Intelligent scheduling method and system for distributed photovoltaic energy storage system based on AI technology

Abstract

The present application relates to the technical field of power intelligent scheduling, in particular to a distributed photovoltaic energy storage system intelligent scheduling method and system based on AI technology, which comprises the following steps: S1, collecting multi-modal data of the distributed photovoltaic system; S2, predicting future power generation and power consumption based on the multi-modal data using a long short-term memory network model; S3, formulating a preliminary charging and discharging strategy for the energy storage system based on the prediction result; S4, optimizing the preliminary strategy using a reinforcement learning method; S5, executing the optimized strategy and performing real-time monitoring to detect abnormal conditions. The present application designs an intelligent scheduling method and system by integrating a long short-term memory network model and a reinforcement learning algorithm, which can accurately predict the power generation of the photovoltaic system and the power consumption of the user side, while considering the influence of the aging attenuation model of the photovoltaic panel and external environmental factors (such as altitude, cloud cover, etc.) on the power generation efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent power dispatching technology, and more specifically, to an intelligent dispatching method and system for distributed photovoltaic energy storage systems based on AI technology. Background Technology

[0002] Solar energy is an intermittent renewable energy source. Its power generation is affected by weather conditions (such as sunny days, cloudy days, cloud cover, etc.), seasonal changes, geographical location, and other factors, exhibiting significant uncertainty and volatility. This characteristic makes the power generation of photovoltaic (PV) systems difficult to predict, posing a challenge to the stable operation of the power grid. To smooth out output fluctuations of PV systems and improve energy utilization efficiency, energy storage devices are typically required. However, how to efficiently manage and schedule these energy storage resources to maximize the economic benefits of the energy storage system while ensuring the reliability of power supply is a complex problem. As the service life of PV panels increases, their photoelectric conversion efficiency gradually declines, negatively impacting the long-term performance and economic benefits of the PV system. Therefore, this paper proposes an intelligent scheduling method and system for distributed PV energy storage systems based on AI technology. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent scheduling method and system for distributed photovoltaic energy storage systems based on AI technology, so as to solve the problems of solar energy volatility and uncertainty, low energy utilization, and decreased efficiency of aging photovoltaic panels mentioned in the background art.

[0004] To achieve the above objectives, the present invention aims to provide an intelligent scheduling method for distributed photovoltaic energy storage systems based on AI technology, comprising:

[0005] S1. Collect multimodal data from distributed photovoltaic systems;

[0006] S2. Based on multimodal data, a long short-term memory network model is used to predict future power generation and power consumption. Since the aging of photovoltaic panels affects power generation efficiency, a photovoltaic panel aging and degradation model is introduced in the prediction process.

[0007] S3. Based on the prediction results, formulate a preliminary charging and discharging strategy for the energy storage system;

[0008] S4. Optimize the initial strategy using reinforcement learning methods;

[0009] S5. Execute the optimized strategy and perform real-time monitoring to detect anomalies.

[0010] As a further improvement to this technical solution, in S1, the multimodal data of the distributed photovoltaic system includes the power generation efficiency of the photovoltaic panels, the status of the energy storage equipment, and the grid load.

[0011] As a further improvement to this technical solution, in step S2, predicting future power generation and power consumption based on multimodal data using a long short-term memory network model includes the following steps:

[0012] S2.1 Preprocess the multimodal data;

[0013] S2.2 Extract features related to power generation and power consumption forecasting from multimodal data;

[0014] S2.3 Select Long Short-Term Memory Network as the prediction model, and divide the multimodal dataset into training set, validation set and test set to construct power generation model and power consumption model. Since the aging of photovoltaic panels affects power generation efficiency, a photovoltaic panel aging degradation model is introduced into the power generation model. The aging of photovoltaic panels is affected by the thickness of dust on the photovoltaic panel and the length of cracks in the photovoltaic panel. The parameters of dust thickness and crack length on the photovoltaic panel are introduced into the photovoltaic panel aging degradation model.

[0015] S2.4. Use the power difference function to measure the difference between the model's predicted value and the actual value, and use the gradient descent method to optimize the power difference function. Optimize the power difference function to take into account the influence of altitude.

[0016] S2.5. Train the electricity consumption model and the power consumption model using the training set;

[0017] S2.6. By receiving new input data in real time through the trained power generation and power consumption models, the future power generation and power consumption can be predicted.

[0018] As a further improvement to this technical solution, in S2.3, the power generation model is as follows:

[0019] ;

[0020] Since the aging of photovoltaic panels affects power generation efficiency, a photovoltaic panel aging degradation model is introduced into the power generation model:

[0021] ;

[0022] The aging of photovoltaic panels is affected by the thickness of dust on the photovoltaic panel and the length of cracks on the photovoltaic panel. The parameters of dust thickness and crack length on the photovoltaic panel are introduced into the aging and degradation model of photovoltaic panels:

[0023] ;

[0024] The electricity consumption model is as follows:

[0025] ;

[0026] in, Indicates at a point in time Projected power generation; Indicates at a point in time The power generation feature vector is input to the model; This represents the predicted power generation after incorporating the photovoltaic panel aging and degradation model. Indicates at a point in time Forecasted electricity consumption; Indicates at a point in time The electricity consumption feature vector is input to the model; Indicates a point in time; This represents a function that maps input features and hidden states to predicted power generation. Indicates the previous time point The hidden state; Represents the parameters of the model; Indicates at a point in time Historical power generation; Indicates at a point in time Historical power generation; Indicates the working status of the photovoltaic panel; Indicates at a point in time The degree of aging of the photovoltaic panels; Indicates at a point in time The thickness of dust on the photovoltaic panel; Indicates at a point in time The length of the crack in the photovoltaic panel; This is a function that calculates the overall degree of aging and degradation.

[0027] As a further improvement to this technical solution, in S2.4, the power difference function is:

[0028] ;

[0029] To address the impact of altitude, the power difference function is optimized.

[0030] ;

[0031] in, This represents the difference between the model's predicted value and the actual value. This represents the optimized power difference function value; Indicates the number of samples; Indicates at a point in time Actual power generation or power consumption; Indicates at a point in time Predicted power generation or power consumption; The weighting coefficient representing the influence of altitude; Indicates the influence coefficient of altitude; Indicates the first The altitude of each sample; Indicates the sample index; Indicates a point in time.

[0032] As a further improvement to this technical solution, in step S3, based on the prediction results, a preliminary charging and discharging strategy for the energy storage system is formulated, including the following steps:

[0033] S3.1 Based on the prediction results, identify the peak and off-peak periods of photovoltaic power generation;

[0034] S3.2 Analyze the user electricity demand in the forecast results and identify peak and off-peak electricity consumption periods;

[0035] S3.3 During peak power generation periods, excess electrical energy is stored in energy storage devices. During peak electricity consumption periods, the electrical energy stored in the energy storage devices is used to meet user demand. During off-peak electricity consumption periods, if the energy storage devices are low in power, electricity is purchased from the grid to supplement the energy storage devices.

[0036] S3.4. Based on the specifications and current status of the energy storage device, set the charging rate; based on the specifications and current status of the energy storage device, set the discharging rate; and set the maximum and minimum allowable capacity of the energy storage device.

[0037] S3.5. For the next few hours, predict the specific situation of power supply and demand, and adjust the charging and discharging plan accordingly.

[0038] As a further improvement to this technical solution, in step S4, a reinforcement learning method is used to optimize the initial strategy, including the following steps:

[0039] S4.1 Define the system's state variables and the actions that the energy storage system can take;

[0040] S4.2 Design a reward function that reflects the system objective and minimizes transmission loss. Consider the different transmission losses caused by different transmission distances and optimize the reward function. Further optimize the reward function for the shading of photovoltaic panels caused by obstructions.

[0041] S4.3 Select a reinforcement learning algorithm and set its hyperparameters;

[0042] S4.4. Use reinforcement learning algorithms to iteratively train the initial policy. In each training cycle, the algorithm selects an action based on the current state, executes the action, observes the new state variables and obtains the reward function, and then updates and optimizes the initial policy.

[0043] As a further improvement to this technical solution, in S4.2, the reward function is:

[0044] ;

[0045] Considering the different transmission losses caused by different transmission distances, the reward function is optimized:

[0046] ;

[0047] To address the issue of photovoltaic panel shadows caused by obstructions blocking the sun, the reward function has been further optimized:

[0048] ;

[0049] in, This indicates the system's transmission loss; This indicates the transmission loss after considering different transmission distances; This indicates the optimized transmission loss after considering the obstruction of sunlight by objects. This represents a coefficient that adjusts the proportion of transmission losses in the total reward. Indicates the length of the time period; Indicates a point in time; Indicates a time interval; Indicates time The transmission distance; Indicates time The electrical power transmitted; This represents a coefficient that adjusts the proportion of zero-load in the total reward. Indicates zero-load time; Indicates the maximum possible load time; This represents a coefficient related to transmission distance, used to adjust for the impact of transmission distance on transmission losses; Indicates the total length of the transmission line; This represents a coefficient related to the transmitted power, used to adjust the impact of transmitted power on transmission losses. Indicates the transmitted electrical power; Indicates time Light correction factor due to obstruction.

[0050] As a further improvement to this technical solution, in step S5, the optimized strategy is executed, and real-time monitoring is performed to detect abnormal situations, including the following steps:

[0051] S5.1. The final optimized scheduling strategy is then distributed to the energy storage system and related equipment.

[0052] S5.2 Adjust the charging and discharging rate and status of the energy storage device in real time through the control system;

[0053] S5.3 Continuously collect real-time data from photovoltaic panels and energy storage devices, and use the collected real-time data to assess the current status of the system;

[0054] S5.4. Based on historical data, set the safety range and threshold for the collected real-time data;

[0055] S5.5 Compare the real-time data collected in real time with the set threshold to detect whether there are any abnormalities. Once an abnormality is detected, an alarm mechanism is immediately triggered.

[0056] On the other hand, the present invention provides an intelligent scheduling system for a distributed photovoltaic energy storage system based on AI technology, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the intelligent scheduling method for a distributed photovoltaic energy storage system based on AI technology described above.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] 1. This AI-based intelligent scheduling method and system for distributed photovoltaic energy storage systems integrates a long short-term memory network model and reinforcement learning algorithms. This intelligent scheduling method and system can accurately predict the power generation of the photovoltaic system and the electricity consumption on the user side, while also considering the aging and degradation model of the photovoltaic panels and the impact of external environmental factors (such as altitude and cloud cover) on power generation efficiency. This method and system not only help to rationally plan the charging and discharging strategies of the energy storage system and reduce unnecessary energy waste, but also effectively reduce users' electricity expenses and improve the overall economic efficiency of the photovoltaic energy storage system.

[0059] 2. This AI-based intelligent scheduling method and system for distributed photovoltaic energy storage systems monitors the operating status of energy storage devices in real time and detects potential anomalies by combining preset safety thresholds. This allows for timely identification and handling of problems during system operation, such as overcharging, over-discharging, or equipment failure, thus ensuring the long-term stable operation of the system. Furthermore, optimized charging and discharging strategies can prevent damage to the batteries from frequent charging and discharging operations, extending the lifespan of the energy storage devices and improving the overall safety of the system. Attached Figure Description

[0060] Figure 1 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1:

[0063] Please see Figure 1 As shown, this embodiment provides an intelligent scheduling method for a distributed photovoltaic energy storage system based on AI technology, including the following steps:

[0064] S1. Collect multimodal data from distributed photovoltaic systems;

[0065] In this embodiment, the multimodal data of the distributed photovoltaic system includes the power generation efficiency of the photovoltaic panels, the status of the energy storage devices, and the grid load.

[0066] S2. Predict future power generation and power consumption using a long short-term memory network model based on multimodal data;

[0067] In this embodiment, the prediction of future power generation and power consumption based on multimodal data using a long short-term memory network model includes the following steps:

[0068] S2.1 Preprocess the multimodal data to remove noise and outliers, handle missing values, and ensure data quality. Different cleaning methods are needed for different types of data to transform data of different scales to the same scale to facilitate model training. For example, scale all numerical data to between 0 and 1 or perform Z-score standardization.

[0069] S2.2 Extract features related to power generation and power consumption forecasts from multimodal data. These features include historical power generation and the operating status of photovoltaic panels.

[0070] S2.3 Select Long Short-Term Memory Network as the prediction model, and divide the multimodal dataset into training set, validation set and test set to construct power generation model and power consumption model. Since the aging of photovoltaic panels affects power generation efficiency, a photovoltaic panel aging degradation model is introduced into the power generation model. The aging of photovoltaic panels is affected by the thickness of dust on the photovoltaic panel and the length of cracks in the photovoltaic panel. The parameters of dust thickness and crack length on the photovoltaic panel are introduced into the photovoltaic panel aging degradation model.

[0071] Long Short-Term Memory (LSTM) networks are a type of recurrent neural network used to process and predict data with time-series characteristics. LSTM networks effectively capture long-term dependencies in time-series data. In photovoltaic (PV) energy storage systems, both power generation and consumption exhibit significant time dependence. For example, PV power generation is affected by sunshine hours, typically peaking during the day, especially from noon to afternoon; while electricity consumption is influenced by people's lifestyles, usually peaking in the morning and evening. LSTM networks can effectively model these temporal patterns and trends, thereby improving prediction accuracy. In distributed PV energy storage systems, data sources are diverse, including the power generation efficiency of PV panels, the status of energy storage devices, and grid load. LSTM networks can handle multimodal data, improving the comprehensiveness and accuracy of predictions by fusing different types of input data together for joint modeling. The power generation model is as follows:

[0072] ;

[0073] Furthermore, in the Long Short-Term Memory network model It is a complex nonlinear function composed of multiple gating mechanisms, including an input gate, a forget gate, and an output gate. The function's purpose is to process the current input... and the hidden state of the previous moment Mapped to the current time output ; The function is decomposed into the following steps:

[0074] The input gate controls which values ​​will update the cell state:

[0075] ;

[0076] in, This represents the Sigmoid activation function. and These are the input weight matrix of the input gate and the hidden state weight matrix, respectively. It is a bias term;

[0077] The forgetting gate determines which information to discard:

[0078] ;

[0079] Candidate cell states compute new information to update the cell state:

[0080] ;

[0081] in, It is the hyperbolic tangent activation function, used to restrict the output to between -1 and 1;

[0082] Cell state updates use the results of forgetting and input gates:

[0083] ;

[0084] in, This represents element-wise multiplication;

[0085] The output gate determines which parts of the cell state will be output:

[0086] ;

[0087] Current output Determined by cell state and output gate:

[0088] ;

[0089] The final predicted power generation yes After an additional linear layer or directly equal to ;

[0090] The aging of photovoltaic (PV) panels gradually reduces their power generation efficiency. If this factor is ignored and power generation predictions rely solely on historical data, the predicted results may deviate significantly from the actual situation. Introducing a PV panel aging and degradation model can more accurately reflect the actual power generation capacity of PV panels, thereby improving the accuracy of power generation predictions. Based on more accurate power generation predictions, more reasonable energy storage system charging and discharging strategies can be formulated. For example, when low power generation is predicted, electricity can be purchased from the grid in advance to supplement energy storage devices; when high power generation is predicted, excess energy can be prioritized for storage as a backup. The PV panel aging and degradation model can be adjusted according to different PV panel types and usage environments, exhibiting strong adaptability. Since PV panel aging affects power generation efficiency, the PV panel aging and degradation model is incorporated into the power generation model.

[0091] ;

[0092] Furthermore, Used to predict at a point in time Electricity generation This function not only considers historical power generation and The working status of the photovoltaic panels was also taken into consideration. The degree of aging of photovoltaic panels And the hidden state at the previous time point Define the input feature vector ,in, The function expansion is:

[0093] Input gate : ;

[0094] Forgotten Gate : ;

[0095] Candidate cell state : ;

[0096] Cell state update : ;

[0097] Output gate : ;

[0098] Current hidden state : ;

[0099] Predicted power generation : ,in, It is a weight matrix. These are bias terms; they are a subset of the model's parameters. ;

[0100] Dust covering the surface of photovoltaic (PV) panels blocks some sunlight from reaching them, reducing light absorption, lowering photoelectric conversion efficiency, and accelerating panel aging. Dust particles also reflect and scatter some light, further reducing the effective light intensity reaching the panels. Dust accumulation increases the surface temperature of the PV panels, which in turn reduces photoelectric conversion efficiency. Incorporating a dust thickness parameter into power generation prediction models can more accurately reflect the actual power generation capacity of PV panels; for example, if a large dust thickness is predicted for a certain period, the predicted power generation value can be appropriately lowered. Cracks can damage the physical structure of PV panels, affecting their internal electrical properties. The continuity of the circuit reduces the photoelectric conversion efficiency and may also lead to a decline in the electrical performance of the photovoltaic panel, such as short circuits and open circuits, further affecting the power generation efficiency. Increased crack length will lead to a significant decrease in the power generation efficiency of the photovoltaic panel, and severe cracks may even cause the photovoltaic panel to fail completely. Introducing the crack length parameter into the power generation prediction model can more accurately reflect the actual power generation capacity of the photovoltaic panel. For example, if the predicted crack length is large within a certain period, the predicted power generation value can be appropriately lowered. By simultaneously considering the dust thickness and crack length parameters, the power generation prediction model can more comprehensively reflect the actual condition of the photovoltaic panel and improve the accuracy of the prediction.

[0101] The aging of photovoltaic panels is affected by the thickness of dust on the photovoltaic panel and the length of cracks on the photovoltaic panel. The parameters of dust thickness and crack length on the photovoltaic panel are introduced into the aging and degradation model of photovoltaic panels:

[0102] ;

[0103] The electricity consumption model is as follows:

[0104] ;

[0105] in, Indicates at a point in time Projected power generation; Indicates at a point in time The power generation feature vector is input to the model; This represents the predicted power generation after incorporating the photovoltaic panel aging and degradation model. Indicates at a point in time Forecasted electricity consumption; Indicates at a point in time The electricity consumption feature vector is input to the model; Indicates a point in time; This represents a function that maps input features and hidden states to predicted power generation. Indicates the previous time point The hidden state contains information from all previous time points; The parameters of the model include the weight matrix and the bias vector; Indicates at a point in time Historical power generation; Indicates at a point in time Historical power generation; Indicates the working status of the photovoltaic panel; Indicates at a point in time The degree of aging of the photovoltaic panels; Indicates at a point in time The thickness of dust on the photovoltaic panel; Indicates at a point in time The length of the crack in the photovoltaic panel; A function representing the overall degree of aging degradation; Weights indicating dust thickness; Weights representing crack length; Indicates the weight of time; It is a constant representing the aging rate;

[0106] Furthermore, electricity consumption model The expansion formula is similar to the power generation model, if Including historical electricity consumption Weather conditions (e.g., temperature, humidity, etc.), user behavior patterns Therefore, the input feature vector is represented as: ;

[0107] Input gate : ;

[0108] Forgotten Gate : ;

[0109] Candidate cell state : ;

[0110] Cell state update : ;

[0111] Output gate: ;

[0112] Current hidden state : ;

[0113] Predicted electricity consumption The final predicted electricity consumption is obtained through an additional linear layer. ;

[0114] S2.4. Use the power difference function to measure the difference between the model's predicted value and the actual value, and use the gradient descent method to optimize the power difference function. Optimize the power difference function to take into account the influence of altitude.

[0115] The energy difference function quantifies the difference between the model's predicted and actual values, thereby evaluating the model's predictive performance. This helps determine whether the model accurately captures patterns and trends in the data. Based on more accurate predictions, more reasonable charging and discharging strategies for energy storage systems can be developed. The energy difference function is:

[0116] ;

[0117] Altitude significantly impacts the power generation efficiency of photovoltaic (PV) systems. Environmental factors such as atmospheric pressure, temperature, and humidity at high altitudes affect the photoelectric conversion efficiency of PV panels. By incorporating the influence of altitude into the power difference function, the impact of these environmental factors on power generation and consumption can be more accurately reflected, thereby improving the model's prediction accuracy. Altitudes vary greatly across different regions; the optimized power difference function can better adapt to PV systems in different geographical environments. This facilitates the promotion and application of intelligent dispatching methods in various regions, improving the system's universality and flexibility. For example, in high-altitude areas, the power generation efficiency of PV panels may differ due to the influence of atmospheric pressure and temperature. The optimized model can more accurately predict these changes, thus enabling the development of more reasonable dispatching strategies. The power difference function is optimized to address the impact of altitude.

[0118] ;

[0119] in, This represents the difference between the model's predicted value and the actual value. This represents the optimized power difference function value, which incorporates the influence of altitude. This indicates the number of samples, i.e., the number of data points in the time series. Indicates at a point in time Actual power generation or power consumption; Indicates at a point in time Predicted power generation or power consumption; The weighting coefficients represent the impact of altitude and are used to adjust the importance of altitude in the loss function. The influence coefficient of altitude represents the degree of influence of a unit altitude on the loss function; Indicates the first The altitude of each sample; Indicates the sample index; Indicates a point in time;

[0120] Furthermore, regression analysis was used to establish a model relating altitude to power generation or consumption, and the coefficients of the model were used to determine... and The value of , if the regression model is: , This indicates the amount of electricity generated or consumed. It refers to altitude. It is the intercept. It is the influence coefficient of altitude. It is the error term, obtained through regression analysis. After setting the value, adjust it according to the model's performance and actual needs. The value;

[0121] S2.5. Use the training set to train the electricity consumption model and the power consumption model. During the training process, use the validation set to evaluate the model performance periodically, and adjust the hyperparameters (such as learning rate, batch size, etc.) according to the performance of the validation set to prevent overfitting.

[0122] S2.6. By receiving new input data in real time through the trained power generation model and power consumption model, the future power generation and power consumption are predicted, providing a basis for subsequent energy storage system scheduling. These predicted values ​​will be used as input to formulate the initial charging and discharging strategy of the energy storage system and further optimize energy management. If the expected power generation is greater than the power consumption, it indicates that there is a power surplus; conversely, if the power consumption is greater than the power generation, there may be a power shortage.

[0123] S3. Based on the prediction results, formulate a preliminary charging and discharging strategy for the energy storage system;

[0124] In this embodiment, based on the prediction results, a preliminary charging and discharging strategy for the energy storage system is formulated, including the following steps:

[0125] S3.1 Based on the prediction results, identify the peak and off-peak periods of photovoltaic power generation. Typically, peak periods of photovoltaic power generation occur during the day, especially from noon to afternoon on sunny days; while off-peak periods occur at night or on cloudy or rainy days.

[0126] S3.2 Analyze the user electricity demand in the forecast results and identify peak and off-peak electricity consumption periods. Generally speaking, the peak electricity consumption of household users is at night, while the peak electricity consumption of commercial users may be during the day.

[0127] S3.3 During peak power generation periods, excess electrical energy is stored in energy storage devices (if the energy storage devices are already at full capacity, they are used for high-energy-consuming devices such as hot water heating and electric vehicle charging to reduce the energy consumption of these devices during peak power consumption periods). During peak power consumption periods, the electrical energy in the energy storage devices is used to meet user needs. During off-peak power consumption periods, if the energy storage devices are not powerful enough, electricity is purchased from the grid to supplement the energy storage devices.

[0128] S3.4. Based on the specifications and current status of the energy storage device, set the charging rate to avoid excessively fast charging that could shorten battery life or cause safety issues. Based on the specifications and current status of the energy storage device, set the discharging rate to avoid excessively fast discharging that could degrade battery performance or fail to meet user needs. Set the maximum and minimum allowable capacity of the energy storage device to ensure that the device operates within a safe range. For example, avoid fully charging or fully discharging the battery to extend battery life.

[0129] S3.5 For the next few hours or day, predict the specific situation of power supply and demand, and adjust the charging and discharging plan to adapt to the actual needs. For example, if it is predicted that user demand will suddenly increase during a certain period of time, increase the charging amount of energy storage equipment in advance.

[0130] S4. Optimize the initial strategy using reinforcement learning methods;

[0131] In this embodiment, reinforcement learning is a method for learning optimal behavioral strategies through interaction with the environment. The reward function is a core concept in reinforcement learning, used to evaluate the quality of each action taken by the agent. The design of the reward function determines the system's learning objective. Preliminary strategies are usually generated based on static rules or simple models, which may not fully consider dynamically changing environmental factors and system states. Reinforcement learning allows for dynamic adjustment and optimization of scheduling strategies, making them more adaptable to actual operating conditions and improving overall system performance. Reinforcement learning can automatically adjust scheduling strategies based on the system's real-time state and environmental changes, achieving adaptive scheduling. This helps the system maintain efficient operation under different operating conditions, improving system flexibility and robustness. Reinforcement learning can more accurately predict and manage the energy supply and demand balance, optimizing the charging and discharging strategies of energy storage devices. Optimizing the preliminary strategy using reinforcement learning includes the following steps:

[0132] S4.1 Define the system's state variables, including the current time, the power generation of the photovoltaic panels, the remaining power of the energy storage device, and the grid load. Define the actions that the energy storage system can take, such as charging, discharging, and maintaining the current state.

[0133] S4.2 Design a reward function that reflects the system's objectives to minimize transmission losses. Consider the different transmission losses caused by different transmission distances and optimize the reward function. Further optimize the reward function to address the shading of photovoltaic panels caused by obstructions. The reward function considers immediate rewards (such as giving positive rewards when the system charges during peak power generation and discharges during peak power consumption; giving positive rewards when the system avoids unnecessary charging and discharging operations and reduces equipment losses) and long-term rewards (such as giving positive rewards when the system maintains stable operation for a long period of time and avoids frequent charging and discharging operations).

[0134] The reward function is:

[0135] ;

[0136] Transmission losses are an unavoidable part of energy transmission, especially significant over long distances. By considering transmission distance, energy transmission paths can be optimized to reduce energy loss during transmission. The system can choose paths with lower transmission losses, thereby improving energy transmission efficiency. Transmission losses lead to the loss of some electrical energy during transmission, increasing energy waste. Optimizing transmission paths and strategies can reduce unnecessary transmission losses, thus improving energy utilization efficiency. Transmission losses not only waste energy but also increase system operating costs. Optimizing transmission paths and strategies can reduce transmission losses, thereby lowering the cost of purchasing electricity from the grid. Considering the different transmission losses caused by different transmission distances, the reward function can be optimized.

[0137] ;

[0138] Obstructions blocking sunlight can cause a sudden drop in the power generation of photovoltaic panels. This instantaneous change places higher demands on power generation forecasting. By optimizing the reward function, such instantaneous changes can be predicted and managed more accurately, improving the accuracy of power generation forecasting. The drop in power generation caused by obstructions blocking sunlight will affect the charging and discharging strategies of energy storage systems. An optimized reward function can better cope with this change, ensuring the stable operation of the system under different weather conditions. For example, when obstruction is predicted, electricity can be purchased from the grid in advance to supplement the energy storage devices, or the discharge strategy of the energy storage devices can be adjusted to cope with the drop in power generation. By optimizing the reward function, energy waste caused by obstruction can be reduced. For example, it can prevent over-discharge when power generation drops, which could cause the energy storage devices to be unable to meet subsequent demand.

[0139] To address the issue of photovoltaic panel shadows caused by obstructions blocking the sun, the reward function has been further optimized:

[0140] ;

[0141] ;

[0142] in, This indicates the system's transmission loss; This indicates the transmission loss after considering different transmission distances; This indicates the optimized transmission loss after considering the obstruction of sunlight by objects. This represents a coefficient that adjusts the proportion of transmission losses in the total reward. Indicates the length of the time period; Indicates a point in time; Indicates a time interval; Indicates time The transmission distance; Indicates time The electrical power transmitted; This represents a coefficient that adjusts the proportion of zero-load in the total reward. This indicates zero-load time, which is the time during which the system has no load. This represents the maximum possible load time, which is the total time the system can theoretically operate continuously. This represents a coefficient related to transmission distance, used to adjust for the impact of transmission distance on transmission losses; Indicates the total length of the transmission line; This represents a coefficient related to the transmitted power, used to adjust the impact of transmitted power on transmission losses. This represents the electrical power transmitted (which is a constant or average value). Indicates time Light correction factor due to obstruction; Indicates time Predicted light intensity; Indicates time The percentage of obstruction by the obstruction, with a value range of [0,1], where 0 represents no obstruction and 1 represents complete obstruction;

[0143] S4.3 Select a reinforcement learning algorithm (Q-Learning) and set the algorithm's hyperparameters, such as learning rate, discount factor, and exploration strategy;

[0144] S4.4. Use reinforcement learning algorithms to iteratively train the initial policy. In each training cycle, the algorithm selects an action based on the current state, executes the action, observes the new state variables and obtains the reward function, and then updates and optimizes the initial policy.

[0145] S5. Execute the optimized strategy and perform real-time monitoring to detect anomalies;

[0146] In this embodiment, the optimized strategy is executed, and real-time monitoring is performed to detect abnormal situations, including the following steps:

[0147] S5.1. The final optimized scheduling strategy is sent to the energy storage system and related equipment to ensure that all equipment performs charging and discharging operations in accordance with the optimized strategy;

[0148] S5.2. Adjust the charging and discharging rate and status of the energy storage equipment in real time through the control system to ensure that the system operates in the optimal state;

[0149] S5.3 Continuously collect real-time data from photovoltaic panels and energy storage devices. Real-time data includes the power generation of photovoltaic panels, the remaining power of energy storage devices, and the grid load. Use the collected real-time data to assess the current status of the system, such as determining whether the energy storage devices are overloaded or whether the photovoltaic panels are working properly.

[0150] S5.4. Based on historical data, set the safety range and threshold for the collected real-time data;

[0151] The photovoltaic panel power generation is set within a safety range based on the average and standard deviation of historical data. For example, assuming the average power generation is 1000 kWh and the standard deviation is 100 kWh, the safety range can be set as follows: Set alarm thresholds to trigger an alarm when power generation exceeds the normal range. For example, the lower threshold is 750 kWh, and the upper threshold is 1250 kWh. If the average remaining power of historical data is 50% and the standard deviation is 10%, then the safe range for the remaining power of the energy storage device can be set as follows: The lower limit for remaining power is 20%, and the upper limit is 80%. If the average grid load in historical data is 500 kWh and the standard deviation is 50 kWh, then the safety range can be set as follows: The lower limit threshold for grid load is 350 kWh, and the upper limit threshold is 650 kWh.

[0152] S5.5 Compare the real-time data collected in real time with the set threshold to detect whether there are any abnormalities. Once an abnormality is detected, an alarm mechanism is immediately triggered.

[0153] Example 2:

[0154] This embodiment provides an AI-based intelligent scheduling system for distributed photovoltaic energy storage systems, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the aforementioned AI-based intelligent scheduling method for distributed photovoltaic energy storage systems.

[0155] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. An intelligent scheduling method for distributed photovoltaic energy storage systems based on AI technology, characterized in that, Includes the following steps: S1. Collect multimodal data from distributed photovoltaic systems; S2. Based on multimodal data, a long short-term memory network model is used to predict future power generation and power consumption, and a photovoltaic panel aging and degradation model is introduced in the prediction process. The method of predicting future power generation and consumption using a long short-term memory network model based on multimodal data includes the following steps: S2.1 Preprocess the multimodal data; S2.2 Extract features related to power generation and power consumption forecasting from multimodal data; S2.3 Select Long Short-Term Memory Network as the prediction model, and divide the multimodal dataset into training set, validation set and test set to construct power generation model and power consumption model. Since the aging of photovoltaic panels affects power generation efficiency, a photovoltaic panel aging degradation model is introduced into the power generation model. The aging of photovoltaic panels is affected by the thickness of dust on the photovoltaic panel and the length of cracks in the photovoltaic panel. The parameters of dust thickness and crack length on the photovoltaic panel are introduced into the photovoltaic panel aging degradation model. S2.

4. Use the power difference function to measure the difference between the model's predicted value and the actual value, and use the gradient descent method to optimize the power difference function. Optimize the power difference function to take into account the influence of altitude. S2.

5. Train the electricity consumption model and the power consumption model using the training set; S2.

6. By receiving new input data in real time through the trained power generation model and power consumption model, predict future power generation and power consumption; S3. Based on the prediction results, formulate a preliminary charging and discharging strategy for the energy storage system; S4. Optimize the initial strategy using reinforcement learning methods; S5. Execute the optimized strategy and perform real-time monitoring to detect anomalies.

2. The intelligent scheduling method for distributed photovoltaic energy storage systems based on AI technology according to claim 1, characterized in that: In S1, the multimodal data of the distributed photovoltaic system includes the power generation efficiency of the photovoltaic panels, the status of the energy storage devices, and the grid load.

3. The intelligent scheduling method for distributed photovoltaic energy storage systems based on AI technology according to claim 1, characterized in that: In S2.3, the power generation model is as follows: ； Since the aging of photovoltaic panels affects power generation efficiency, a photovoltaic panel aging degradation model is introduced into the power generation model: ； The aging of photovoltaic panels is affected by the thickness of dust on the photovoltaic panel and the length of cracks on the photovoltaic panel. The parameters of dust thickness and crack length on the photovoltaic panel are introduced into the aging and degradation model of photovoltaic panels: ； The electricity consumption model is as follows: ； in, Indicates at a point in time Projected power generation; Indicates at a point in time The power generation feature vector is input to the model; This represents the predicted power generation after incorporating the photovoltaic panel aging and degradation model. Indicates at a point in time Forecasted electricity consumption; Indicates at a point in time The electricity consumption feature vector is input to the model; Indicates a point in time; This represents a function that maps input features and hidden states to predicted power generation. Indicates the previous time point The hidden state; Represents the parameters of the model; Indicates at a point in time Historical power generation; Indicates at a point in time Historical power generation; Indicates the working status of the photovoltaic panel; Indicates at a point in time The degree of aging of the photovoltaic panels; Indicates at a point in time The thickness of dust on the photovoltaic panel; Indicates at a point in time The length of the crack in the photovoltaic panel; This is a function that calculates the overall degree of aging and degradation.

4. The intelligent scheduling method for a distributed photovoltaic energy storage system based on AI technology according to claim 1, characterized in that: In S2.4, the power difference function is: ； To address the impact of altitude, the power difference function is optimized. ； in, This represents the difference between the model's predicted value and the actual value. This represents the optimized power difference function value; Indicates the number of samples; Indicates at a point in time Actual power generation or power consumption; Indicates at a point in time Predicted power generation or power consumption; The weighting coefficient representing the influence of altitude; Indicates the influence coefficient of altitude; Indicates the first The altitude of each sample; Indicates the sample index; Indicates a point in time.

5. The intelligent scheduling method for distributed photovoltaic energy storage systems based on AI technology according to claim 1, characterized in that: In step S3, based on the prediction results, a preliminary charging and discharging strategy for the energy storage system is formulated, including the following steps: S3.1 Based on the prediction results, identify the peak and off-peak periods of photovoltaic power generation; S3.2 Analyze the user electricity demand in the forecast results and identify peak and off-peak electricity consumption periods; S3.3 During peak power generation periods, excess electrical energy is stored in energy storage devices. During peak electricity consumption periods, the electrical energy stored in the energy storage devices is used to meet user demand. During off-peak electricity consumption periods, if the energy storage devices are low in power, electricity is purchased from the grid to supplement the energy storage devices. S3.

4. Based on the specifications and current status of the energy storage device, set the charging rate; based on the specifications and current status of the energy storage device, set the discharging rate; and set the maximum and minimum allowable capacity of the energy storage device. S3.

5. For the next few hours, predict the specific situation of power supply and demand, and adjust the charging and discharging plan accordingly.

6. The intelligent scheduling method for a distributed photovoltaic energy storage system based on AI technology according to claim 1, characterized in that: In step S4, reinforcement learning is used to optimize the initial strategy, including the following steps: S4.1 Define the system's state variables and the actions that the energy storage system can take; S4.2 Design a reward function that reflects the system objective and minimizes transmission loss. Consider the different transmission losses caused by different transmission distances and optimize the reward function. Further optimize the reward function for the shading of photovoltaic panels caused by obstructions. S4.3 Select a reinforcement learning algorithm and set its hyperparameters; S4.

4. Use reinforcement learning algorithms to iteratively train the initial policy. In each training cycle, the algorithm selects an action based on the current state, executes the action, observes the new state variables and obtains the reward function, and then updates and optimizes the initial policy.

7. The intelligent scheduling method for a distributed photovoltaic energy storage system based on AI technology according to claim 6, characterized in that: In S4.2, the reward function is: ； Considering the different transmission losses caused by different transmission distances, the reward function is optimized: ； To address the issue of photovoltaic panel shadows caused by obstructions blocking the sun, the reward function has been further optimized: ； in, This indicates the system's transmission loss; This indicates the transmission loss after considering different transmission distances; This indicates the optimized transmission loss after considering the obstruction of sunlight by objects. This represents a coefficient that adjusts the proportion of transmission losses in the total reward. Indicates the length of the time period; Indicates a point in time; Indicates a time interval; Indicates time The transmission distance; Indicates time The electrical power transmitted; This represents a coefficient that adjusts the proportion of zero-load in the total reward. Indicates zero-load time; Indicates the maximum possible load time; This represents a coefficient related to transmission distance, used to adjust for the impact of transmission distance on transmission losses; Indicates the total length of the transmission line; This represents a coefficient related to the transmitted power, used to adjust the impact of transmitted power on transmission losses. Indicates the transmitted electrical power; Indicates time Light correction factor due to obstruction.

8. The intelligent scheduling method for a distributed photovoltaic energy storage system based on AI technology according to claim 1, characterized in that: In step S5, the optimized strategy is executed, and real-time monitoring is performed to detect abnormal situations, including the following steps: S5.

1. The final optimized scheduling strategy is then distributed to the energy storage system and related equipment. S5.2 Adjust the charging and discharging rate and status of the energy storage device in real time through the control system; S5.3 Continuously collect real-time data from photovoltaic panels and energy storage devices, and use the collected real-time data to assess the current status of the system; S5.

4. Based on historical data, set the safety range and threshold for the collected real-time data; S5.5 Compare the real-time data collected in real time with the set threshold to detect whether there are any abnormalities. Once an abnormality is detected, an alarm mechanism is immediately triggered.

9. An intelligent scheduling system for a distributed photovoltaic energy storage system based on AI technology, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the intelligent scheduling method for distributed photovoltaic energy storage systems based on AI technology as described in any one of claims 1-8.

Images