid under the premise of meeting its own physical operating limits and the influence of the climate factor.

[0054] Based on the constructed distributed energy storage cluster scheduling model of the power grid considering the influence of the climate factor, the distributed energy storage cluster can realize the value of electric energy and the value of regulation service by participating in the operation of the power grid, thereby recovering the investment in distributed energy storage technology.

[0055] The present application solves the corresponding constraint by using the power balance constraint of the power distribution network with climate correction, the power flow constraint established by the transfer distribution factor, and the regulation service related constraint to obtain the electrical energy technical parameters, frequency modulation capacity and frequency modulation mileage technical parameters in the power distribution network, wherein the marginal node power parameter considering the power flow limit penalty is equal to the sum of the dual multipliers of the power distribution network load balance constraint plus the product of the node n load of all branches l multiplied by the generator output power transfer distribution factor of the branch l minus the maximum reverse power flow constraint dual multiplier. Unlike the marginal node power parameter calculation in the prior art, the present application considers the dual multipliers of the climate factor influence and the regulation service related constraint to obtain the frequency modulation service technical parameters under the unified scheduling of the power distribution network, and realizes the determination of the scarce technical parameters of the frequency modulation capacity and the frequency modulation mileage in the regulation service.

[0056] (4) Climate-aware distributed energy storage cluster scheduling algorithm based on reinforcement learning Based on the technical parameter declaration and pre-scheduling operation results of the distributed energy storage participating in the coordinated scheduling of the power distribution network, a climate-aware distributed energy storage cluster scheduling sequential decision model is constructed under the DRL framework. The model takes the PPO algorithm as the core and is combined with the LSTM architecture. By using the PPO module in the Stable Baselines3 library based on pytorch, the model can be combined with the LSTM network to form a sequential decision model. This integration enables the decision-making agent to extract the charging and discharging characteristics of the distributed energy storage under the influence of climate factors through the policy_kwargs interface.

[0057] In this study, each type of distributed energy storage participating in the operation of the power distribution network is modeled as a Markov decision process (MDP), defined as a five-tuple: state, action, reward, policy, and parameter. The LSTM network is used to optimize the parameters to extract the technical parameters and flexibility of the distributed energy storage participating in scheduling, providing sufficient information for the agent during the training process.

[0058] 1) State space The state represents the environmental information observed by the agent, including climate information, power distribution network information, distributed energy storage cluster state, and each distributed energy storage unit state (input to the LSTM module to capture sequence features).

[0059] a. Climate information: current climate factor value; historical climate data trend; climate prediction information.

[0060] b. Power distribution network information (coordinated scheduling results): power parameters (electrical energy + regulation service technical parameters) at different time periods; total load response requirements (or total charging and discharging amount to be completed) at each time period.

[0061] c. Energy storage cluster overall state: the weighted average of SOC (State of Charge) under the current cluster state; the total capacity, the remaining available capacity of the cluster; the current time period.

[0062] d. Each distributed energy storage unit state: the current SOC of each energy storage unit; the charging and discharging behavior and SOC evolution of the last n time steps; geographical distribution / physical constraints (such as maximum power, charging and discharging power, etc.).

[0063] 2) Action space The action space is defined as the charging and discharging power of each energy storage unit at the current time step under the influence of climate factors. The power value is first normalized to the interval of negative 1 to positive 1, and then mapped to the maximum allowed charging and discharging range of the energy storage unit after the correction of the climate factor. The normalized action value in the interval of negative 1 to positive 1 corresponds to the actual power of the energy storage unit in the interval of negative climate-corrected maximum discharge power to positive climate-corrected maximum charge power. In addition, the sum of the charging and discharging power of all energy storage units should meet the total power requirement of the distribution network coordinated scheduling, i.e. after the action output, a normalization adjustment or penalty constraint should be added to ensure that the total power of the cluster is consistent with the distribution network scheduling target.

[0064] 3) Reward function The reward function is defined as guiding the reinforcement learning model to achieve the optimality and coordination of cluster scheduling under the premise of meeting the distribution network scheduling results and climate factor constraints. The core purpose is to guide the model to achieve the optimality and coordination of cluster scheduling under the premise of meeting the distribution network scheduling results. The function contains five main components: the climate adaptability reward is set as a weight coefficient multiplied by the adaptability of the energy storage device to the current climate conditions, the distribution network execution matching degree reward is set as a negative weight coefficient multiplied by the absolute value of the deviation between the actual total charging and discharging of the cluster and the distribution network scheduling target, the running technical performance reward is the sum of the technical benefits of each energy storage unit charging and discharging, i.e. the power parameter minus the distributed energy storage unit charging and discharging technical parameter multiplied by the power, the penalty considering battery life and loss technical parameter is a negative weight coefficient multiplied by the sum of the absolute values of the charging and discharging power of all energy storage units, the distribution network balance and running safety reward is a negative weight coefficient multiplied by the sum of the squares of the differences between the SOC of all energy storage units and the target SOC to reward the cluster SOC to remain in the middle interval to reserve the adjustment ability, and the total reward function is equal to the climate adaptability reward plus the running technical performance reward plus the distribution network balance reward plus the distribution network execution matching degree reward plus the battery life loss penalty.

[0065] In order to obtain a robust operation strategy, a combined strategy optimization framework combining PPO and LSTM is introduced. In the PPO algorithm, the Actor network is responsible for generating the operation behavior, and the Critic network is responsible for evaluating the value of the distribution network state under the current policy; both networks are trained through LSTM.

[0066] The Actor and Critic networks are based on the training results of LSTM, and improve the policy through mutual feedback. The running behavior is improved through the policy function, and then the environment returns the reward and the next state, thus completing a round of policy combination. This process is iterated until the training is completed.

[0067] In the LSTM-based policy optimization loop, the environment passes the state sequence to the Actor and Critic networks. In the Actor network, the LSTM is trained by quantifying the difference between the new and old policies, thereby optimizing the policy parameters. In the Critic network, the LSTM is trained by minimizing the mean squared error (MSE) between the advantage policy and the existing policy.

[0068] The PPO algorithm constraint definition is that the PPO algorithm introduces a truncated agent objective function, which improves the power and robustness of the optimization process by constraining the range of policy changes. The PPO clipping agent objective function is the expected value of the minimum of the estimated advantage function multiplied by the probability ratio of the new policy to the old policy, multiplied by the estimated advantage function after applying the clipping function to the probability ratio. The probability ratio is equal to the action probability under the new policy divided by the action probability under the old policy. The estimated advantage function is equal to the reward of the current state plus the discount factor multiplied by the next state value function minus the current state value function. The state value function represents the state value function under the policy. The clipping function limits the probability ratio to the range of 1 minus the clipping parameter to 1 plus the clipping parameter to prevent the policy from updating too much. The discount factor is used to balance the importance of immediate rewards and future rewards.

Claims

1. A distributed energy storage cluster coordination scheduling method based on climate change, characterized in that, The method comprises: A multi-dimensional climate impact factor model is established according to historical climate data to dynamically correct the upper limit of photovoltaic / wind power output and the load characteristics of the power distribution network; A climate-aware scheduling model is constructed, a state space is constructed by fusing climate factors and SOC time sequence characteristics through LSTM, and a PPO algorithm is used to generate charge / discharge instructions of climate-sensitive energy storage in the climate-corrected power interval; The charge / discharge mode switching of a non-climate-sensitive energy storage cluster is triggered by the climate-corrected photovoltaic power threshold, and the dynamic balance of photovoltaic excess power storage and power supply gap compensation is assisted in scheduling; Through a multi-dimensional reward function, climate dynamic adaptation is coupled and instruction tracking optimization is performed, and dynamic closed-loop response is performed.

2. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1, characterized in that, The multi-dimensional climate impact factor model is constructed in the following manner: The temperature factor is used to correct the charge / discharge power and load curve of the energy storage device, the sunshine factor is used to correct the upper limit of the photovoltaic unit output, and the wind speed factor is used to correct the upper limit of the wind power unit output; The weight coefficients of the temperature factor, the sunshine factor and the wind speed factor are obtained by training the historical climate data.

3. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1 or 2, characterized in that, The method uses a time series analysis method to process historical climate data to establish a correlation model of climate parameters and load characteristics of the power distribution network; The initial weights of the temperature factor, the sunshine factor and the wind speed factor are determined through multiple regression analysis; The weight coefficients are dynamically optimized using a neural network, and the climate impact factor reflects the real-time correction requirements of the current environment on the charge / discharge power.

4. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1, characterized in that, When the PPO algorithm generates the charge / discharge instructions: The action space is normalized to the [-1, 1] interval and mapped to the actual charge / discharge power range after climate correction; The sum of the charge / discharge power needs to meet the total power demand of the coordinated scheduling of the power distribution network, and the target consistency is obtained through penalty constraints.

5. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1 or 4, characterized in that, The method adds a scheduling matching degree penalty term in the reinforcement learning reward function to calculate the absolute deviation of the actual total charge / discharge power from the power distribution network scheduling target; Through the back propagation mechanism, the PPO algorithm adjusts the action output, and the total power of the cluster converges to the target power demand range.

6. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1, characterized in that, The time sequence characteristics extracted by the LSTM include: Historical fluctuation trend of climate factors, historical charge / discharge behavior and SOC evolution trajectory of each energy storage unit, and current period climate prediction value; The state space includes the frequency modulation capacity demand and the power quantity technical parameters of the power distribution network.

7. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1 or 6, characterized in that, The method corrects the maximum / minimum capacity limit of the energy storage device based on the climate factor; The SOC optimal working interval is calculated in real time in combination with the frequency modulation capacity demand of the power distribution network; When an extreme temperature environment is detected, the SOC adjustment interval is automatically compressed to prevent overcharging / overdischarging of the battery.

8. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1, characterized in that, The multi-dimensional reward function includes: A climate adaptability reward term to improve the charge / discharge power control accuracy in an extreme temperature environment; An SOC balance penalty term to maintain the state of charge of the cluster within the set adjustment interval; A scheduling matching degree penalty term, which takes the deviation of the actual charge / discharge curve from the target value as the optimization basis.

9. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1 or 8, characterized in that, The method extracts the power quantity technical parameters and regulation service technical parameters as reference values; The deviation amount of the actual charge / discharge power of each energy storage unit from the declared parameters is calculated; The deviation amount is multiplied by the distributed energy storage unit charge / discharge technical parameter as the core coefficient of the penalty term.

10. The distributed energy storage cluster coordinated scheduling method based on climate change according to claim 1, characterized in that, The objective function of the climate-aware scheduling model is: The total power loss of climate correction is minimized, including the start-stop loss of conventional units, the power loss of power generation, and the charging and discharging power loss of energy storage; The marginal node power parameter is obtained by solving the load balance constraint dual multiplier, and the frequency modulation service technical parameter is determined in combination with the branch transfer distribution factor.

Images