Optical storage charging multi-scene adaptive control and load adjustment method

Abstract

The invention discloses a light storage charging multi-scene adaptive control and load adjustment method, and relates to the technical field of electric vehicle charging, and the method comprises the steps: recognizing a plurality of scenes, such as peak and valley electricity prices and sufficient illumination, and collecting photovoltaic, energy storage, power grid and vehicle related parameters at different frequencies; preprocessing and fusing multi-source data, and extracting key features to generate a standardized matrix; constructing a hybrid intelligent model, predicting a related trend and learning an optimal control strategy; generating a charging power distribution and mode selection scheme according to the scene; energy storage and the charging pile are linked to adjust the load, and power balance is maintained; monitoring the operation state, and feeding back an optimization strategy when the operation state is abnormal; and continuously updating the data and the model to adapt to scene change. According to the invention, multi-scene accurate adaptation and multi-component cooperation are realized, the photovoltaic absorption capability and the charging service quality are improved, the operation cost and the equipment loss are reduced, the requirements of different scales of optical storage and charging systems are adapted, and the power source, network, load and storage integrated operation is assisted.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle charging technology, and in particular to a method for adaptive control and load regulation of photovoltaic-storage charging in multiple scenarios. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the number of electric vehicles continues to grow, and the demand for charging infrastructure is increasing daily. Integrated photovoltaic-storage-charging systems, as a new solution integrating photovoltaic power generation, energy storage, and vehicle charging, have become an important development direction for charging infrastructure due to their advantages of being clean, low-carbon, and capable of peak shaving and valley filling. However, existing photovoltaic-storage-charging systems mostly adopt fixed control strategies, lacking the ability to dynamically adapt to complex scenarios. In actual operation, charging scenarios vary significantly due to factors such as electricity pricing policies, sunlight conditions, grid operation status, and user demand. Peak-valley pricing scenarios require cost optimization, abundant sunlight scenarios require maximizing photovoltaic absorption, grid load scenarios require responding to grid regulation, and emergency charging scenarios require ensuring rapid energy replenishment. However, existing systems struggle to develop precise control strategies for different scenarios, often resulting in insufficient photovoltaic absorption, excessive grid load fluctuations, high charging costs, or delayed response to user demand.

[0003] The limitations of data processing and decision-making models further restrict the improvement of system performance. The operation of photovoltaic-storage-charging systems involves multi-dimensional data from photovoltaics, energy storage, the power grid, and vehicles. Existing technologies mostly use fixed frequencies for data acquisition, and there is no difference in the sampling of core and non-core parameters, resulting in excessive data transmission and storage pressure. At the same time, there is a lack of effective data filtering and fusion mechanisms, and high-frequency noise from photovoltaic output fluctuations and grid voltage and frequency fluctuations directly affects the accuracy of decision-making. The application of AI models is mostly limited to single prediction or control functions and is mainly based on offline training, which cannot adapt to changes in scenarios and equipment status evolution in real time. The model decision accuracy gradually decreases over time, making it difficult to maintain optimal control performance in the long term. In addition, load regulation is mostly concentrated on local optimization within a single station, lacking deep collaboration between energy storage systems, charging piles, and the power grid. When photovoltaic output is excessive or insufficient, it cannot be flexibly adjusted through energy storage. During peak charging loads, it is easy to cause local grid overload, affecting the safe and stable operation of the power grid.

[0004] The contradiction between the diversity of user charging needs and the constraints of grid operation is becoming increasingly prominent. Different users have varying charging urgency and vehicle battery status. Existing systems lack a scientific charging priority assessment mechanism, leading to disorderly competition when charging resources are scarce, preventing users with urgent needs from receiving priority service; conversely, when charging resources are abundant, it is difficult to achieve balanced allocation, resulting in low charging efficiency. In multi-site deployment scenarios, each photovoltaic-storage-charging station operates independently with fragmented data, lacking regional-level global optimization and scheduling. This easily leads to situations where some stations are saturated while others are idle, causing uneven photovoltaic consumption and grid load imbalance within the region. Simultaneously, charging strategies do not fully consider the impact of battery characteristics and environmental factors. Inappropriate charging parameter settings often result in decreased charging efficiency, excessive battery temperature rise, shortened battery life, and negatively impact user charging experience and equipment reliability. These problems severely restrict the promotion and application of photovoltaic-storage-charging systems and the realization of their comprehensive benefits, urgently requiring a technical solution that can adapt to multiple scenarios and achieve precise control and coordinated regulation. Summary of the Invention

[0005] The present invention proposes a multi-scenario adaptive control and load regulation method for photovoltaic energy storage and charging to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a multi-scenario adaptive control and load regulation method for photovoltaic energy storage charging, comprising the following steps: Scene recognition and multi-source parameter acquisition: Based on electricity price, solar irradiance, grid dispatch instructions, user charging request priority, and battery status data, real-time identification of peak-valley electricity price, abundant solar irradiance, grid load shortage, emergency charging, and regular charging scenarios; simultaneous acquisition of multi-dimensional operating parameters; Multi-source data fusion processing: The original data is preprocessed, outliers are removed using the 3σ criterion, and missing data is filled using linear interpolation; key features are extracted through feature engineering, and a weighted fusion algorithm is used to integrate and generate a standardized data matrix; AI Adaptive Decision Model Construction: Construct a hybrid intelligent model that integrates time series prediction and reinforcement learning. The LSTM network completes short-term prediction of photovoltaic output, grid load and charging demand for the next 1-4 hours. The DQN algorithm uses multi-objective dynamic learning to learn the optimal control strategy under different scenarios. Dynamic generation of charging strategies: Based on AI model prediction results and scenario types, an adaptive charging control strategy including charging power allocation and charging mode selection is generated. Load coordination regulation: Combining charging strategy with grid operating status, the load coordination regulation mechanism is activated to suppress fluctuations by controlling the charging and discharging levels of the energy storage system; Real-time monitoring and feedback optimization: Real-time monitoring of various operating parameters and adjustment effects, comparison with preset thresholds, and triggering online adjustment of AI model parameters and iterative optimization of strategies when data is abnormal or the effect does not meet expectations; Scene adaptive iteration: Continuously collect new scene data and policy execution effect data, regularly update the training sample set, and start model retraining.

[0007] Furthermore, it also includes a dynamic optimization process for scene adaptation weights, which adjusts the weights of each feature in real time through multi-factor coupling analysis, and the optimization formula is as follows: ,in For the scene The feature weight matrix, for The weighting adjustment coefficient at time step, For the weight update cycle, The total number of feature dimensions. for Time of the first Importance coefficient of dimensional features for Time of the first Dimensional features and scenarios The fitness function.

[0008] Furthermore, it also includes a dynamic evaluation mechanism for charging demand priority. Based on the electric vehicle's battery SOC status, charging reservation time, vehicle type, and user level, a priority evaluation system is constructed. The charging priority of each vehicle is quantitatively evaluated through the analytic hierarchy process. In scenarios where charging resources are scarce, charging power and charging time are allocated according to priority. In scenarios where charging resources are abundant, charging resources are allocated evenly.

[0009] Furthermore, an adaptive data filtering algorithm is adopted in the multi-source data fusion process. By dynamically adjusting the filtering parameters to adapt to the data fluctuation characteristics under different scenarios, high-frequency noise in photovoltaic output and grid load data is effectively filtered out. A data evaluation mechanism is introduced, and a confidence score method is used to quantitatively score the fused data. When the score is lower than the set standard, the data re-acquisition process is triggered.

[0010] Furthermore, a distributed load balancing algorithm is introduced into the load coordination and adjustment process, using power allocation formulas. To achieve power coordination, in which for The total power balance value of the system at any given time. The number of charging stations. for Time of the first The charging power of each charging station for The charging and discharging power of the energy storage system at all times. for The power exchanged with the power grid at any time is taken as a positive value when absorbing power from the grid and a negative value when feeding power back to the grid.

[0011] Furthermore, the dynamic generation process of the charging strategy includes a charging efficiency optimization module, which dynamically adjusts the matching relationship between charging voltage, current and charging time based on the characteristics of electric vehicle batteries and environmental parameters; and introduces a charging temperature rise control mechanism to monitor the battery charging temperature rise rate in real time.

[0012] Furthermore, the reinforcement learning process of the AI ​​adaptive decision-making model adopts a dual reward mechanism, setting immediate rewards and long-term rewards respectively. The immediate reward is calculated based on the photovoltaic absorption rate and load regulation effect in the current scenario, while the long-term reward is calculated based on the operating cost and service satisfaction over a period of time. The total reward value is obtained by weighted summation, guiding the model to achieve a balance between short-term optimization and long-term benefits.

[0013] Furthermore, it also includes a power grid interaction optimization mechanism, which establishes a real-time communication link with the power grid dispatching system to obtain power grid dispatching instructions and operating constraints, and integrates power grid instructions into the optimization objectives of the AI ​​model; when the power grid needs to shave peaks and fill valleys, it actively adjusts the charging load and energy storage charging and discharging status; at the same time, it feeds back charging load forecast information and adjustable load capacity to the power grid.

[0014] Furthermore, in emergency charging scenarios, a rapid response mechanism is activated to prioritize cutting off unnecessary auxiliary loads, maximize the allocation of charging resources to emergency charging vehicles according to emergency priority, and optimize the discharge power and discharge duration of the energy storage system. After the emergency scenario ends, it automatically switches to the normal control strategy and restores normal load allocation in a gradient manner.

[0015] Furthermore, it also includes a multi-station coordinated regulation mechanism. When multiple photovoltaic, energy storage, and charging stations are located in the same power grid area, a high-speed communication link is established between the stations to share data on photovoltaic output, energy storage status, charging demand, and load regulation of each station.

[0016] Compared with existing technologies, the beneficial effects of this invention are: This invention comprehensively improves the operational performance and overall benefits of photovoltaic-storage-charging systems by constructing a multi-scenario adaptive control and load-coordinated adjustment system, demonstrating significant technical advantages. In the scene identification and multi-source parameter acquisition stages, scene types are accurately classified based on multi-dimensional data. Differentiated sampling frequencies are used for core and non-core parameters, ensuring both real-time and comprehensive data while reducing system operational pressure, providing high-quality data support for subsequent decision-making. The multi-source data fusion processing incorporates various algorithms to remove outliers, fill in missing data, and filter high-frequency noise, improving data reliability and continuity. This enables AI models to accurately capture core features and avoid decision-making biases caused by false data.

[0017] The AI ​​adaptive decision-making model integrates time-series forecasting and reinforcement learning to achieve accurate predictions of photovoltaic power output, grid load, and charging demand. Simultaneously, it dynamically learns optimal strategies guided by multi-objective optimization, with a dual-reward mechanism guiding the model to balance short-term performance and long-term benefits, avoiding local optima and significantly improving the model's scenario adaptability and decision-making scientific rigor. The dynamic charging strategy generation stage develops personalized solutions for different scenarios, combining charging efficiency optimization and temperature rise control mechanisms to ensure charging speed and service quality while reducing charging losses and protecting battery life, achieving a synergy between charging demand and equipment safety.

[0018] The load coordination and regulation mechanism, through precise power allocation between energy storage systems and charging piles, smooths out load fluctuations in photovoltaic and grid systems, maintains dynamic power balance, strengthens deep collaboration between photovoltaic, energy storage, charging, and grid systems, and improves the safe and stable operation of the grid. Real-time monitoring and feedback optimization enable full-process tracking of operational status, promptly triggering strategy iterations and parameter adjustments to ensure the system continuously outputs optimal control effects. The multi-station coordinated regulation and grid interaction optimization mechanism breaks down the limitations of data fragmentation and local optimization, achieving global resource allocation and integrated operation of power generation, grid, load, and energy storage within the region, thereby improving the regional photovoltaic absorption rate and grid load balance.

[0019] The charging demand priority assessment mechanism ensures the fairness and efficiency of charging services, prioritizing users with urgent needs and achieving balanced allocation when resources are sufficient. Overall, this invention achieves precise adaptation to multiple scenarios, deep collaboration among multiple components, and optimal balance across multiple objectives, significantly improving photovoltaic absorption capacity and charging service quality, optimizing grid load distribution, reducing operating costs and equipment losses, extending battery and equipment lifespan, and adapting to the application needs of photovoltaic-storage-charging systems of different scales. This provides strong support for the promotion of integrated photovoltaic-storage-charging technology and the development of the new energy vehicle industry. Attached Figure Description

[0020] Figure 1 This is a schematic block diagram of the multi-scenario adaptive control and load regulation method for photovoltaic energy storage and charging proposed in this invention; Figure 2 A bar chart comparing core performance indicators before and after implementation; Figure 3 Pie chart showing the percentage of charging efficiency across various scenarios; Figure 4 This is a scatter plot of regional load balance under multi-station coordination. Detailed Implementation

[0021] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0023] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0024] Reference Figures 1 to 4 A method for adaptive control and load regulation of photovoltaic energy storage charging in multiple scenarios includes the following steps: Scene recognition and multi-source parameter acquisition: Based on electricity price data, solar irradiance data, grid dispatch instructions, user charging request priority, and battery status data, the system identifies charging scenario types in real time, covering peak-valley electricity price scenarios, abundant solar irradiance scenarios, grid load stress scenarios, emergency charging scenarios, and regular charging scenarios. Simultaneously, it collects multi-dimensional operating parameters, including photovoltaic module output power and conversion efficiency, remaining energy storage system capacity, charging and discharging efficiency and health status, real-time grid load, voltage frequency and transmission capacity, electric vehicle charging power demand, battery SOC status, battery type and charging cutoff voltage, ambient temperature, and solar irradiance. The sampling interval for core parameters is set to the second level, while the sampling interval for non-core parameters is dynamically adjusted to the minute level according to the scenario, ensuring that data acquisition meets real-time decision-making needs while reducing data transmission and storage pressure. Multi-source data fusion processing: The collected raw data is preprocessed, outliers are removed using the 3σ criterion, and missing data is filled in using linear interpolation; key features such as photovoltaic power output fluctuation coefficient, grid load change rate, and vehicle charging demand urgency are extracted using feature engineering, and multi-dimensional features are integrated using a weighted fusion algorithm to generate a standardized data matrix, providing high-quality input for subsequent decision-making models; AI Adaptive Decision Model Construction: A hybrid intelligent model integrating time series forecasting and reinforcement learning is constructed. The time series forecasting model adopts an LSTM network, which is trained based on historical data to achieve short-term forecasts of photovoltaic output, grid load and charging demand in the next 1-4 hours. The reinforcement learning model adopts the DQN algorithm, with the optimization objectives of photovoltaic absorption rate, grid load balance, charging service satisfaction and minimization of operating costs, and dynamically learns the optimal control strategy under different scenarios. Dynamic charging strategy generation: Based on AI model predictions and scenario types, an adaptive charging control strategy is generated. The strategy includes electric vehicle charging power allocation schemes and charging mode selection schemes. Charging modes cover constant current charging, constant voltage charging, constant power charging, and segmented charging. In peak-valley electricity price scenarios, charging time periods are allocated according to electricity price fluctuations, with charging power increased during off-peak hours and decreased during peak hours. In scenarios with abundant sunshine, charging power is adjusted in real time according to photovoltaic output to maximize the absorption of photovoltaic power. In scenarios with tight grid load, the total charging power is dynamically reduced according to grid dispatch requirements. Load Coordination and Regulation: Combining charging strategies with grid operating status, a load coordination and regulation mechanism is activated to smooth out fluctuations in photovoltaic output and grid load by controlling the charging and discharging status of the energy storage system; when photovoltaic output is excessive, the energy storage system charges and stores electrical energy at its rated power; when photovoltaic output is insufficient or grid load is at its peak, the energy storage system discharges to supplement electrical energy as needed; the charging power of each charging pile is dynamically adjusted and controlled within the rated range to achieve balanced regional load distribution and coordinated operation of photovoltaic, energy storage, charging loads and the grid; Real-time monitoring and feedback optimization: Real-time monitoring of photovoltaic output, energy storage status, grid parameters, vehicle charging status and load regulation effect; comparison of monitoring data with preset thresholds; when data exceeds the normal range or regulation effect does not meet expectations, a feedback signal is generated and input into the AI ​​adaptive decision-making model to trigger online adjustment of model parameters and iterative optimization of strategy. Scenario-adaptive iteration: Continuously collect new scenario data and control strategy execution effect data, regularly update the AI ​​model training sample set, start the model retraining process, optimize the model's scenario adaptability and decision accuracy, so that the system can output the optimal control and load adjustment scheme in various scenarios.

[0025] This invention also includes a dynamic optimization process for scene adaptation weights, which adjusts the weights of each feature in real time through multi-factor coupling analysis. The optimization formula is as follows: ,in For the scene The feature weight matrix, for The weighting adjustment coefficient at time step, For the weight update cycle, The total number of feature dimensions. for Time of the first Importance coefficient of dimensional features for Time of the first Dimensional features and scenarios The adaptation function enhances the AI ​​decision-making model's ability to capture the core features of the scene, enabling the control strategy to accurately match the scene requirements.

[0026] This invention also includes a dynamic evaluation mechanism for charging demand priority. Based on the electric vehicle's battery SOC status, charging reservation time, vehicle type, and user level, a priority evaluation system is constructed. The charging priority of each vehicle is quantitatively evaluated using the analytic hierarchy process. In scenarios where charging resources are scarce, charging power and charging time are allocated according to priority, with higher-priority vehicles receiving higher charging power quotas. In scenarios where charging resources are abundant, charging resources are allocated in a balanced manner to achieve a balance between charging service efficiency and user experience.

[0027] In this invention, an adaptive data filtering algorithm is used in the multi-source data fusion processing. By dynamically adjusting the filtering parameters to adapt to the data fluctuation characteristics under different scenarios, high-frequency noise in photovoltaic power output and grid load data is effectively filtered out, core change trends are preserved, and the temporal continuity and numerical accuracy of the data are optimized. A data reliability assessment mechanism is introduced, and a confidence score method is used to quantify the reliability of the fused data. When the score is lower than the set standard, the data re-acquisition process is triggered, providing high-quality input data support for the AI ​​decision model.

[0028] In this invention, a distributed load balancing algorithm is introduced into the load coordination and regulation process, using a power allocation formula. To achieve power coordination, in which for The total power balance value of the system at any given time. The number of charging stations. for Time of the first The charging power of each charging station for The charging and discharging power of the energy storage system is recorded at all times, with positive values ​​taken during charging and negative values ​​taken during discharging. for The power exchange with the grid is constantly monitored, with positive values ​​taken when absorbing power from the grid and negative values ​​taken when feeding power back to the grid. This allocation mechanism maintains the dynamic balance of system power and supports the safe operation of the grid.

[0029] In this invention, the charging strategy dynamic generation process includes a charging efficiency optimization module, which dynamically adjusts the matching relationship between charging voltage, current and charging time based on the characteristics of electric vehicle batteries and environmental parameters, reducing charging losses and battery thermal stress caused by charging parameter mismatch, and extending battery cycle life; a charging temperature rise control mechanism is introduced to monitor the battery charging temperature rise rate in real time. When the temperature rise rate exceeds the set range, the charging power is dynamically reduced or the charging mode is switched to maintain the thermal stability of the battery during the charging process.

[0030] In this invention, the reinforcement learning process of the AI ​​adaptive decision-making model adopts a dual reward mechanism, setting immediate rewards and long-term rewards respectively. The immediate reward is calculated based on the photovoltaic absorption rate and load regulation effect under the current scenario, while the long-term reward is calculated based on the operating cost and service satisfaction over a period of time. The total reward value is obtained by weighted summation, which guides the model to achieve a balance between short-term optimization and long-term benefits, prevents the model from getting trapped in local optima, and achieves synergistic optimization of short-term performance and long-term benefits.

[0031] This invention also includes a power grid interaction optimization mechanism. By establishing a real-time communication link with the power grid dispatching system, it obtains power grid dispatching instructions and operational constraints, and integrates the power grid instructions into the optimization objectives of the AI ​​model. When the power grid needs to shave peaks and fill valleys, it actively adjusts the charging load and energy storage charging and discharging status to respond to the power grid dispatching needs. At the same time, it feeds back charging load forecast information and adjustable load capacity to the power grid, providing data support for power grid dispatching, strengthening the dispatching coordination capability between the photovoltaic-storage-charging system and the power grid, and promoting the integrated operation of power generation, grid, load and storage.

[0032] In this invention, a fast response mechanism is activated in the emergency charging scenario. Non-essential auxiliary loads are preferentially cut off, and charging resources are maximally allocated to emergency charging vehicles according to the emergency priority. Meanwhile, the discharge power and discharge duration of the energy storage system are optimized to ensure the rapidity and stability of emergency charging. After the emergency scenario ends, it automatically switches to the conventional control strategy, restores the normal load distribution in a gradient manner, reduces the grid fluctuation amplitude during the load switching process, and maintains the stability of grid operation parameters.

[0033] This invention also includes a multi-station collaborative regulation mechanism. When multiple photovoltaic-energy storage-charging stations are in the same power grid area, a high-speed communication link between stations is established to share the photovoltaic power output, energy storage status, charging demand, and load regulation data of each station. Through a global optimization algorithm, the charging resources and energy storage scheduling plan in the area are allocated to maximize the photovoltaic power consumption in the area, balance the grid load distribution, eliminate the regional load distribution deviation caused by independent regulation of single stations, and improve the operation economy, safety, and stability of the regional power grid.

[0034] The following further illustrates the specific implementation manners of this system through two embodiments: Embodiment 1: Implementation of multi-scenario adaptive control and load regulation for a photovoltaic-energy storage-charging station in an urban commercial complex This embodiment is applied to a photovoltaic-energy storage-charging station supporting an urban commercial complex in the core area of the city. The station is equipped with 100 kW photovoltaic modules, a 500 kWh energy storage system, and 10 60 kW DC charging piles, serving electric vehicle users for shopping in surrounding malls and commuting in office buildings. It usually faces multi-scenario switches such as peak-valley electricity prices, conventional charging, and emergency charging. The operation state of the power grid fluctuates significantly under the influence of urban electricity loads, and multi-objective collaborative optimization needs to be achieved.

[0035] Scenario recognition and multi-source parameter collection Based on the power grid electricity price publishing system, photovoltaic inverter data interface, communication link with the power grid dispatching center, charging pile user interaction terminal, and battery management system, a multi-source data collection network is constructed. Scenario recognition is judged through multi-dimensional data fusion: for the peak-valley electricity price scenario, it is based on the time periods published by the local power grid. The peak periods are from 8:00 to 10:00 and from 18:00 to 22:00, the normal periods are from 10:00 to 18:00 and from 22:00 to 24:00, and the valley periods are from 0:00 to 8:00; for the scenario of abundant sunlight, the judgment criterion is that the output power of the photovoltaic modules is not less than 80 kW and the duration is not less than 10 minutes; for the scenario of tight power grid load, it is triggered by receiving the load limit instruction issued by the power grid dispatching center; for the emergency charging scenario, it is based on the user submitting an emergency charging request through the terminal and checking the emergency label; the remaining periods are judged as the conventional charging scenario.

[0036] Parameter acquisition employs differentiated frequencies: core parameters include photovoltaic module output power and conversion efficiency, remaining energy storage system capacity, charge / discharge efficiency and health status, real-time grid load, voltage frequency and transmission capacity, electric vehicle charging power demand, battery SOC status, battery type and charging cutoff voltage, with a sampling interval of 1 second; non-core parameters include ambient temperature and light intensity, with a sampling interval dynamically adjusted to 5 minutes. All acquired data is transmitted to the central controller via industrial Ethernet, with transmission latency controlled within 50ms.

[0037] Multi-source data fusion processing The collected raw data undergoes a preprocessing procedure: outliers are identified using the 3σ criterion, the mean and standard deviation of each parameter are calculated, and data points exceeding the range of mean minus three standard deviations to mean plus three standard deviations are removed. The outlier removal rate for parameters such as photovoltaic module output power and grid voltage frequency is controlled within 0.5%. For missing data, linear interpolation is used to fill in the missing data based on three valid data points before and after the missing data. The continuity error of the filled data does not exceed 2%.

[0038] Key features were extracted during the feature engineering phase: the photovoltaic output fluctuation coefficient equals the maximum output minus the minimum output during a certain period, divided by the average output during that period; the grid load change rate equals the current load minus the previous load, divided by the previous load; and the urgency of vehicle charging demand equals the charging cutoff voltage minus the current voltage, divided by the charging cutoff voltage, and multiplied by the user priority coefficient. A weighted fusion algorithm was used to integrate the multi-dimensional features, with the weights for photovoltaic output-related features set to 0.3, grid parameter-related features set to 0.3, and vehicle status-related features set to 0.4, generating a 12-dimensional standardized data matrix, with the data normalized to the 0-1 range.

[0039] An adaptive data filtering algorithm is employed during multi-source data fusion processing, with filtering parameters dynamically adjusted according to the scenario: a 5-second filtering window is set for scenarios with ample sunlight to enhance the response to fluctuations in photovoltaic output; a 10-second filtering window is set for scenarios with tight grid load to reduce the impact of grid parameter fluctuations on decision-making. Simultaneously, a confidence score is used to assess data reliability; data with a score of 0.85 or higher is considered reliable, while data with a score below 0.85 triggers a re-acquisition process for the corresponding parameters, ensuring the quality of the input data.

[0040] Building an AI Adaptive Decision Model A hybrid intelligent model integrating LSTM time-series prediction and DQN reinforcement learning was constructed. The LSTM network structure was set with 12 neurons in the input layer corresponding to the normalized data matrix dimension, 3 hidden layers with 64 neurons each, and 3 neurons in the output layer, corresponding to the predicted values ​​of photovoltaic power output, grid load, and charging demand, respectively. It was trained based on historical data from the past 30 days, with a training batch size of 32 and a learning rate of 0.001. The model converged after 1000 iterations, and the prediction error was controlled within 5%, achieving short-term predictions for the next 1 to 4 hours.

[0041] The DQN reinforcement learning model aims to minimize photovoltaic (PV) grid integration rate, grid load balancing, charging service satisfaction, and operating costs. It employs a dual-reward mechanism: the immediate reward is calculated as 0.3 times the PV integration rate plus 0.2 times the grid load balancing plus 0.3 times the charging service satisfaction minus 0.2 times the unit charging cost; the long-term reward is the total immediate reward divided by the period length (set to 24 hours), with a weighted summation of the total reward. The immediate reward has a weight of 0.6, and the long-term reward has a weight of 0.4. The model's experience replay pool capacity is set to 10,000, and the initial exploration rate is 0.9, gradually decreasing to 0.1 with iterations to ensure the model balances exploration and utilization.

[0042] Dynamic generation of charging strategy Based on AI model predictions and scenario types, an adaptive charging control strategy is generated: In peak-valley electricity pricing scenarios, the charging power of charging piles is increased to 90% to 100% of the rated power during off-peak hours, maintained at 70% to 80% of the rated power during normal hours, and reduced to 40% to 60% of the rated power during peak hours; In scenarios with ample sunlight, the charging power is adjusted in real time according to the photovoltaic output. When the photovoltaic output is not less than 80kW, the charging pile operates at full power. When the output is between 40kW and 80kW, the power is matched to 1.2 times the photovoltaic output. When the output is less than 40kW, the power is supplemented by energy storage discharge; In scenarios with tight grid load, the total charging power is reduced to 50% to 70% of the rated total power according to grid dispatch instructions; In emergency charging scenarios, the charging power of a single pile is increased to 100% of the rated power to prioritize rapid energy replenishment.

[0043] The charging mode selection is based on battery type and SOC status: when the lithium battery SOC is below 30%, a constant current charging mode is used with a current set to 0.3C; when the SOC is between 30% and 80%, a constant power charging mode is used; when the SOC is above 80%, a constant voltage charging mode is switched, with the voltage gradually approaching the charging cutoff voltage. A charging efficiency optimization module is integrated into the charging strategy, dynamically adjusting charging parameters based on ambient temperature: when the ambient temperature is below 10℃, the initial charging current is reduced to 0.2C to avoid damaging the battery with high current at low temperatures; when the temperature is between 10℃ and 35℃, charging is performed according to standard parameters; when the temperature is above 35℃, the charging power is reduced by 10% for every 5℃ increase, and the charging pile's cooling system is activated.

[0044] Load Coordination The load coordination and adjustment mechanism is activated, and the energy storage system and charging piles are linked through the central controller: When the photovoltaic output is excessive, that is, the output is greater than the total charging demand plus the grid's allowable power supply, the energy storage system charges at 80% of the rated charging and discharging power until the remaining power reaches 90% and then stops; when the photovoltaic output is insufficient, that is, the output plus the energy storage discharge power is less than the total charging demand or the grid load is at its peak, the energy storage system discharges according to the charging demand gap, the discharge power does not exceed 70% of the rated power, and the remaining power is maintained at more than 10% to ensure emergency backup.

[0045] Distributed load balancing algorithms are applied during load coordination and regulation, using power allocation formulas. Achieve power coordination. At a certain moment... When it is 10 seconds, There are 10 charging piles, of which 8 are operational and 2 are idle. The charging power of the operational charging piles is 54kW, 50kW, 48kW, 52kW, 46kW, 50kW, 49kW, and 51kW respectively. The total charging power of the 10 charging piles is 54+50+48+52+46+50+49+51+0+0=400kW. The energy storage system is in a discharging state. 80kW; grid interconnection power If the power absorbed from the grid is 20kW, then =400+80+20=500kW), maintaining dynamic power balance in the system. Simultaneously, the charging power of each charging pile is dynamically adjusted to ensure that the power of a single pile does not exceed the rated value of 60kW, and the load difference between phases within the area does not exceed 10%.

[0046] Real-time monitoring and feedback optimization The central controller monitors the output data of the photovoltaic inverter, the BMS data of the energy storage system, the data of the power grid monitoring terminal, the operation data of the charging pile, and the vehicle battery data in real time, and sets the normal range of each parameter: the conversion efficiency of the photovoltaic module is not less than 15%, the charging and discharging efficiency of the energy storage system is not less than 85%, the remaining power is 10% to 90%, the grid voltage deviation is not more than ±5%, the frequency deviation is not more than ±0.2Hz, the charging power fluctuation of the charging pile is not more than 5%, and the battery temperature rise rate is not more than 2℃ per minute.

[0047] When the grid voltage deviation reaches 6%, a feedback signal is generated and input into the AI ​​adaptive decision-making model. The model triggers online parameter adjustments: the feature weights of grid parameters are increased from 0.3 to 0.4, the output weights of neurons in the hidden layer of the LSTM network are adjusted, the weight of grid load balance in the DQN algorithm's instant reward is increased from 0.2 to 0.3, and the charging strategy is iteratively optimized by reducing the total power of charging piles by 10% to restore the grid voltage deviation to the normal range. All abnormal data and processing procedures are recorded in the system log in real time.

[0048] Scene adaptation iteration and other mechanisms The scene adaptation weight dynamic optimization process is based on the formula. Execution, in which The time interval is 3600 seconds, which is equivalent to one hour. The total number of feature dimensions is 12. The value ranges from 0.9 to 1.1 and is dynamically adjusted according to the stability of the scene. It takes 1.0 when the scene is stable and 1.1 when it fluctuates. Fixed values ​​are assigned based on the importance of the features: 0.15 for photovoltaic output, 0.15 for grid load, 0.2 for battery SOC, 0.2 for charging demand, and 0.3 for other features. Electricity price characteristics under peak-valley electricity pricing scenarios are calculated by correlation calculation between features and scenarios. The value is 0.9, which represents the photovoltaic characteristics under abundant sunlight. It is 0.9.

[0049] The dynamic assessment of charging demand priority adopts the analytic hierarchy process (AHP). The target layer is the charging priority, the criteria layer is the battery SOC status, charging reservation time, vehicle type, user level, and the solution layer is the vehicles to be charged. The weights of the criteria layer are set to 0.4, 0.3, 0.15 and 0.15, respectively. A judgment matrix is ​​constructed by pairwise comparison, the priority score of each vehicle is calculated and sorted, and power is allocated from high to low score when charging resources are scarce.

[0050] Table 1 Comparison of system performance before and after implementation of Example 1

[0051] Table 1 shows that the overall system performance has been significantly improved after the implementation of this invention. The photovoltaic absorption rate increased from 68% to 92%, thanks to dynamic power matching and energy storage coordination under abundant sunlight conditions, maximizing the utilization of clean photovoltaic energy. The unit charging cost decreased from RMB 1.8 per kWh to RMB 1.2 per kWh, attributed to the achievement of time-of-use optimization and operating cost optimization goals under peak-valley electricity pricing scenarios. The grid load fluctuation amplitude decreased from 25% to 8%, demonstrating the smoothing effect of the load coordination mechanism on grid fluctuations and reducing the impact on the urban power grid. The emergency charging response time was shortened from 15 minutes to 5 minutes, verifying the effectiveness of the rapid response mechanism and priority evaluation system in emergency scenarios. The average battery cycle life was significantly improved, due to the protection of the battery by the charging efficiency optimization and temperature rise control mechanism. These improvements fully demonstrate the technical advantages of this invention in multi-scenario adaptation and multi-objective coordinated optimization.

[0052] Example 2: Implementation of Multi-Scenario Adaptive Control and Load Regulation for Suburban Distributed Photovoltaic Storage and Charging Stations This embodiment is applied to a distributed photovoltaic-storage-charging station in a suburban industrial park. The station is equipped with 200kW photovoltaic modules, a 1000kWh energy storage system, and 15 60kW DC charging piles. It also forms a regional cluster with another photovoltaic-storage-charging station located 1 kilometer away, serving commuter vehicles of enterprises in the park and vehicles of surrounding residents. The scenario is characterized by abundant sunshine and conventional charging, with relatively relaxed grid load but large fluctuations in sunshine conditions, requiring the maximization of photovoltaic absorption and regional collaborative optimization.

[0053] Scene recognition and multi-source parameter acquisition Scene recognition is based on multi-source data linkage judgment: the standard for a scene with sufficient sunlight is that the output power of photovoltaic modules is not less than 150kW and the duration is not less than 15 minutes; the peak-valley electricity price scene is based on the local power grid time period, with peak periods being 7:00 to 11:00 and 17:00 to 21:00, normal periods being 11:00 to 17:00 and 21:00 to 24:00, and valley periods being 0:00 to 7:00; the power grid load shortage scene is triggered by the power grid voltage and frequency exceeding the normal range for 1 minute; the emergency charging scene is based on the charging application submitted by the company for official emergency vehicles; the remaining time periods are regular charging scenes.

[0054] Parameter acquisition employs differentiated frequencies: core parameters include photovoltaic module output power and conversion efficiency, remaining energy storage system capacity, charge / discharge efficiency and health status, real-time grid load, voltage frequency and transmission capacity, electric vehicle charging power demand, battery SOC status, battery type and charging cutoff voltage, with a sampling interval of 1 second; non-core parameters include ambient temperature, light intensity, and wind speed, with a sampling interval of 10 minutes. Data transmission is achieved through 5G industrial modules, with a transmission latency of no more than 100ms, ensuring cross-site data sharing requirements.

[0055] Multi-source data fusion processing Raw data preprocessing process: Outliers are removed using the 3σ criterion, the mean and standard deviation of each parameter are calculated, and data points exceeding the range of mean minus three standard deviations to mean plus three standard deviations are removed to ensure data validity; missing data are filled using linear interpolation, calculated based on 5 valid data points before and after, with the filling error not exceeding 3%.

[0056] Feature engineering extracts key features: photovoltaic output fluctuation coefficient, grid load change rate, vehicle charging demand urgency, and light intensity change rate. A weighted fusion algorithm is used to integrate these features, with a weight of 0.4 for photovoltaic output-related features, 0.2 for grid parameter-related features, 0.3 for vehicle status-related features, and 0.1 for environmental parameter-related features, generating a 14-dimensional standardized data matrix, which is normalized to the 0 to 1 range.

[0057] The adaptive data filtering algorithm dynamically adjusts parameters according to the scene: the filtering window is set to 8 seconds when the light fluctuation is large and 12 seconds when the power grid is stable; the data reliability assessment adopts the confidence score method, and the data is considered reliable if the score is not lower than 0.9. If the score is lower than 0.9, re-acquisition is triggered to ensure the quality of the data input to the AI ​​model.

[0058] Building an AI Adaptive Decision Model The hybrid intelligent model consists of LSTM time-series prediction and DQN reinforcement learning. The LSTM network structure has 14 neurons in the input layer, 3 hidden layers with 80 neurons each, and 3 neurons in the output layer. It is trained based on historical data from the past 60 days, with a batch size of 64, a learning rate of 0.0008, and converges after 1500 iterations. The prediction error does not exceed 4%, enabling accurate prediction of photovoltaic power output, grid load, and charging demand for the next 1 to 4 hours.

[0059] The DQN reinforcement learning model aims to optimize photovoltaic (PV) grid integration rate, grid load balancing, charging service satisfaction, and operating cost. It employs a dual-reward mechanism: the immediate reward is calculated as 0.4 times the PV integration rate plus 0.2 times the grid load balancing rate plus 0.2 times the charging service satisfaction rate minus 0.2 times the unit charging cost; the long-term reward is calculated by weighting the total immediate reward over the period by 24 hours, with both immediate and long-term rewards having a weight of 0.5. The model's experience replay pool has a capacity of 20,000, with an initial exploration rate of 0.8, gradually decreasing to 0.05 to balance exploration and utilization.

[0060] Dynamic generation of charging strategy Charging strategies are generated based on scenario type and prediction results: In scenarios with abundant sunlight, the charging pile operates at full power when the photovoltaic output is not less than 150kW; when the output is between 80kW and 150kW, the power is matched to 1.1 times the photovoltaic output; when the output is less than 80kW, it is supplemented by energy storage discharge; in scenarios with peak-valley electricity prices, the charging pile power is 80% to 100% during valley hours, 60% to 80% during flat hours, and 40% to 50% during peak hours; in scenarios with tight grid load, the total charging power is reduced to 60% to 80% of the rated total power according to grid requirements; in emergency charging scenarios, each charging pile operates at full power.

[0061] Charging mode selection: Constant current charging at 0.25C when the lithium battery SOC is below 20%; constant power charging when the SOC is between 20% and 85%; constant voltage charging when the SOC is above 85%. The charging efficiency optimization module adjusts parameters based on ambient temperature: initial current of 0.15C when the temperature is below 5℃; standard parameters from 10℃ to 30℃; and power reduced by 8% for every 5℃ increase above 30℃, working in conjunction with the charging pile's cooling system to maintain a battery temperature rise rate of no more than 1.5℃ per minute.

[0062] Load Coordinated Regulation and Multi-Station Coordination The load coordination and regulation mechanism links the energy storage system and charging piles: when the photovoltaic output is excessive, the energy storage system charges at 90% of the rated power and stops when the remaining power reaches 95%; when the photovoltaic output is insufficient or the grid load is at its peak, the energy storage system discharges according to demand, with the power not exceeding 80% of the rated power and the remaining power not less than 15%.

[0063] Power distribution formula Applied to load regulation, at a certain moment When it is 20 seconds, There are 15 charging piles, 12 of which are in operation, with charging powers of 58kW, 55kW, 52kW, 56kW, 53kW, 57kW, 54kW, 55kW, 51kW, 53kW, 56kW, and 52kW respectively. The total charging power of the 15 charging piles is 58+55+52+56+53+57+54+55+51+53+56+52+0+0+0=662kW; the energy storage system discharges... 120kW; grid interconnection power The power absorbed is 18kW, meaning it absorbs power from the grid. =662+120+18=800kW, maintaining power balance.

[0064] The multi-station coordinated regulation mechanism achieves data sharing between the two stations through a 5G communication link. The shared content includes photovoltaic output, energy storage status, charging demand, and load regulation data. A particle swarm optimization algorithm is used for regional global optimization. The objective function is to maximize the regional photovoltaic absorption and the highest load balance. The optimal scheduling scheme is obtained after 50 iterations, which allocates charging resources and energy storage scheduling strategies between the two stations to avoid resource idleness and load imbalance in the region.

[0065] Real-time monitoring and feedback optimization Real-time monitoring covers the entire process of photovoltaics, energy storage, power grid, charging piles, and vehicles, with the following normal parameter ranges: photovoltaic module conversion efficiency not less than 16%, energy storage charging and discharging efficiency not less than 86%, power grid voltage deviation not exceeding ±5%, frequency deviation not exceeding ±0.2Hz, charging pile power fluctuation not exceeding 4%, and battery temperature rise rate not exceeding 1.5℃ per minute.

[0066] When the photovoltaic output fluctuation coefficient exceeds 0.3, the feedback signal triggers the adjustment of AI model parameters: the photovoltaic feature weight of the LSTM network is increased by 0.1, the photovoltaic absorption rate weight in the DQN algorithm's instant reward is increased from 0.4 to 0.5, and the charging power allocation strategy is optimized to mitigate the impact of photovoltaic fluctuations. Abnormal data and processing are recorded in real time, supporting subsequent traceability analysis.

[0067] Scene adaptation and interaction with power grid Dynamic optimization of scene adaptation weights according to formula implement, The time interval is 7200 seconds, which is equivalent to a two-hour period. The total number of feature dimensions is 14. The value ranges from 0.85 to 1.15. Based on the importance of the characteristics, the allocation is as follows: photovoltaic output characteristics 0.2, grid load characteristics 0.15, charging demand characteristics 0.25, environmental characteristics 0.1, and other characteristics 0.3. Calculate based on scene relevance to improve the accuracy of model scene adaptation.

[0068] The power grid interaction optimization mechanism communicates with the power grid dispatching system in real time to obtain dispatching instructions and constraints, which are then integrated into the optimization objectives of the AI ​​model. During peak shaving, the power grid proactively reduces the charging load by 20% to 30%, and during valley filling, it increases the charging load by 10% to 20%. At the same time, it provides feedback on charging load forecasts and adjustable capacity to support power grid dispatching.

[0069] Table 2 Comparison of system performance before and after implementation of Example 2

[0070] Table 2 shows that the performance of the regional photovoltaic-storage-charging system has been comprehensively optimized after the implementation of this invention. The regional photovoltaic absorption rate has increased from 72% to 95%, mainly due to precise power matching, energy storage coordinated regulation, and multi-station global optimization under abundant sunshine scenarios, fully tapping the potential of abundant photovoltaic resources in suburban areas. The regional grid load balance has increased from 65% to 90%, reflecting the effectiveness of the multi-station coordinated regulation mechanism and eliminating load imbalance caused by independent operation of a single station. The unit charging cost has decreased from RMB 1.7 per kWh to RMB 1.1 per kWh, attributed to cost optimization under peak-valley electricity pricing scenarios and the improvement of photovoltaic absorption rate. The emergency charging response time has been shortened from 12 minutes to 4 minutes, verifying the practicality of the emergency response mechanism and priority evaluation system. The charging pile utilization rate has increased from 60% to 85%, reflecting the effectiveness of the charging resource balanced allocation strategy and avoiding resource idleness. The improvement of various indicators proves that this invention can effectively adapt to the operational needs of distributed photovoltaic-storage-charging stations in suburban areas, achieving multi-scenario adaptive control and regional coordinated optimization.

[0071] Reference Figure 2 This figure visually demonstrates the significant performance improvements of the core components before and after the implementation of this invention, fully validating the effectiveness of the technical solution. The photovoltaic absorption rate increased from 70% to 93%, stemming from the precise matching of photovoltaic output and charging power in scenarios with abundant sunlight, as well as the storage and reuse of excess photovoltaic energy by the energy storage system, maximizing the potential of clean energy. The grid load fluctuation amplitude decreased from 23% to 9%, reflecting the core role of the load coordination and regulation mechanism in suppressing fluctuations through the energy storage charging and discharging level, reducing the impact of charging load on the grid. The reduction in unit charging cost is attributed to the optimized allocation of power during peak-valley pricing scenarios; the shortened emergency charging response time reflects the practical value of the priority assessment system and rapid response mechanism; and the increased utilization rate of charging piles proves the effectiveness of the balanced allocation strategy for charging resources. The improvements in all indicators are directly related to core technologies such as multi-scenario adaptive decision-making and load coordination and regulation, highlighting the comprehensive advantages of this invention in multi-objective optimization.

[0072] Reference Figure 3This figure illustrates the distribution of charging efficiency under different scenarios, highlighting the targeted and effective multi-scenario adaptive control of this invention. The scenario with abundant sunlight accounts for the highest proportion (35%), indicating that the system achieves the highest charging efficiency in this scenario through precise matching of photovoltaic output and charging power, consistent with the optimization goal of maximizing photovoltaic power consumption. The scenario with peak-valley electricity prices accounts for 28%, reflecting the system's efficient operation in cost-optimized scenarios, reducing operating costs while maintaining high charging efficiency through charging with low-priced electricity during off-peak hours. The conventional charging scenario accounts for 20%, ensuring stable and efficient daily charging services. The scenarios with stable grid load and emergency charging have relatively low proportions, but still maintain stable efficiency, corresponding to grid-friendly operation and rapid energy replenishment needs, respectively. This chart demonstrates that this invention can formulate personalized strategies for the core needs of different scenarios, maintaining high charging efficiency throughout multi-scenario switching and achieving multi-objective collaborative optimization.

[0073] Reference Figure 4 This figure illustrates the effect of the multi-station collaborative mechanism on improving regional load balance, highlighting the regional optimization capabilities of this invention. Before collaboration, the load balance of each station was between 58% and 67%, all below the excellent threshold of 85%, indicating a significant regional load imbalance, with some stations at saturation and others underutilized. After collaboration, the load balance of each station improved to 88%-92%, all reaching excellent levels, and the difference between stations narrowed to less than 4%, achieving a uniform distribution of regional load. This effect stems from the multi-station collaborative adjustment mechanism sharing data on photovoltaic output, energy storage status, and charging demand of each station through high-speed communication links, and employing a global optimization algorithm to allocate charging resources and energy storage scheduling schemes within the region, eliminating resource waste and load imbalance caused by independent operation of a single station. This figure demonstrates that this invention can overcome the limitations of a single station, maximizing photovoltaic absorption and grid load balance at the regional level, and improving overall operating efficiency and grid security and stability.

[0074] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for adaptive control and load regulation of photovoltaic energy storage charging in multiple scenarios, characterized in that, Includes the following steps: Scene recognition and multi-source parameter acquisition: Based on electricity price, solar irradiance, grid dispatch instructions, user charging request priority, and battery status data, real-time identification of peak-valley electricity price, abundant solar irradiance, grid load shortage, emergency charging, and regular charging scenarios; simultaneous acquisition of multi-dimensional operating parameters; Multi-source data fusion processing: The original data is preprocessed, outliers are removed using the 3σ criterion, and missing data is filled using linear interpolation; key features are extracted through feature engineering, and a weighted fusion algorithm is used to integrate and generate a standardized data matrix; AI Adaptive Decision Model Construction: Construct a hybrid intelligent model that integrates time series prediction and reinforcement learning. The LSTM network completes short-term prediction of photovoltaic output, grid load and charging demand for the next 1-4 hours. The DQN algorithm uses multi-objective dynamic learning to learn the optimal control strategy under different scenarios. Dynamic generation of charging strategies: Based on AI model prediction results and scenario types, an adaptive charging control strategy including charging power allocation and charging mode selection is generated. Load coordination regulation: Combining charging strategy with grid operating status, the load coordination regulation mechanism is activated to suppress fluctuations by controlling the charging and discharging levels of the energy storage system; Real-time monitoring and feedback optimization: Real-time monitoring of various operating parameters and adjustment effects, comparison with preset thresholds, and triggering online adjustment of AI model parameters and iterative optimization of strategies when data is abnormal or the effect does not meet expectations; Scene adaptive iteration: Continuously collect new scene data and policy execution effect data, regularly update the training sample set, and start model retraining.

2. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, It also includes a dynamic optimization process for scene adaptation weights, which adjusts the weights of each feature in real time through multi-factor coupling analysis. The optimization formula is as follows: ,in For the scene The feature weight matrix, for The weighting adjustment coefficient at time step, For the weight update cycle, The total number of feature dimensions. for Time of the first Importance coefficient of dimensional features for Time of the first Dimensional features and scenarios The fitness function.

3. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, It also includes a dynamic evaluation mechanism for charging demand priority, which builds a priority evaluation system based on the electric vehicle's battery SOC status, charging reservation time, vehicle type and user level, and uses the analytic hierarchy process to quantify the charging priority of each vehicle; in scenarios where charging resources are scarce, charging power and charging time are allocated according to priority; in scenarios where charging resources are abundant, charging resources are allocated evenly.

4. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, The multi-source data fusion process employs an adaptive data filtering algorithm, which dynamically adjusts the filtering parameters to adapt to the data fluctuation characteristics under different scenarios, effectively filtering high-frequency noise in photovoltaic output and grid load data. A data evaluation mechanism is introduced, using a confidence score method to quantitatively score the fused data. When the score is lower than the set standard, a data re-acquisition process is triggered.

5. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, Distributed load balancing algorithms are introduced into the load coordination and regulation process, using power allocation formulas. Complete power coordination, among which for The total power balance value of the system at any given time. The number of charging stations. for Time of the first The charging power of each charging station for The charging and discharging power of the energy storage system at all times. for The power exchanged with the power grid at any time is taken as a positive value when absorbing power from the grid and a negative value when feeding power back to the grid.

6. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, The dynamic generation of the charging strategy includes a charging efficiency optimization module, which dynamically adjusts the matching relationship between charging voltage, current and charging time based on the characteristics of electric vehicle batteries and environmental parameters; and introduces a charging temperature rise control mechanism to monitor the battery charging temperature rise rate in real time.

7. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, The reinforcement learning process of the AI ​​adaptive decision-making model adopts a dual reward mechanism, setting immediate rewards and long-term rewards respectively. The immediate reward is calculated based on the photovoltaic absorption rate and load regulation effect in the current scenario, while the long-term reward is calculated based on the operating cost and service satisfaction over a period of time. The total reward value is obtained by weighted summation, which guides the model to achieve a balance between short-term optimization and long-term benefits.

8. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, It also includes a power grid interaction optimization mechanism, which establishes a real-time communication link with the power grid dispatch system to obtain power grid dispatch instructions and operating constraints, and integrates power grid instructions into the optimization objectives of the AI ​​model; when the power grid needs to shave peaks and fill valleys, it actively adjusts the charging load and energy storage charging and discharging status; at the same time, it feeds back charging load forecast information and adjustable load capacity to the power grid.

9. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, In emergency charging scenarios, a rapid response mechanism is activated to prioritize cutting off unnecessary auxiliary loads, allocate charging resources to emergency charging vehicles to the maximum extent according to emergency priority, and optimize the discharge power and discharge duration of the energy storage system. After the emergency scenario ends, it automatically switches to the normal control strategy and restores normal load distribution in a tiered manner.

10. The photovoltaic-storage-charging multi-scenario adaptive control and load regulation method according to claim 1, characterized in that, It also includes a multi-station coordinated regulation mechanism. When multiple photovoltaic, energy storage and charging stations are located in the same power grid area, a high-speed communication link is established between the stations to share data on photovoltaic output, energy storage status, charging demand and load regulation of each station.

Images