Multi-source photovoltaic energy storage collaborative management method and system

Abstract

The invention relates to the technical field of energy management, and discloses a multi-source photovoltaic energy storage collaborative management method and system. The method comprises the steps of obtaining a multi-source photovoltaic system data generation task set; constructing a multi-dimensional data fusion model, and defining a collaborative optimization coordinate system; determining an initial cooperative constraint condition; co-scheduling simulation is executed based on the model to generate a preliminary strategy; and dynamically correcting and generating a multi-source collaborative optimization strategy by using a self-adaptive optimization algorithm. The method also relates to refining processing of each axis in the model, adoption of an improved algorithm optimization strategy, a calculation framework and a fault-tolerant mechanism in simulation, anomaly detection and re-optimization, construction of a scene library for predicting conflicts and the like. The system comprises a data acquisition and classification module, a model construction module, a constraint setting module, a simulation scheduling module and a strategy optimization module. The energy utilization efficiency of the multi-source photovoltaic energy storage system can be effectively improved, stable operation of the system is guaranteed, and equipment loss and cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of energy management, and specifically to a multi-source photovoltaic energy storage collaborative management method and system. Background Art

[0002] With the growing global demand for clean energy, solar energy, as an abundant and renewable energy source, has been increasingly widely used in the energy field. The multi-source photovoltaic energy storage system combines photovoltaic power generation and energy storage technologies and has become a key solution to address the issues of intermittent and unstable energy supply. However, there are still many challenges in the actual operation and management of this field.

[0003] The variability of light intensity brings great uncertainty to photovoltaic power generation. Since solar radiation is affected by factors such as weather and time, the power generation of photovoltaic arrays fluctuates significantly. On sunny days, the power generation is high when the light is sufficient; however, once the clouds block the sunlight, the power generation will drop rapidly. This kind of fluctuation makes it difficult for photovoltaic power to be stably connected to the power grid, causing difficulties in the dispatching and stable operation of the power grid. For example, when the photovoltaic power generation suddenly drops significantly, if the power grid fails to adjust the output of other power sources in time, it may lead to insufficient power supply in some areas, affecting the normal electricity consumption of users.

[0004] There are also many problems in the management of energy storage systems. On the one hand, the charge and discharge efficiency of energy storage devices is not constant and is affected by various factors such as charge and discharge rates and environmental temperatures. In high or low temperature environments, the charge and discharge efficiency of energy storage batteries will be significantly reduced, thereby affecting the overall performance of the energy storage system. On the other hand, the cycle life of energy storage devices is limited, and frequent charge and discharge will accelerate their aging, increasing the operating cost. If the charge and discharge strategies of the energy storage system are not reasonably planned, it may lead to premature damage of the energy storage devices, requiring frequent replacement, which not only wastes resources but also increases the economic burden.

[0005] The complex and variable load demand curve is also a major challenge. The electricity consumption habits of different user groups vary greatly, and the peak electricity consumption periods of commercial users, industrial users, and residential users are different. Industrial users have a large and relatively stable electricity consumption during production; the peak electricity consumption of commercial users usually concentrates during the daytime business hours; residential users have a higher electricity demand in the morning and evening. This complex load change pattern makes it difficult to accurately predict the load demand and difficult to achieve efficient coordination among photovoltaic power generation, energy storage, and the load. If the energy cannot be reasonably dispatched according to the load demand, there may be a situation where the photovoltaic power generation is excessive while the energy storage cannot store it in time, or the energy storage discharge cannot meet the load demand, resulting in energy waste and insufficient supply problems.

[0006] In addition, the dynamic nature of grid dispatching instructions further increases the difficulty of energy management. The grid continuously adjusts the dispatching requirements for distributed energy according to its own operating conditions and the demand for power supply and demand balance. The multi-source photovoltaic energy storage system needs to respond to these instructions in a timely manner. However, due to problems such as communication delays between components within the system and limitations in data processing capabilities, the system has a slow response speed and poor accuracy in responding to grid dispatching instructions. This not only affects the stable operation of the grid but also reduces the economic efficiency and reliability of the multi-source photovoltaic energy storage system. Summary of the Invention

[0007] The purpose of the present invention is to provide a multi-source photovoltaic energy storage collaborative management method and system to solve the problems raised in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A multi-source photovoltaic energy storage collaborative management method, the method includes: Obtain multi-source photovoltaic system data, classify the data according to preset dimensions, and generate a structured energy management task set; the preset dimensions include light intensity, energy storage capacity, load demand curve, and grid dispatching instructions; Construct a multi-dimensional data fusion model, define a collaborative optimization coordinate system based on the classification dimensions of the energy management task set, and the coordinate system includes a time axis, an energy axis, a device axis, and an environment axis; Determine the initial collaborative constraint conditions according to photovoltaic power generation prediction and energy storage charge and discharge efficiency, and the constraint conditions include power balance threshold, energy storage cycle life rule, and grid connection priority; Perform collaborative scheduling simulation on the energy management task based on the multi-dimensional data fusion model, simulate the interaction behavior of photovoltaic, energy storage, and load through a dynamic energy flow simulation engine, and generate a preliminary scheduling strategy; Use an adaptive optimization algorithm to dynamically correct the preliminary scheduling strategy, adjust the photovoltaic output curve and the energy storage charge and discharge timing, and generate a multi-source collaborative optimization strategy.

[0009] Preferably, the construction of the multi-dimensional data fusion model includes: Divide the time axis into optimization intervals synchronized with the light cycle, and associate each interval with a photovoltaic prediction error compensation coefficient; Define the state of charge transfer matrix of the energy storage system based on the energy axis, and integrate the load demand fluctuation characteristics; Embed meteorological prediction data in the environment axis, and dynamically associate the cloud cover rate with the photovoltaic output correction factor.

[0010] Preferably, the adaptive optimization algorithm uses an improved NSGA-II algorithm, including: Encode the photovoltaic output deviation and the energy storage loss rate into a Pareto solution set, and define an objective function to evaluate the matching degree of grid frequency modulation demand and equipment aging cost; Generate a multi-objective optimization solution set through the elite retention strategy and non-dominated sorting, and use the fuzzy decision-making method to screen the optimal cooperation strategy.

[0011] Preferably, the cooperative scheduling simulation includes: Establish a distributed cooperative computing framework, and model the photovoltaic array, energy storage cluster, and microgrid controller as intelligent agent nodes; Use the consistent hashing algorithm to achieve fast synchronization of energy distribution instructions between nodes, and introduce a fault tolerance mechanism to handle communication delays.

[0012] Preferably, the method further includes: Deploy edge computing nodes to collect the working status of inverters and battery health data in real time; Identify photovoltaic output mutation events through time series anomaly detection algorithms, and trigger the policy re-optimization process.

[0013] Preferably, the dynamic correction includes: Construct a deep reinforcement learning model, with the state of charge of the energy storage, the deviation of the grid frequency, and the light intensity as the state space; Train the intelligent agent through the twin-delayed deep deterministic policy gradient algorithm to generate charge and discharge action strategies to maximize the long-term economic benefits.

[0014] Preferably, the definition of the environmental axis further includes: Integrate short-term meteorological forecast data into the photovoltaic output prediction model, and use a wavelet neural network to correct the irradiance time series curve; Dynamically adjust the reserve capacity threshold of the energy storage system and the grid connection power ramp rate according to the rainfall probability.

[0015] Preferably, the method further includes: Construct a typical scenario library based on historical energy scheduling data, and extract high-irradiance - low-load and rainy - high-load combination patterns; Predict future 24-hour energy supply and demand conflicts through a graph convolutional network, and pre-generate energy storage charge and discharge plans.

[0016] Preferably, the method further includes: Adopt a dynamic priority mechanism to manage the photovoltaic curtailment rate and the risk of overcharging of the energy storage, and adjust the weight coefficient in real time according to the grid frequency regulation instruction; Configure independent energy channels for key load devices.

[0017] Preferably, the present invention further includes a multi-source photovoltaic energy storage cooperative management system, and the system includes: Data acquisition and classification module: used to obtain multi-source photovoltaic system data, classify the data according to preset dimensions such as light intensity, energy storage capacity, load demand curve, and grid scheduling instruction, and generate a structured energy management task set; Model construction module: Construct a multi-dimensional data fusion model, and define a collaborative optimization coordinate system including a time axis, an energy axis, a device axis, and an environmental axis based on the classification dimensions of the energy management task set; Constraint setting module: Determine the initial collaborative constraint conditions according to the photovoltaic power generation prediction and the charge and discharge efficiency of the energy storage, and the constraint conditions cover the power balance threshold, the energy storage cycle life rule, and the grid connection priority; Simulation scheduling module: Perform collaborative scheduling simulation on the energy management tasks based on the multi-dimensional data fusion model, simulate the interaction behaviors of the photovoltaic, energy storage, and load with the help of a dynamic energy flow simulation engine, and generate a preliminary scheduling strategy; Strategy optimization module: Dynamically correct the preliminary scheduling strategy by using an adaptive optimization algorithm, adjust the photovoltaic output curve and the energy storage charge and discharge timing, and generate a multi-source collaborative optimization strategy.

[0018] Compared with the prior art, the beneficial effects of the present invention are: In terms of data processing and task planning, multi-source photovoltaic system data is acquired and classified according to preset dimensions such as light intensity, energy storage capacity, load demand curve, and grid dispatching instruction to generate a structured energy management task set. This process enables the system to clearly sort out complex and diverse data, providing a solid data foundation for subsequent accurate energy scheduling. Compared with traditional methods, instead of simply summarizing and processing data, it deeply mines the associations between data. For example, based on the change trend of light intensity, combined with the energy storage capacity and load demand, the charge and discharge strategy of the energy storage is planned in advance to improve the forward-looking and rationality of energy utilization.

[0019] The construction of a multi-dimensional data fusion model is of great significance. The time axis is divided into optimized intervals synchronized with the light cycle, and the photovoltaic prediction error compensation coefficient is associated, significantly improving the accuracy of photovoltaic power generation prediction. By analyzing historical data, the prediction error pattern under different light cycles is determined, and the predicted value is corrected in real time during actual operation. For example, on cloudy days when the light intensity changes frequently, using this coefficient can more accurately predict the photovoltaic power generation, avoiding energy scheduling mistakes caused by prediction errors. Based on the energy axis, the state of charge transfer matrix of the energy storage system is defined and the characteristics of load demand fluctuations are integrated, enabling the energy storage system to respond more intelligently to load changes. When the load demand suddenly increases, the energy storage system quickly adjusts the discharge strategy according to the state of charge transfer matrix to ensure the stability of power supply. Meteorological prediction data is embedded in the environmental axis, dynamically associating the cloud cover rate with the photovoltaic output correction factor, integrating short-term meteorological forecast data into the photovoltaic output prediction model and using a wavelet neural network to correct the irradiance time series curve, and dynamically adjusting the reserve capacity threshold and grid connection power ramp rate of the energy storage system according to the rainfall probability, etc. The impact of environmental factors on energy management is fully considered. Adjusting the operating parameters of energy storage and photovoltaic in advance according to meteorological changes, increasing the energy storage reserve capacity before rainfall to prevent power supply shortages caused by weakened light, ensuring the stable operation of the system under various environmental conditions.

[0020] Initial collaborative constraint conditions are determined, including power balance thresholds, energy storage cycle life rules, grid connection priorities, etc., providing guarantee for the stable operation of the system. The setting of the power balance threshold ensures that during the energy scheduling process, the power generation, load consumption, and energy storage charge and discharge powers are balanced, avoiding system failures caused by power imbalance. Following the energy storage cycle life rules, reasonably arranging the charge and discharge times of the energy storage can extend the service life of the energy storage device and reduce the operating cost. Defining the grid connection priority enables the system to grid-connect and supply power to the grid in the optimal order under different conditions, ensuring the stable operation of the grid.

[0021] Through the establishment of a distributed collaborative computing framework, the photovoltaic array, energy storage cluster, and microgrid controller are modeled as agent nodes in the collaborative scheduling simulation, achieving efficient collaboration among the components of the system. Agent nodes can make autonomous decisions based on their own states and system requirements. For example, the photovoltaic array adjusts the power generation according to the light intensity, and the energy storage cluster charges and discharges according to the state of charge and load demand. The consistent hashing algorithm is used to achieve rapid synchronization of energy distribution instructions among nodes, and a fault tolerance mechanism is introduced to handle communication delays, ensuring the reliable operation of the system in a complex network environment. Even in case of communication failures, the fault tolerance mechanism can ensure the accurate transmission of energy distribution instructions, avoiding energy scheduling chaos caused by communication problems.

[0022] An adaptive optimization algorithm is used to dynamically correct the preliminary scheduling strategy, generate a multi-source collaborative optimization strategy, and further improve the system performance. Taking the improved NSGA-II algorithm as an example, the photovoltaic output deviation and energy storage loss rate are encoded into the Pareto solution set, and the objective function is defined to evaluate the matching degree of the power grid frequency regulation demand and the equipment aging cost. A multi-objective optimization solution set is generated through the elitist retention strategy and non-dominated sorting, and the fuzzy decision-making method is used to screen the optimal collaborative strategy, achieving the balanced optimization among multiple objectives. It not only meets the power grid frequency regulation demand but also reduces the equipment aging cost, improving the economic efficiency and reliability of the system.

[0023] In addition, edge computing nodes are deployed to collect the working status of the inverter and the battery health data in real time. The time series anomaly detection algorithm is used to identify the photovoltaic output mutation events and trigger the strategy re-optimization process, which can timely detect and handle the abnormal situations in the system. When the photovoltaic output mutates, the energy scheduling strategy is quickly adjusted to ensure the stable operation of the system. A typical scenario library is constructed based on the historical energy scheduling data, and the graph convolutional network is used to predict the energy supply-demand conflicts in the next 24 hours and pre-generate the energy storage charge and discharge plan, improving the system's ability to respond to future energy demand changes. The dynamic priority mechanism is adopted to manage the photovoltaic curtailment rate and the energy storage overcharge risk, and the weight coefficient is adjusted in real time according to the power grid frequency regulation instruction. An independent energy channel is configured for the key load equipment, further enhancing the stability and reliability of the system and ensuring the normal power consumption of the key loads. Brief Description of the Drawings

[0024] Figure 1 It is the working principle diagram of the multi-source photovoltaic energy storage collaborative management method described in the present invention; Figure 2 It is the optimization flow chart of the improved NSGA-II algorithm; Figure 3 It is the flow chart of the collaborative scheduling simulation; Figure 4 It is the flow chart of the dynamic correction. Detailed Embodiment

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0026] Please refer to Figures 1 - 4 , a multi-source photovoltaic energy storage collaborative management method proposed by the present invention mainly realizes the collaborative scheduling and optimal management of energy by effectively processing and analyzing the data of the multi-source photovoltaic system, constructing appropriate models and constraint conditions. The specific steps are as follows: There are various types of data in a multi-source photovoltaic system, including light intensity data, which can be obtained in real time by light sensors installed around the photovoltaic site; energy storage capacity data, which can be directly obtained from the monitoring system of the energy storage device and reflects the remaining power of the current energy storage device; load demand curve data, which is predicted using data analysis algorithms based on historical load data and real-time load monitoring data and is used to describe the changing trend of the electricity demand of the load during different time periods; and grid dispatching instruction data, which is issued by the grid management department and includes requirements for power regulation and power transmission of the photovoltaic energy storage system. After obtaining these data, they are classified according to several preset dimensions such as light intensity, energy storage capacity, load demand curve, and grid dispatching instructions, so as to generate a structured energy management task set, providing a clear data basis for subsequent energy management.

[0027] Based on the classification dimensions of the energy management task set, a coordinated optimization coordinate system is defined. Among them, the time axis is used to record the time information in the energy management process, which is the time framework for the entire energy dispatch; the energy axis mainly focuses on the energy changes in the energy storage system and the energy interaction with other devices; the device axis covers various devices in the photovoltaic system, such as photovoltaic arrays, energy storage devices, inverters, etc., facilitating the management and control of different devices; and the environment axis integrates environmental factors related to the operation of the photovoltaic system, such as meteorological data. By constructing such a multi-dimensional data fusion model, the multi-source photovoltaic energy storage system can be comprehensively and systematically described and analyzed.

[0028] According to the photovoltaic power generation prediction, which usually combines historical light data, meteorological forecast data, and the performance parameters of photovoltaic devices, etc., and uses specific prediction algorithms to obtain the predicted power value of photovoltaic power generation for a period of time in the future. At the same time, considering the charge and discharge efficiency of the energy storage, which is affected by factors such as the technical characteristics of the energy storage device and the charge and discharge rate. Combining these factors, the initial coordinated constraint conditions are determined, including the power balance threshold, which is used to ensure that during the energy dispatch process, the power generation power of the system, the power consumption power of the load, and the charge and discharge power of the energy storage device are balanced, avoiding power imbalance and resulting in system instability; the energy storage cycle life rule, which is used to limit the charge and discharge times of the energy storage device to extend its service life and reduce costs; and the grid connection priority, which clarifies the priority order of the photovoltaic system to connect to the grid and supply power under different circumstances to ensure the stable operation of the grid.

[0029] Based on the multi-dimensional data fusion model, a collaborative scheduling simulation of energy management tasks is carried out. With the help of the dynamic energy flow simulation engine, the interaction behaviors among photovoltaic, energy storage and load are simulated. For example, when the light intensity changes, the power generation of the photovoltaic array changes accordingly. At this time, the energy storage system decides whether to charge or discharge according to its own state of charge and the load demand to maintain the stable operation of the system. Through this simulation process, a preliminary scheduling strategy is generated, providing a basis for subsequent optimization.

[0030] An adaptive optimization algorithm is used to dynamically correct the preliminary scheduling strategy. During the correction process, the photovoltaic output curve is adjusted. According to different times, light intensities and system requirements, the power generation output of the photovoltaic array is optimized. At the same time, the charging and discharging time sequence of the energy storage is adjusted, and the charging and discharging times of the energy storage device are reasonably arranged to improve the energy utilization efficiency and the economic benefits of the system. Finally, a multi-source collaborative optimization strategy is generated to achieve the efficient management of the multi-source photovoltaic energy storage system.

[0031] The present invention will be further described below in conjunction with Embodiments 1 to 6: Embodiment 1: When constructing the multi-dimensional data fusion model, more detailed processing is carried out on the time axis, energy axis and environment axis. The time axis is divided into optimization intervals synchronized with the light cycle, and the light cycle can be determined according to the local sunrise and sunset times and historical light data. For example, in a certain area, the light cycle is longer in summer and shorter in winter. By accurately dividing the light cycle, each interval is associated with a photovoltaic prediction error compensation coefficient. This coefficient is obtained through the comparative analysis of historical photovoltaic prediction data and actual power generation data, and is used to correct the errors generated in the photovoltaic prediction process. For example, through statistical analysis, it is found that in the interval from 10:00 to 11:00 in the morning, due to factors such as cloud changes, there is a certain deviation between the predicted photovoltaic power and the actual power. By adjusting the predicted power with the associated error compensation coefficient, the prediction accuracy can be improved.

[0032] Based on the energy axis, the state of charge transfer matrix of the energy storage system is defined. The state of charge (SOC) reflects the proportion of the current remaining power of the energy storage system in its rated capacity. The state of charge transfer matrix of the energy storage system describes the change relationship of the state of charge over time under different charging and discharging conditions. For example, when the energy storage system is charging at a certain charging power, according to its internal chemical reactions and physical characteristics, the state of charge will increase according to a specific law; conversely, during discharge, the state of charge will decrease. At the same time, the load demand fluctuation characteristics are integrated. By analyzing the historical load data, the fluctuation laws and characteristics of the load demand are extracted and incorporated into the state of charge transfer matrix, enabling the energy storage system to better adapt to the changes in the load.

[0033] In the environmental axis, meteorological prediction data is embedded. The meteorological prediction data includes information such as temperature, humidity, cloud cover rate, etc. Among them, the cloud cover rate is dynamically associated with the PV output correction factor. The cloud cover rate affects the light intensity, and thus affects the power generation of the PV array. By establishing an association model between the cloud cover rate and the PV output correction factor, when the cloud cover rate changes, the PV output prediction value can be adjusted in a timely manner. For example, when the cloud cover rate is high, the PV output correction factor will decrease accordingly, thereby reducing the predicted power generation of the PV array.

[0034] In addition, short-term meteorological forecast data is integrated into the PV output prediction model. The short-term meteorological forecast data can be obtained from professional meteorological forecast agencies. The wavelet neural network is used to correct the irradiance time series curve. The wavelet neural network combines the multi-resolution analysis ability of wavelet transform and the self-learning ability of neural network. By learning and training the historical irradiance data and meteorological data, the changing trend of irradiance can be predicted more accurately, and then the PV output prediction value can be corrected. The reserve capacity threshold and grid connection power ramp rate of the energy storage system are dynamically adjusted according to the rainfall probability. When the rainfall probability is high, the power generation of the PV array may be greatly affected. At this time, the reserve capacity threshold of the energy storage system is increased to ensure the normal power supply of the load; at the same time, the grid connection power ramp rate is reduced to avoid the impact on the power grid caused by the sudden drop of PV power.

[0035] Embodiment 2: When optimizing the preliminary scheduling strategy, the improved NSGA-II algorithm is adopted. The PV output deviation and the energy storage loss rate are encoded into the Pareto solution set. The PV output deviation refers to the difference between the actual PV power generation and the predicted PV power generation, which reflects the stability of the PV system power generation; the energy storage loss rate reflects the energy loss of the energy storage device during the charge and discharge process. By encoding these two indicators into the Pareto solution set, trade-offs and optimizations can be made among multiple objectives.

[0036] The objective function is defined to evaluate the grid frequency regulation demand matching degree and the equipment aging cost. The grid frequency regulation demand matching degree is used to measure the contribution of the PV energy storage system to the grid frequency regulation, and the equipment aging cost takes into account the aging loss of the PV equipment and the energy storage equipment during operation. For example, by establishing a mathematical model, the fluctuation of the grid frequency is associated with the power regulation ability of the PV energy storage system to calculate the grid frequency regulation demand matching degree; for the equipment aging cost, a corresponding cost model is established according to factors such as the service life and charge-discharge times of the equipment.

[0037] Generate a multi-objective optimization solution set through the elitist retention strategy and non-dominated sorting. The elitist retention strategy means that during the evolutionary process, the excellent individuals in the current population are directly retained in the next generation to avoid the loss of excellent solutions; non-dominated sorting is to sort the individuals in the population according to the domination relationship, and divide the individuals not dominated by other individuals into the first level, and so on. In this way, multiple non-dominated solution sets are generated, and these solution sets represent different optimization schemes.

[0038] Adopt the fuzzy decision-making method to screen the optimal cooperation strategy. The fuzzy decision-making method is an effective method for dealing with uncertain problems. Among multiple non-dominated solution sets, since the performance of each scheme on different objectives varies, it is difficult to directly determine the optimal solution. Through the fuzzy decision-making method, the importance degrees of each objective are fuzzified, and according to the set fuzzy rules and membership functions, the comprehensive evaluation index of each scheme is calculated, so as to screen out the optimal cooperation strategy. For example, in a certain situation, for the two objectives of the matching degree of power grid frequency regulation demand and equipment aging cost, according to the actual demand, determine their fuzzy weights, and then evaluate the schemes in each non-dominated solution set, and select the scheme with the optimal comprehensive evaluation index as the final cooperation strategy.

[0039] Example 3: In the collaborative scheduling simulation session, building a distributed collaborative computing framework is the core task. First, model the photovoltaic array, energy storage cluster, and microgrid controller as agent nodes. As an agent node, the photovoltaic array has the ability to autonomously adjust the power generation according to the change of light intensity. For example, when the light intensity increases, the photovoltaic array can adjust the working parameters of internal components to improve the power generation efficiency and increase the power generation output; conversely, when the light intensity weakens, the power output is correspondingly reduced.

[0040] The energy storage cluster agent node also plays an important role. It can flexibly decide the charge and discharge operations according to its own state of charge, system energy demand, and grid dispatching instructions. For example, when the photovoltaic power generation is excessive and the load demand is low, the energy storage cluster agent node will start the charging program to store the excess electric energy; while when the photovoltaic power generation is insufficient or the load demand increases, it will perform the discharge operation to supplement energy for the system.

[0041] The microgrid controller agent node acts as a coordinator. It is responsible for collecting information of each agent node, including the power generation of the photovoltaic array, the state of charge of the energy storage cluster, and the real-time demand of the load, etc., and formulating a reasonable energy allocation plan according to this information.

[0042] The consistent hashing algorithm is adopted to achieve fast synchronization of energy distribution instructions among nodes. The working principle of the consistent hashing algorithm is to map the data space onto a circular space according to certain rules. In this system, each intelligent agent node has a corresponding position on this circular space. When the microgrid controller intelligent agent node generates an energy distribution instruction, it will quickly determine which intelligent agent nodes the instruction should be sent to according to the consistent hashing algorithm. For example, when it is necessary to adjust the power generation of the photovoltaic array to meet the load demand, the microgrid controller will calculate the corresponding photovoltaic array intelligent agent node according to the consistent hashing algorithm and send the instruction accurately, ensuring that each node can receive the instruction in time and work collaboratively.

[0043] In the actual operation process, communication delay is an inevitable problem. To solve this problem, a fault tolerance mechanism is introduced. When communication delay is detected, the timeout retransmission mechanism will be started first. For example, if the energy distribution instruction sent by the microgrid controller to the photovoltaic array does not receive an acknowledgment message within the specified time, the system will automatically resend the instruction to ensure that the instruction can be correctly received. At the same time, the system also sets up a backup communication link. When the main communication link has a delay or fails, it will automatically switch to the backup communication link to ensure the stability of data transmission. In addition, the communication delay will be monitored and analyzed, and information such as the time, frequency, and impact range of the delay occurrence will be recorded for subsequent optimization and improvement of the system, avoiding energy distribution errors caused by communication delay and ensuring the stable operation of the system.

[0044] Embodiment 4: In the entire multi-source photovoltaic energy storage collaborative management system, edge computing nodes are distributed at various key positions in the system, such as near the inverter and the energy storage battery pack. These nodes can collect the working status of the inverter and the battery health data in real time.

[0045] For the collection of the working status of the inverter, it mainly covers multiple key parameters. The output power reflects the ability of the inverter to convert direct current into alternating current and output it to the power grid or supply the load; the conversion efficiency reflects the loss situation of the inverter during the energy conversion process, and high conversion efficiency helps to improve energy utilization; the working temperature is an important factor affecting the performance and lifespan of the inverter, and too high a temperature may cause the inverter to malfunction. By real-time monitoring of these parameters, abnormal situations of the inverter can be detected in time. For example, when the output power of the inverter suddenly drops, the conversion efficiency significantly decreases, and the working temperature rises sharply, it indicates that the inverter may have a fault and needs to be repaired or adjusted in time.

[0046] The collection of battery health data is also crucial. The capacity attenuation of the battery is one of the important indicators to measure battery health. As the number of battery charge and discharge cycles increases, its actual capacity will gradually decrease. The change in internal resistance can also reflect the health status of the battery. An increase in internal resistance means an increase in energy loss inside the battery and a decline in performance. By monitoring these data, the battery life can be predicted in advance. For example, when it is found that the rate of battery capacity attenuation accelerates and the internal resistance continues to increase, a battery replacement plan can be arranged in advance to avoid abnormal system operation caused by battery failure.

[0047] Identify photovoltaic output mutation events through time series anomaly detection algorithms. This algorithm builds a model based on historical photovoltaic output data to learn the variation law of photovoltaic output over time under normal conditions. For example, based on historical data under different seasons and weather conditions, determine the typical variation curve of photovoltaic output within a day. When there is a large deviation between the real-time monitored photovoltaic output data and the curve predicted by the model, and this deviation exceeds the preset threshold, the anomaly detection mechanism will be triggered and it will be determined as a photovoltaic output mutation event.

[0048] Once a mutation event is detected, immediately trigger the strategy re-optimization process. In this process, the system will re-collect and analyze the current system state data, including the latest light intensity, energy storage capacity, load demand, and grid dispatching instructions and other information. Based on these new data, combined with the operation objectives and constraints of the system, re-generate the optimization strategy. For example, if the photovoltaic output mutation is caused by cloud cover, the system may adjust the discharge strategy of the energy storage system to increase the discharge power to meet the load demand; at the same time, according to the new light prediction situation, re-plan the subsequent photovoltaic output and energy storage charge and discharge plans to ensure that the system can still operate stably under abnormal conditions.

[0049] Example 5: When dynamically correcting the preliminary scheduling strategy, the deep reinforcement learning model uses the state of charge of the energy storage, the grid frequency deviation, and the light intensity as the state space.

[0050] The state of charge of the energy storage (State of Charge, abbreviated as SOC) is the ratio of the current remaining charge of the energy storage system to its rated capacity. It directly affects the charge and discharge decisions of the energy storage system. For example, when the SOC is low, the energy storage system is more inclined to charge to reserve energy; while when the SOC is high, discharging operations can be considered to meet the load demand or participate in grid frequency regulation.

[0051] Grid frequency deviation is an important indicator to measure the stability of power grid operation. The rated frequency of the power grid is generally 50Hz or 60Hz. When the power generation power and load power in the power grid are unbalanced, the power grid frequency will fluctuate. Excessive grid frequency deviation may affect the normal operation of various equipment in the power grid and even lead to power grid failures. In this model, the grid frequency deviation is incorporated into the state space so that the system can adjust the operation strategies of photovoltaic and energy storage in a timely manner according to the changes in the grid frequency and maintain the stability of the grid frequency.

[0052] Light intensity is a key factor determining the power generation of a photovoltaic array. Under different light intensities, the power generation efficiency and power output of the photovoltaic array are different. By monitoring the light intensity in real time and taking it as part of the state space, the model can more accurately predict the power generation capacity of the photovoltaic and thus formulate a more reasonable scheduling strategy.

[0053] The Twin Delayed Deep Deterministic Policy Gradient algorithm (TD3) is used to train the agent to generate charge and discharge action strategies. The TD3 algorithm effectively reduces the overestimation problem in the algorithm training process by introducing two target networks with delayed updates, improving the stability and convergence speed of the algorithm.

[0054] During the training process, the agent selects appropriate charge and discharge actions according to the current system state, namely the state of charge of the energy storage, the grid frequency deviation and the light intensity. For example, when the grid frequency is low and the state of charge of the energy storage is high, the agent may choose to discharge the energy storage system and inject power into the grid to increase the grid frequency; when the light intensity is strong and the state of charge of the energy storage is low, the agent may choose to give priority to charging the energy storage system.

[0055] The agent adjusts its strategy according to the obtained reward signal by interacting with the environment. The setting of the reward signal is closely centered around the operation goal of the system. For example, if the system can reduce the curtailment rate of photovoltaic power, reduce the energy storage loss and improve the matching degree of the grid frequency regulation demand while meeting the load demand, the agent will receive a higher reward. Through multiple trainings, the agent continuously learns and optimizes its strategy, gradually finds the optimal charge and discharge action strategy, realizes the dynamic correction of the preliminary scheduling strategy, and improves the overall performance and economic benefits of the system.

[0056] Example 6: Construct a typical scenario library based on historical energy dispatch data. The historical energy dispatch data includes information such as light intensity, load demand, energy storage status, and grid dispatch instructions in different time periods. By analyzing and mining these data, combined patterns such as high irradiation - low load and rainy - high load are extracted. For example, in some periods of summer, there may be a situation where the light intensity is very strong but the load demand is low, which is the high irradiation - low load pattern; while in rainy weather, the light intensity is very weak, and at the same time, the load demand may increase due to the increase of indoor electrical equipment, forming the rainy - high load pattern.

[0057] Predict the energy supply - demand conflict in the next 24 hours through a graph convolutional network. The graph convolutional network is a neural network specifically used to process graph - structured data. Each device and node in the energy system is regarded as a node in the graph, and the energy flow and mutual relationship between them are regarded as edges in the graph to construct the graph structure of the energy system. Use the graph convolutional network to learn and analyze the historical energy dispatch data and the current system state to predict the possible energy supply - demand conflict situation in the next 24 hours. For example, it is predicted that in a certain future time period, due to insufficient light and increasing load, there may be a conflict of insufficient energy supply. Once a conflict is predicted, a pre - generated energy storage charge - discharge plan is made. According to different conflict situations, corresponding energy storage charge - discharge strategies are formulated. For example, when the energy supply is insufficient, arrange the energy storage system to discharge in advance to meet the load demand; when the energy supply is excessive, arrange the energy storage system to charge to avoid curtailment of solar power.

[0058] Adopt a dynamic priority mechanism to manage the curtailment of solar power rate and the risk of over - charging of energy storage. Adjust the weight coefficient in real - time according to the grid frequency regulation instruction. The grid frequency regulation instruction will change continuously according to the operation of the grid. When the grid needs to increase the power generation, appropriately increase the priority of the photovoltaic system to reduce the curtailment of solar power rate; when the grid needs to reduce the power generation, adjust the priority of the energy storage system to avoid over - charging of energy storage. For example, by establishing a weight coefficient adjustment model, according to the specific requirements of the grid frequency regulation instruction, calculate the weight coefficients of the curtailment of solar power rate and the risk of over - charging of energy storage, and then adjust the energy dispatch strategy according to these coefficients.

[0059] Configure independent energy channels for critical load devices. Critical load devices refer to devices with high requirements for the stability of power supply, such as life - support devices in hospitals and servers in data centers. Configure independent energy channels for these devices so that when abnormal situations occur in the system, the power supply of critical load devices can be guaranteed preferentially. The independent energy channels can be connected to the energy storage system or backup power supply through separate lines to ensure that critical load devices can still operate normally when the photovoltaic system fails or the grid power outage occurs, improving the reliability and safety of the system.

[0060] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0061] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-source photovoltaic energy storage collaborative management method, characterized in that: include: Acquire multi-source photovoltaic system data, classify the data according to preset dimensions, and generate a structured energy management task set; the preset dimensions include light intensity, energy storage capacity, load demand curve, and grid dispatch instructions; Construct a multi-dimensional data fusion model, and define a collaborative optimization coordinate system based on the classification dimension of the energy management task set, wherein the coordinate system includes a time axis, an energy axis, an equipment axis, and an environment axis; Determine initial coordination constraints based on photovoltaic power generation forecast and energy storage charging and discharging efficiency, wherein the constraints include power balance threshold, energy storage cycle life rule and grid connection priority; Based on the multi-dimensional data fusion model, the collaborative scheduling simulation of energy management tasks is performed, and the interactive behavior of photovoltaic, energy storage and load is simulated through the dynamic energy flow simulation engine to generate a preliminary scheduling strategy; An adaptive optimization algorithm is used to dynamically modify the preliminary scheduling strategy, adjust the photovoltaic output curve and energy storage charging and discharging timing, and generate a multi-source collaborative optimization strategy.

2. The multi-source photovoltaic energy storage collaborative management method according to claim 1, characterized in that: The construction of the multidimensional data fusion model comprises: The time axis is divided into optimization intervals synchronized with the photoperiod, and each interval is associated with a photovoltaic prediction error compensation coefficient; Define the state of charge transfer matrix of the energy storage system based on the energy axis and integrate the load demand fluctuation characteristics; Meteorological forecast data is embedded in the environmental axis to dynamically associate cloud coverage with photovoltaic output correction factors.

3. The multi-source photovoltaic energy storage collaborative management method according to claim 1, characterized in that: The adaptive optimization algorithm adopts an improved NSGA-II algorithm, including: The PV output deviation and energy storage loss rate are encoded as Pareto solution sets, and the objective function is defined to evaluate the matching degree of grid frequency regulation demand and equipment aging cost; The multi-objective optimization solution set is generated through elite retention strategy and non-dominated sorting, and the fuzzy decision method is used to screen the optimal collaborative strategy.

4. The multi-source photovoltaic energy storage collaborative management method according to claim 1, characterized in that: The collaborative scheduling simulation includes: Establish a distributed collaborative computing framework and model the photovoltaic array, energy storage cluster and microgrid controller as intelligent agent nodes; A consistent hashing algorithm is used to achieve fast synchronization of energy allocation instructions between nodes, and a fault-tolerant mechanism is introduced to handle communication delays.

5. The multi-source photovoltaic energy storage collaborative management method according to claim 1, characterized in that: The method further comprises: Deploy edge computing nodes to collect inverter working status and battery health data in real time; The time series anomaly detection algorithm is used to identify sudden changes in photovoltaic output and trigger the strategy re-optimization process.

6. The multi-source photovoltaic energy storage collaborative management method according to claim 1, characterized in that: The dynamic correction includes: Construct a deep reinforcement learning model with energy storage state of charge, grid frequency deviation and light intensity as the state space; The agent is trained through a double-delayed deep deterministic policy gradient algorithm to generate charging and discharging action strategies to maximize long-term economic benefits.

7. The multi-source photovoltaic energy storage collaborative management method according to claim 2, characterized in that: The definition of the environment axis also includes: Integrate short-term weather forecast data into the photovoltaic output prediction model and use wavelet neural network to correct the irradiance time series curve; The reserve capacity threshold and grid-connected power ramp rate of the energy storage system are dynamically adjusted according to the rainfall probability.

8. The multi-source photovoltaic energy storage collaborative management method according to claim 1, characterized in that: The method further comprises: Build a typical scenario library based on historical energy dispatch data to extract high irradiation-low load and cloudy-high load combination modes; The graph convolutional network is used to predict the energy supply and demand conflicts in the next 24 hours, and to pre-generate energy storage charging and discharging plans.

9. The multi-source photovoltaic energy storage collaborative management method according to claim 1, characterized in that: The method further comprises: A dynamic priority mechanism is used to manage the PV abandonment rate and energy storage overcharging risk, and the weight coefficient is adjusted in real time according to the grid frequency regulation instructions; Configure independent energy channels for key load equipment.

10. A multi-source photovoltaic energy storage collaborative management system, characterized in that: include: Data collection and classification module: used to obtain multi-source photovoltaic system data, classify the data according to the preset dimensions of light intensity, energy storage capacity, load demand curve and grid dispatch instructions, and generate a structured energy management task set; Model building module: build a multi-dimensional data fusion model, and define a collaborative optimization coordinate system including the time axis, energy axis, equipment axis and environment axis based on the classification dimension of the energy management task set; Constraint setting module: Determines initial coordination constraints based on photovoltaic power generation forecast and energy storage charging and discharging efficiency. The constraints include power balance threshold, energy storage cycle life rules and grid connection priority; Simulation scheduling module: Based on the multi-dimensional data fusion model, it performs collaborative scheduling simulation on energy management tasks, simulates the interactive behavior of photovoltaics, energy storage and loads with the help of dynamic energy flow simulation engine, and generates preliminary scheduling strategies; Strategy optimization module: Adopts adaptive optimization algorithm to dynamically modify the initial scheduling strategy, adjust the photovoltaic output curve and energy storage charging and discharging timing, and generate a multi-source collaborative optimization strategy.

Images