A power intelligent distribution and anti-collision method for optical storage and charging system

Abstract

The application provides a power intelligent distribution and conflict prevention method for a light storage charging system in the technical field of new energy micro-grid energy management, which comprises the following steps: collecting operation, environment, user and market data in real time, predicting photovoltaic output power, charging pile demand, dispatchable load power and energy storage battery life attenuation cost in a future period; establishing a multi-objective optimization model, and setting constraints such as grid-connected point power, battery state and basic demand; solving the model based on the predicted data, generating an optimized power distribution plan and control instruction of the energy storage, charging pile and load, and executing the charging and discharging power and load control in real time; continuously monitoring the actual power deviation in the control period, triggering rolling optimization when the limit is exceeded, and regularly updating the battery health state and user habit model, and adaptively adjusting the optimization parameters to maintain the efficient operation of the system. The application has the advantage that the multi-objective dynamic collaborative optimization capability of the system under the power safety constraint is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of energy management technology for new energy microgrids, and specifically to a method for intelligent power allocation and anti-collision for photovoltaic-storage-charging systems. Background Technology

[0002] With the continued advancement of the "dual-carbon" strategy, photovoltaic-storage-charging systems, integrating photovoltaic power generation, energy storage batteries, electric vehicle charging, and household electricity consumption, have been widely applied in residential communities, industrial parks, and other scenarios. This system helps increase the self-consumption rate of clean energy, reduces users' electricity costs, and plays a supplementary regulatory role in peak shaving and valley filling of the power grid. However, in actual operation, an increasingly prominent contradiction arises: the total power capacity of the system's connection point (such as the rated current of the inlet switch) has an upper limit, while various loads within the system—especially high-power electric vehicle charging piles and household appliances such as air conditioners and water heaters—may overlap in their electricity consumption periods, causing the total power demand to momentarily exceed the safety limit, triggering circuit breaker tripping and power outages.

[0003] To address the aforementioned power conflict issue, current solutions can be mainly divided into two categories: The first type is a priority-based power-off scheme based on real-time monitoring. This scheme monitors the total load power in real time and, when approaching the power limit, shuts down some loads according to a preset fixed priority (such as prioritizing residential power supply and cutting off power to charging stations). While this type of scheme is simple and direct, it is a "passive response" strategy with the following obvious drawbacks: First, the user experience is poor, as the charging process may be forcibly interrupted; second, the strategy lacks flexibility and cannot be dynamically adjusted based on user habits and system status, easily leading to unnecessary power outages and energy waste.

[0004] The second type is a power allocation scheme based on fixed rules, such as setting fixed charging periods or power limits for charging piles. This scheme lacks intelligent adaptability and is difficult to cope with fluctuations in photovoltaic power generation, the randomness of user electricity consumption behavior, and the dynamic changes in the state of energy storage batteries. Therefore, it cannot achieve comprehensive optimization of system energy efficiency and economy while ensuring power safety.

[0005] More importantly, most existing solutions focus only on instantaneous power balance, neglecting the long-term economic efficiency and equipment health management of the system. For example, the cycle life of energy storage batteries is closely related to their charge and discharge strategies; frequent high-power charge and discharge or shallow charge and discharge will accelerate battery capacity degradation. Existing anti-collision strategies often do not take into account the life degradation model of such critical equipment when making decisions, and therefore are not optimal from the perspective of total life cycle cost.

[0006] In summary, existing technologies mostly employ passive, rigid, and locally optimized strategies, lacking dynamic, collaborative, and intelligent decision-making capabilities for the overall system state and long-term goals. Specifically, this manifests in two ways: First, the strategies are passive and crude, relying on post-event power outages or fixed rules, impacting user experience and failing to adaptively adjust based on real-time power generation, demand, and energy storage status, resulting in low system energy efficiency. Second, the decision-making dimensions are singular, focusing solely on instantaneous power balance as the control objective, failing to comprehensively consider multi-dimensional dynamic factors such as photovoltaic output fluctuations, random user behavior, and the health status of energy storage batteries (e.g., cycle life). Therefore, it is difficult to achieve global optimization of the system's economy and reliability throughout its entire lifecycle while ensuring power safety.

[0007] Therefore, how to provide a method for intelligent power allocation and anti-collision in photovoltaic energy storage and charging systems, and improve the system's multi-objective dynamic collaborative optimization capability under power safety constraints, has become an urgent technical problem to be solved. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a method for intelligent power allocation and anti-conflict in photovoltaic energy storage and charging systems, so as to improve the system’s multi-objective dynamic collaborative optimization capability under power safety constraints.

[0009] This invention is implemented as follows: A method for intelligent power allocation and anti-collision in an optical energy storage and charging system, comprising the following steps: Step S1: Collect system operation data, environmental data, user data and market data of the photovoltaic-storage-charging system in real time, and predict the output power curve of the photovoltaic unit, the adjustable charging demand curve of the charging pile, the power demand curve of each schedulable load, and the life decay cost of the energy storage battery unit under different charging and discharging strategies within a preset period in the future. Step S2: Establish a multi-objective optimization model with a future preset time period as the optimization cycle. The multi-objective optimization model takes the lowest total system operating cost, the highest user energy comfort, and the lowest energy storage battery unit life decay cost as the comprehensive optimization objectives, and takes the following constraints as the total power of the grid connection point not exceeding the limit in real time, the charging and discharging power and state of charge of the energy storage battery unit not exceeding the limit, and meeting the basic requirements of charging piles and dispatchable loads. Step S3: Based on the output power curve, adjustable charging demand curve, power demand curve, and lifetime degradation cost, solve the multi-objective optimization model to obtain the optimized power allocation plan for energy storage battery units, charging piles, and various schedulable loads in the future preset time period, and then generate the power control command for the current control cycle. Step S4: Execute the power control command to control the charging and discharging power of the energy storage battery unit, the charging power of the charging pile, and the activation status and power of the schedulable load in real time. Step S5: During the control cycle, continuously monitor the deviation between the actual power and the optimized power allocation plan. If the deviation exceeds the set deviation threshold, trigger the rolling optimization process. Step S6: Regularly update the health status model and user habit model of the energy storage battery unit, and adaptively adjust the relevant parameters of the multi-objective optimization model.

[0010] Furthermore, step S1 specifically includes: Real-time acquisition of system operation data, environmental data, user data, and market data of the photovoltaic energy storage and charging system; The system operation data includes: the actual power generation of the photovoltaic unit, the real-time state of charge, health parameters, internal resistance, temperature and historical cycle data of the energy storage battery unit, the real-time charging power, connection status and vehicle battery information of the charging pile, the real-time power and switching status of each schedulable load, and the total power limit and real-time total power of the system grid connection point. The environmental data includes: real-time meteorological data and weather forecast information for a future preset period; The user data includes: historical charging records of the charging pile, the user's expected vehicle retrieval time and target battery level, and historical usage records and user preference settings for each schedulable load. The market data includes: current and future time-of-use electricity prices and feed-in electricity prices; Based on the weather forecast information and the historical actual power generation of the photovoltaic unit, the output power curve of the photovoltaic unit is predicted within a future preset period. Based on the user data, the adjustable charging demand curve of the charging pile and the power demand curve of each schedulable load are predicted within a preset time period in the future. Based on the real-time state of charge, health parameters, internal resistance, temperature, and historical cycle data of the energy storage battery cell, the lifetime degradation cost under different charge and discharge strategies is predicted.

[0011] Furthermore, in step S1, the output power curve is predicted based on a photovoltaic power output prediction model. The photovoltaic power output prediction model takes irradiance, temperature, and cloud cover from weather forecast information as inputs and is trained using the historical actual power generation of the photovoltaic unit and the corresponding historical meteorological data. The photovoltaic power output prediction model is a long short-term memory network model based on an attention mechanism, and its training process specifically includes: Time alignment and cleaning of historical actual power generation data with historical meteorological data were performed to construct a training dataset; The training dataset is input into the photovoltaic power output prediction model, and the attention mechanism automatically learns and weights the historical meteorological features and power change patterns most relevant to the prediction of a specific future period. The network parameters of the photovoltaic power output prediction model are optimized by backpropagation algorithm to minimize the error between the output power curve predicted by the photovoltaic power output prediction model and the historical actual power generation.

[0012] Furthermore, in step S1, the prediction process of the adjustable charging demand curve specifically includes: Based on the user data, obtain the user's expected vehicle retrieval time T_target, the vehicle's current battery charge SoC_current, and the total battery capacity C; Based on the maximum allowable charging power P_max_charge of the charging pile, calculate the shortest time T_min required to fully charge at constant power. The charging demand time window is defined as [T_target-T_min-ΔT_buffer,T_target], where ΔT_buffer is a time buffer amount preset based on the uncertainty of user behavior; Within the charging demand time window, a segmented charging power demand function is constructed with the charging completion time T_finish as the variable. The segmented charging power demand function satisfies that the cumulative charging amount from the current time to T_finish is equal to C*(1-SoC_current), and the charging power is continuously adjustable within the range of [0, P_max_charge] at any time.

[0013] Furthermore, in step S1, the prediction of the power demand curve specifically includes: For air conditioners and water heaters, a hybrid prediction model based on physical models and data-driven approaches is established: using a building thermodynamic model and combining future temperature and sunshine forecasts, the basic heat load is calculated; at the same time, a neural network is trained using historical user adjustment data to predict the random adjustment behavior of user-set temperatures, and the power demand curve is output by combining the basic heat load and random adjustment behavior. For washing machines and dishwashers, based on the historical operation time patterns of users, a hidden Markov model is used to predict their start-up probability distribution during future optimization periods, and then converted into a power demand curve. The specific process for predicting the lifetime degradation cost is as follows: Based on the real-time state of charge, health parameters, internal resistance, temperature and historical cycle data of the energy storage battery cell, the current health status of the energy storage battery cell is calculated. Based on the electrochemical age model, the equivalent influence factor K_degrade of different charge-discharge depths and average charge-discharge rates on the battery cycle life degradation under the current health state is calculated. Quantify the cost of lifespan decay as follows: Lifetime degradation cost = K_degrade * |P_bss(t)| * Δt * C_battery; Where P_bss(t) is the charging and discharging power of the energy storage battery unit at time t; Δt is the control period; and C_battery is the purchase cost conversion factor corresponding to the unit battery capacity.

[0014] Furthermore, in step S2, the formula for calculating the total operating cost of the system is: Total system operating cost = Electricity purchase cost - Electricity sales revenue + Demand response penalty cost + Lifetime degradation cost; Among them, the cost of purchasing electricity is calculated based on the time-of-use price and the power absorbed from the grid; the revenue from selling electricity is calculated based on the grid connection price and the power fed back to the grid; the demand response penalty cost is activated when the total demand power of the system exceeds the demand response threshold agreed with the grid. The user's energy comfort level is quantitatively represented by constructing a comprehensive dissatisfaction index D_unsat: D_unsat=Σ_i(w_i*(ΔE_i / E_i_nom)^2)+ w_ev*((T_finish_ev-T_target_ev) / T_target_ev)^2; Where i represents each schedulable load; E_i_nom represents the rated energy consumption of the schedulable load; ΔE_i represents the deviation between the actual energy consumption of the schedulable load and E_i_nom within an optimization cycle; w_i represents the weighting coefficient; T_finish_ev represents the actual charging completion time of the electric vehicle; T_target_ev represents the user's expected vehicle pickup time; and w_ev represents the weighting coefficient for charging delay.

[0015] Furthermore, in step S3, the solution of the multi-objective optimization model employs an improved non-dominated sorting genetic algorithm, specifically including: The optimization variables are encoded, including the power of the energy storage battery unit in each control time slot, the power of the charging pile in each control time slot, and the start-stop time sequence of each schedulable load. Initialize the population and introduce initial individuals generated based on historical best solutions and heuristic rules to improve the quality of the initial population; In each generation of evolution, the total system operating cost, user energy comfort, and lifespan degradation cost of individual energy storage battery units are calculated, and fast non-dominated sorting and congestion calculation are performed. The offspring population is generated using tournament selection, simulated binary crossover, and polynomial mutation operators. An elite retention strategy is introduced, merging the parent and offspring populations, and selecting the next generation of population based on non-dominance level and crowding. After reaching the maximum number of iterations, the final optimal compromise solution is determined from the first non-dominated front solution set, based on the preset target weight or the compromise point selected by user interaction, as the optimized power allocation plan.

[0016] Furthermore, in step S4, based on the real-time total power margin of the system's grid connection points, different operating modes are dynamically entered, including: Ample Mode: When the real-time total power of the system grid connection point is detected to be lower than 60% of the total power limit, the charging piles and dispatchable loads operate at the maximum power required by the user, and the surplus photovoltaic power and / or grid power charge the energy storage battery unit. The charging power adopts the planned value of the optimized power allocation plan or the current maximum value. Optimization mode: When the real-time total power of the system's grid connection point is between 60% and 95% of the total power limit, the power control command of the current control cycle is strictly executed; Emergency Mode: When the real-time total power of the system's grid connection points exceeds 95% of the total power limit, the power control command is immediately interrupted, and a rapid reduction strategy based on dynamic priority is initiated: First, the power of all charging piles is reduced to a safe level proportionally according to a preset ratio. If the limit is still exceeded, the power is sequentially delayed and shut down according to the real-time interruptibility level of the schedulable load, and an alarm is sent to the user. After the real-time total power recovers to below 85% of the total power limit and stabilizes for more than a preset time, the system automatically switches back to optimization mode.

[0017] Furthermore, in step S5, the rolling optimization process specifically includes: Record the deviation between the current actual power and the optimized power allocation plan, including photovoltaic output deviation ΔP_pv and load demand deviation ΔP_load; Starting from the current moment, the prediction step is re-executed with a rolling optimization window shorter than the optimization period. Within the rolling optimization window, the planned trajectory of the state of charge of the energy storage battery cells is corrected to ensure that it returns to near the target trajectory of the optimized power allocation plan at the end of the rolling optimization window.

[0018] Furthermore, in step S6, the adaptive adjustment specifically includes: Based on the complete charge-discharge cycle data of the energy storage battery cell within a preset period, the relationship between its capacity decay and cumulative throughput and average depth of discharge is fitted, and the model parameters of the electrochemical age model are updated online. Based on user feedback ratings regarding past charging completion time delays, dynamically adjust the weighting coefficient w_ev within the overall dissatisfaction index D_unsat; Based on the Pareto front distribution of historical optimization results, dynamically adjust the weighting coefficients when transforming a multi-objective optimization problem into a single-objective problem in a multi-objective optimization model, or adjust the preference rules for selecting the optimal compromise solution.

[0019] The advantages of this invention are: 1. Enhance the system's multi-objective dynamic collaborative optimization capability under power security constraints by constructing a closed-loop decision-making framework of "prediction-optimization-execution-monitoring-adaptation": First, predict photovoltaic and flexible load demand and battery life degradation costs based on multi-dimensional data. Then, with the comprehensive goal of minimizing the total system operating cost, user comfort, and battery life degradation costs, establish and solve a global optimization model covering energy storage, charging piles, and all schedulable loads to generate and execute a collaborative power allocation plan. Ensure real-time power security by introducing rolling optimization triggered by actual deviations and a working mode that dynamically switches according to the power margin of the grid connection point (including emergency rapid reduction strategy). At the same time, regularly update the battery health and user habit models and adaptively adjust optimization parameters so that the entire system can continuously and dynamically balance multiple objectives such as economy, experience, and equipment health while meeting strict power constraints.

[0020] 2. Establish a fully intelligent closed-loop management and control architecture of "prediction-optimization-control-rolling-self-learning" with strong systematicity and adaptability: Construct a complete intelligent management and control closed loop, from multi-source data collection and high-precision prediction, to optimization decision-making based on multi-objective models, to precise execution and dynamic mode switching, and finally to deviation correction through rolling optimization and self-evolution through regular model updates. This closed-loop architecture ensures that the system can not only make optimal decisions based on the current state, but also cope with uncertainties in actual operation (rolling optimization), and continuously learn from historical operating data to optimize its own model parameters (adaptive adjustment), so that the system can maintain a high-efficiency and reliable operating state in the long term, significantly improving the robustness and intelligence level of the overall solution.

[0021] 3. A multi-objective optimization model that comprehensively considers economy, comfort, and equipment lifespan is proposed to achieve refined coordination and optimal balance of benefits from multiple parties: "lowest total system operating cost", "highest user energy comfort" and "minimum cost of energy storage battery cell lifespan degradation" are taken as optimization objectives. By constructing a precise quantitative model (such as using the comprehensive dissatisfaction index D_unsat to quantify comfort), the multi-objective problem is transformed into a solvable optimization problem. Under the premise of meeting the hard constraints of the power grid and equipment, Pareto optimal solutions are found in multiple dimensions such as reducing user electricity costs, ensuring user electricity / vehicle experience, and extending the lifespan of expensive energy storage equipment. This achieves synergistic optimization of multiple stakeholders, and the comprehensive benefits far exceed those of single-objective optimization, making it more commercially valuable.

[0022] 4. Advanced, fine-grained, and targeted forecasting methods are employed to provide a highly reliable data foundation for optimized decision-making: Advanced forecasting models are used for different objects, significantly improving forecast accuracy; for example, an attention-based LSTM network is used to predict photovoltaic power, which can effectively capture the complex spatiotemporal correlation between meteorological sequences and power generation; an "adjustable charging demand curve" and flexible time window are constructed for electric vehicle charging demand, which respects users' vehicle pickup time and provides flexible space for power scheduling; a hybrid "physical + data-driven" forecasting model is established for loads such as air conditioners and water heaters, taking into account both physical laws and the randomness of user behavior; these high-precision forecasting results are the prerequisite for multi-objective optimization models to generate effective plans, ensuring the feasibility and superiority of the optimization scheme.

[0023] 5. The design incorporates a dynamic priority-based safety anti-conflict and rapid response mechanism, greatly enhancing the system's operational safety and stability: Based on the real-time power margin at the grid connection point, the system dynamically enters three operating modes: "sufficient," "optimized," and "emergency." In emergency mode, a rapid reduction strategy based on dynamic priority is initiated. This mechanism effectively prevents the total system power from exceeding limits from both "prevention" (strict execution of the plan in optimized mode) and "emergency" (rapid intervention in emergency mode), ensuring grid safety. The rapid reduction strategy follows a progressive logic of "proportionally reducing charging pile power → shutting down adjustable loads according to priority," balancing response speed and differentiated handling of the impact on different loads. It can quickly pull the system back to a safe range in extreme situations, preventing system-wide power outages caused by protection device activation, demonstrating high reliability.

[0024] 6. Integrating rolling optimization and adaptive adjustment mechanisms for model parameters enables the system to continuously optimize and adapt over the long term: Through rolling optimization and adaptive adjustment, the system is equipped with the ability to dynamically respond to uncertainties and improve itself; the rolling optimization process is triggered when the deviation between actual and planned deviations exceeds the limit, and short-cycle optimization is re-executed with the latest data, which can promptly correct plan failures caused by prediction errors or sudden disturbances, ensuring real-time optimal control; the battery health status model and user habit model are updated regularly, and the internal parameters of the optimization model (such as weight coefficients) are adjusted accordingly, so that the optimization target can be dynamically adjusted with battery aging and changes in user behavior, avoiding the performance degradation of the model over time due to "fixation", and ensuring the long-term efficient operation of the system throughout its entire life cycle. Attached Figure Description

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] Figure 1 This is a flowchart of a power intelligent allocation and anti-collision method for a photovoltaic energy storage and charging system according to the present invention. Detailed Implementation

[0027] The overall concept of the technical solution in this application embodiment is as follows: An intelligent closed-loop management and control system of "prediction-optimization-execution-rolling-adaptation" is constructed. This system collects multi-source data in real time and uses advanced models for high-precision prediction. Based on this, an optimization model is established with multiple objectives: minimizing total system operating cost, maximizing user energy comfort, and minimizing the cost of energy storage battery lifespan degradation. An improved algorithm is used to solve for the globally optimal power allocation plan covering energy storage, charging piles, and schedulable loads. Real-time power safety is ensured by strictly executing the plan and dynamically switching operating modes (sufficient, optimized, emergency) based on the real-time power margin of the grid connection point. Rolling optimization triggered by actual deviations is introduced to address uncertainties. Furthermore, by regularly updating battery health and user habit models and adaptively adjusting optimization parameters, the system possesses the ability to continuously learn and self-optimize during operation. Thus, under strict power constraints, it achieves long-term, dynamic, and collaborative optimization of economy, comfort, and equipment health.

[0028] Please refer to Figure 1 As shown, a preferred embodiment of the intelligent power allocation and anti-collision method for a photovoltaic energy storage and charging system of the present invention includes the following steps: Step S1: Collect system operation data, environmental data, user data and market data of the photovoltaic-storage-charging system in real time, and predict the output power curve of the photovoltaic unit, the adjustable charging demand curve of the charging pile, the power demand curve of each schedulable load, and the life decay cost of the energy storage battery unit under different charging and discharging strategies within a preset period in the future. Step S2: Establish a multi-objective optimization model with a future preset time period as the optimization cycle. The multi-objective optimization model takes the lowest total system operating cost, the highest user energy comfort, and the lowest energy storage battery unit life decay cost as the comprehensive optimization objectives, and takes the following constraints as the total power of the grid connection point not exceeding the limit in real time, the charging and discharging power and state of charge of the energy storage battery unit not exceeding the limit, and meeting the basic requirements of charging piles and dispatchable loads. Step S3: Based on the output power curve, adjustable charging demand curve, power demand curve, and lifetime degradation cost, solve the multi-objective optimization model to obtain the optimized power allocation plan for energy storage battery units, charging piles, and various schedulable loads in the future preset time period, and then generate the power control command for the current control cycle. Step S4: Execute the power control command to control the charging and discharging power of the energy storage battery unit, the charging power of the charging pile, and the activation status and power of the schedulable load in real time. Step S5: During the control cycle, continuously monitor the deviation between the actual power and the optimized power allocation plan. If the deviation exceeds the set deviation threshold, trigger the rolling optimization process. Step S6: Regularly update the health status model and user habit model of the energy storage battery unit, and adaptively adjust the relevant parameters of the multi-objective optimization model.

[0029] Step S1 specifically involves: Real-time acquisition of system operation data, environmental data, user data, and market data of the photovoltaic energy storage and charging system; The system operation data, environmental data, user data, and market data are collected in real time by sensors, monitoring terminals, and communication modules (such as 4G / 5G, power line carrier, LoRa, etc.) deployed in various units of the photovoltaic-storage-charging system (photovoltaic inverters, energy storage converters, charging pile controllers, smart meters, smart home gateways, etc.), and aggregated to the central processing unit of the energy management system (EMS) on the local or cloud.

[0030] The system operation data includes: the actual power generation of the photovoltaic unit, the real-time state of charge, health parameters, internal resistance, temperature and historical cycle data of the energy storage battery unit, the real-time charging power, connection status and vehicle battery information of the charging pile, the real-time power and switching status of each schedulable load, and the total power limit and real-time total power of the system grid connection point. The environmental data includes: real-time meteorological data and weather forecast information for a future preset period; The user data includes: historical charging records of the charging pile, the user's expected vehicle retrieval time and target battery level, and historical usage records and user preference settings for each schedulable load. The market data includes: current and future time-of-use electricity prices and feed-in electricity prices; Based on the weather forecast information and the historical actual power generation of the photovoltaic unit, the output power curve of the photovoltaic unit is predicted within a future preset period. Based on the user data, the adjustable charging demand curve of the charging pile and the power demand curve of each schedulable load are predicted within a preset time period in the future. Based on the real-time state of charge, health parameters, internal resistance, temperature, and historical cycle data of the energy storage battery cell, the lifetime degradation cost under different charge and discharge strategies is predicted.

[0031] In step S1, the output power curve is predicted based on a photovoltaic power output prediction model. The photovoltaic power output prediction model takes irradiance, temperature and cloud cover from weather forecast information as input and is trained with the historical actual power generation of the photovoltaic unit and the corresponding historical meteorological data. The photovoltaic power output prediction model is a long short-term memory network model based on an attention mechanism, and its training process specifically includes: Time alignment and cleaning of historical actual power generation data with historical meteorological data were performed to construct a training dataset; The training dataset is input into the photovoltaic power output prediction model, and the attention mechanism automatically learns and weights the historical meteorological features and power change patterns most relevant to the prediction of a specific future period. The network parameters of the photovoltaic power output prediction model are optimized by backpropagation algorithm to minimize the error between the output power curve predicted by the photovoltaic power output prediction model and the historical actual power generation.

[0032] The attention-based Long Short-Term Memory (Attention-LSTM) model comprises a multi-layer LSTM network as an encoder to extract temporal features from historical meteorological sequences (irradiance, temperature, cloud cover) and power sequences; an attention layer to calculate the correlation weights between the features of the future prediction time and each historical time; and a fully connected layer as a decoder to output the photovoltaic output power curve for a predetermined future time period (e.g., the next 24 hours, at 15-minute intervals). During model training, historical aligned data from the past few months is used, with mean squared error (MSE) as the loss function, and training is performed using the Adam optimizer.

[0033] In step S1, the prediction process of the adjustable charging demand curve is specifically as follows: Based on the user data, obtain the user's expected vehicle retrieval time T_target, the vehicle's current battery charge SoC_current, and the total battery capacity C; Based on the maximum allowable charging power P_max_charge of the charging pile, calculate the shortest time T_min required to fully charge at constant power. The charging demand time window is defined as [T_target-T_min-ΔT_buffer,T_target], where ΔT_buffer is a time buffer amount preset based on the uncertainty of user behavior; Within the charging demand time window, a segmented charging power demand function is constructed with the charging completion time T_finish as the variable. The segmented charging power demand function satisfies that the cumulative charging amount from the current time to T_finish is equal to C*(1-SoC_current), and the charging power is continuously adjustable within the range of [0, P_max_charge] at any time.

[0034] The segmented charging power demand function can be specifically defined as follows: Within the charging demand time window [T_target-T_min-ΔT_buffer,T_target], a planned charging completion time T_finish (T_finish∈[current time,T_target]) is set. Then, the charging power P_charge(t) from the current time t_now to T_finish must satisfy the total charging amount constraint: ∫_{t_now}^{T_finish}P_charge(t)dt=C*(1-SoC_current).

[0035] A simple implementation is to model it as a function whose power is adjustable during the [t_now, T_finish] time period and zero during the [T_finish, T_target] time period. The optimization algorithm will dynamically allocate the power value of P_charge(t) within the time window, with the value ranging from [0, P_max_charge], provided that the total charging amount is satisfied.

[0036] In step S1, the prediction of the power demand curve specifically includes: For air conditioners and water heaters, a hybrid prediction model based on physical models and data-driven approaches is established: using a building thermodynamic model and combining future temperature and sunshine forecasts, the basic heat load is calculated; at the same time, a neural network is trained using historical user adjustment data to predict the random adjustment behavior of user-set temperatures, and the power demand curve is output by combining the basic heat load and random adjustment behavior. For washing machines and dishwashers, based on the historical operation time patterns of users, a hidden Markov model is used to predict their start-up probability distribution during future optimization periods, and then converted into a power demand curve. For air conditioners, the building thermodynamic model can employ an equivalent thermal parameter model, combined with building envelope parameters and indoor-outdoor temperature difference to predict the base load. Simultaneously, a neural network (such as a fully connected network) is used to learn the relationship between historical user temperature setpoint changes and factors such as time and outdoor temperature, predicting power fluctuations caused by random adjustments. These two methods are then superimposed to generate the final power demand probability distribution or curve. For washing machines, the Hidden Markov Model (HMM) states can be defined as "not started," "washing," "rinsing," "spinning," etc. Based on the user's historical start-up time (e.g., higher probability of start-up at 8 PM), the probability of the machine being in each state in the future is predicted, and combined with its rated power, converted into a desired power demand curve.

[0037] The specific process for predicting the lifetime degradation cost is as follows: Based on the real-time state of charge, health parameters, internal resistance, temperature and historical cycle data of the energy storage battery cell, the current health status of the energy storage battery cell is calculated. Based on the electrochemical age model, the equivalent influence factor K_degrade of different charge-discharge depths and average charge-discharge rates on the battery cycle life degradation under the current health state is calculated. Quantify the cost of lifespan decay as follows: Lifetime degradation cost = K_degrade * |P_bss(t)| * Δt * C_battery; Where P_bss(t) is the charging and discharging power of the energy storage battery unit at time t; Δt is the control period; and C_battery is the purchase cost conversion factor corresponding to the unit battery capacity.

[0038] The electrochemical age model can be a degradation model based on the Arrhenius and Peukert equations. The equivalent influencing factor K_degrade can be expressed as a function of depth of charge / discharge (DoD) and average charge / discharge rate (C-rate), for example, K_degrade=f(DoD)*g(C-rate), where the specific parameters of the functions f and g are obtained by fitting life test data or historical operating data provided by the battery manufacturer. C_battery is a coefficient obtained by dividing the purchase cost per unit capacity battery by the estimated total cycle life, used to quantify life degradation into economic cost.

[0039] In summary, step S1 constructs a multi-dimensional, high-precision, and scheduling-oriented prediction system. Unlike existing technologies that only focus on instantaneous power or use simple prediction models, this invention employs customized advanced prediction models tailored to the different physical characteristics and behavioral patterns of the source (photovoltaics), load (charging piles, home appliances), and storage (batteries) in a photovoltaic-storage-charging system. For example, for photovoltaic output, which has strong weather dependence and time-series correlation, an attention-based LSTM model is used to capture long-range dependence; for electric vehicle charging, which has clear user time preferences and rigid energy demands, an adjustable charging demand curve within a flexible time window is constructed, providing crucial decision-making freedom for subsequent optimization; and for home appliance loads, which are influenced by both physical laws and user habits, a hybrid model driven by physics and data is used. In particular, the invention innovatively quantifies the life-cycle degradation cost of energy storage batteries as a predictable variable and incorporates it into the optimization framework, thereby shifting the health management of the entire equipment lifecycle from post-event assessment to in-process decision-making factors. This lays a precise and reliable data foundation for multi-objective collaborative optimization, which is a key prerequisite for achieving global optimization in this invention.

[0040] In step S2, the formula for calculating the total operating cost of the system is: Total system operating cost = Electricity purchase cost - Electricity sales revenue + Demand response penalty cost + Lifetime degradation cost; Among them, the cost of purchasing electricity is calculated based on the time-of-use price and the power absorbed from the grid; the revenue from selling electricity is calculated based on the grid connection price and the power fed back to the grid; the demand response penalty cost is activated when the total demand power of the system exceeds the demand response threshold agreed with the grid. The user's energy comfort level is quantitatively represented by constructing a comprehensive dissatisfaction index D_unsat: D_unsat=Σ_i(w_i*(ΔE_i / E_i_nom)^2)+ w_ev*((T_finish_ev-T_target_ev) / T_target_ev)^2; Where i represents each schedulable load; E_i_nom represents the rated energy consumption of the schedulable load; ΔE_i represents the deviation between the actual energy consumption of the schedulable load and E_i_nom within an optimization cycle; w_i represents the weighting coefficient; T_finish_ev represents the actual charging completion time of the electric vehicle; T_target_ev represents the user's expected vehicle pickup time; and w_ev represents the weighting coefficient for charging delay.

[0041] The core innovation of the multi-objective optimization model established in step S2 lies in breaking away from the traditional single-dimensional perspective that focuses solely on instantaneous power balance (anti-conflict). By simultaneously considering the minimum total system operating cost, the maximum user energy comfort, and the minimum lifespan degradation cost of energy storage battery units as comprehensive optimization objectives, it achieves multi-dimensional dynamic trade-offs under strict power safety constraints (hard constraints such as grid connection power and battery state). The total system operating cost includes the benefits and penalties of interaction with the grid, achieving linkage with the external electricity market. The comprehensive dissatisfaction index D_unsat, for the first time, unifies and quantifies the energy consumption deviation of different loads (such as air conditioners and water heaters) and the charging delay of electric vehicles, making the fuzzy comfort a concrete optimizable objective. The introduction of lifespan degradation cost makes the long-term implicit cost of battery health explicit. This model systematically integrates the four major operational objectives of anti-conflict (safety constraints), economy, comfort, and equipment durability, aiming to find a globally Pareto optimal solution, fundamentally solving the problems mentioned in the background technology of existing schemes having a single decision-making dimension and neglecting long-term economy and equipment health.

[0042] In step S3, the solution of the multi-objective optimization model is obtained using an improved non-dominated sorting genetic algorithm, specifically including: The optimization variables are encoded, including the power of the energy storage battery unit in each control time slot, the power of the charging pile in each control time slot, and the start-stop time sequence of each schedulable load. Initialize the population and introduce initial individuals generated based on historical best solutions and heuristic rules to improve the quality of the initial population; In each generation of evolution, the total system operating cost, user energy comfort, and lifespan degradation cost of individual energy storage battery units are calculated, and fast non-dominated sorting and congestion calculation are performed. The offspring population is generated using tournament selection, simulated binary crossover, and polynomial mutation operators. An elite retention strategy is introduced, merging the parent and offspring populations, and selecting the next generation of population based on non-dominance level and crowding. After reaching the maximum number of iterations, the final optimal compromise solution is determined from the first non-dominated front solution set, based on the preset target weight or the compromise point selected by user interaction, as the optimized power allocation plan.

[0043] The improved Non-Dominated Sorting Genetic Algorithm (NSGA-II) is chosen to solve the multi-objective optimization model for the following reasons: This optimization problem is characterized by multiple objectives, nonlinearity, complex constraints, and a large solution space, which traditional optimization algorithms struggle to handle effectively. NSGA-II is renowned for its ability to efficiently obtain uniformly distributed Pareto fronts in multi-objective problems. The improvements in this invention are: 1) Population initialization optimization: In addition to random initialization, it introduces initial individuals generated based on historical best solutions and heuristic rules (such as prioritizing charging during off-peak electricity prices), significantly improving the quality of initial solutions and the algorithm's convergence speed; 2) Elite retention strategy: This ensures that excellent individuals are not lost, guaranteeing the convergence of iterations. Finally, the optimal compromise solution is selected from the first non-dominated front based on preset weights or user preferences. This demonstrates the flexibility of this invention—it can either automatically execute a preset balancing strategy (such as prioritizing economy) or interact with the user (e.g., the user selects an economy-first or comfort-first mode on a mobile app), thereby generating an optimized power allocation plan and short-cycle control instructions that take into account the interests of all parties.

[0044] Regarding the construction and solution of multi-objective optimization models: Handling Multi-Objective Problems: The multi-objective optimization model comprises three objectives: minimizing the total system operating cost F1, minimizing user dissatisfaction D_unsat(F2), and minimizing the energy storage lifetime degradation cost F3. In practical solutions, this can be transformed into a solvable optimization problem using the weighted summation method, the ε-constraint method, or by directly obtaining the Pareto front. For example, in the weighted summation method, a single-objective function is constructed: Minimize: w1*F1 + w2*F2 + w3*F3, where w1, w2, and w3 are preset weight coefficients that satisfy w1 + w2 + w3 = 1. The weight coefficients reflect different preferences for economy, comfort, and equipment lifespan, and can be preset by the system or selected by the user.

[0045] Specification of constraints: The real-time non-exceeding limit constraint of the total power at the grid connection point can be expressed as: |P_grid(t)|≤P_grid_max; Where P_grid(t) represents the real-time total power at the grid-connected points of the system; P_grid_max represents the total power limit at the grid-connected points of the system; P_grid(t) = P_pv(t) + P_bss(t) - P_charge(t) - P_load(t); P_pv(t) represents the actual power generation of the photovoltaic unit at time t; P_bss(t) represents the charging and discharging power of the energy storage battery unit at time t. Positive values ​​indicate discharging (supplying power to the system), and negative values ​​indicate charging (absorbing power from the system); P_charge(t) represents the total charging power of all electric vehicle charging piles at time t; P_load(t) represents the total power of all unschedulable conventional loads in the system at time t.

[0046] Energy storage battery constraints include: SOC_min≤SOC(t)≤SOC_max; -P_bss_discharge_max≤P_bss(t)≤P_bss_charge_max; SOC(t+1)=SOC(t)+(η_charge*P_bss_charge(t)-P_bss_discharge(t) / η_discharge)*Δt / Capacity; Where η represents efficiency; SOC(t) represents the state of charge of the energy storage battery cell at time t, i.e., the percentage of remaining capacity; SOC_min and SOC_max are the upper and lower limits of the allowed state of charge of the energy storage battery set by the system, respectively, to prevent overcharging or over-discharging and protect battery health; P_bss_discharge_max represents the maximum allowable discharge power (positive value) of the energy storage battery cell; P_bss_charge_max represents the maximum allowable charging power (positive value) of the energy storage battery cell. Therefore, the value range of P_bss(t) is from -P_bss_charge_max (maximum charging power) to P_bss_discharge_max (maximum discharge power). The values ​​are: SOC(t+1) represents the battery state of charge at the next moment (t+1); η_charge and η_discharge represent the charging efficiency and discharging efficiency of the energy storage battery (0<η<1), used to quantify energy loss during charging and discharging; P_bss_charge(t) represents the charging power of the energy storage battery at time t (the absolute value when P_bss(t)<0); P_bss_discharge(t) represents the discharging power of the energy storage battery at time t (the part when P_bss(t)>0); Δt represents the time length of a control time slot in the optimization model; Capacity represents the rated capacity of the energy storage battery cell (unit: kWh).

[0047] Implementation details of the improved Non-Dominated Sorting Genetic Algorithm (NSGA-II): Optimization variable encoding can be done using real numbers. For energy storage and charging pile power, the power value of each control time slot (e.g., 96 time slots corresponding to 24 hours) is treated as a gene. For the start and stop of schedulable loads, their on / off states in each time slot can be represented by a 0-1 sequence. Heuristic rules introduced during population initialization may include: prioritizing charging energy storage during off-peak electricity price periods; prioritizing the use of photovoltaic power to supply loads and charge electric vehicles during peak photovoltaic output periods; and staggering the operating times of high-power loads (such as water heaters). The solutions generated by these heuristic rules, together with randomly generated solutions, constitute the initial population to accelerate convergence.

[0048] In step S4, different operating modes are dynamically entered based on the real-time total power margin of the system's grid connection points, including: Ample Mode: When the real-time total power of the system grid connection point is detected to be lower than 60% of the total power limit, the charging piles and dispatchable loads operate at the maximum power required by the user, and the surplus photovoltaic power and / or grid power charge the energy storage battery unit. The charging power adopts the planned value of the optimized power allocation plan or the current maximum value. Optimization mode: When the real-time total power of the system's grid connection point is between 60% and 95% of the total power limit, the power control command of the current control cycle is strictly executed; Emergency Mode: When the real-time total power of the system's grid connection points exceeds 95% of the total power limit, the power control command is immediately interrupted, and a rapid reduction strategy based on dynamic priority is initiated: First, the power of all charging piles is reduced to a safe level proportionally according to a preset ratio. If the limit is still exceeded, the power is sequentially delayed and shut down according to the real-time interruptibility level of the schedulable load, and an alarm is sent to the user. After the real-time total power recovers to below 85% of the total power limit and stabilizes for more than a preset time, the system automatically switches back to optimization mode.

[0049] Step S4 does not mechanically execute the optimized power allocation plan, but introduces a dynamic mode switching and anti-collision mechanism based on real-time power margin, constituting the real-time security layer of this invention. This mechanism is designed as a three-level response system: 1. Ample Power Mode (>40% margin): When the power is sufficient, it fully meets the user's needs and charges the energy storage, demonstrating the system's user-friendliness.

[0050] 2. Optimization mode (5%-40% margin): This is the main operating state of the system, which strictly implements the optimized power allocation plan and achieves multi-objective collaboration. This is the core optimization layer of this invention.

[0051] 3. Emergency Mode (<5% Margin): This is the system's safety protection layer. Once a momentary risk of power exceeding the limit (reaching 95% of the limit) is detected, the optimization plan is immediately interrupted, and a rapid load reduction strategy based on dynamic priority is initiated. This strategy adopts a progressive logic of first adjusting voltage, then cutting off loads: first, the power of all charging piles is reduced proportionally (the impact is relatively small and recoverable); if the limit is still exceeded, then the scheduleable loads are shut down with a delay according to priority. This design ensures millisecond-level rapid response capability, preventing over-limit tripping, while minimizing the impact on users' critical loads, fundamentally different from the real-time monitoring-based priority power-off scheme (brutal disconnection) mentioned in the background technology. After the power recovers to below 85% of the limit and stabilizes, the system automatically switches back to optimization mode, resuming intelligent optimization operation, thus achieving intelligent coordination and seamless switching between safety protection and optimization operation.

[0052] In step S5, the rolling optimization process specifically includes: Record the deviation between the current actual power and the optimized power allocation plan, including photovoltaic output deviation ΔP_pv and load demand deviation ΔP_load; Starting from the current moment, the prediction step is re-executed with a rolling optimization window shorter than the optimization period. Within the rolling optimization window, the planned trajectory of the state of charge of the energy storage battery cells is corrected to ensure that it returns to near the target trajectory of the optimized power allocation plan at the end of the rolling optimization window.

[0053] The deviation threshold can be set according to the power level, for example, 5% or 10% of the rated power. When the photovoltaic output deviation ΔP_pv or the load demand deviation ΔP_load continuously exceeds the threshold for several control cycles (e.g., 3 cycles), rolling optimization is triggered. The length of the rolling optimization window (e.g., the next 4 hours) is usually much smaller than the global optimization cycle (e.g., 24 hours). In rolling optimization, the current actual energy storage SOC value is used as the initial state, and the optimization problem with a shortened time window is resolved. However, constraints are imposed on the SOC state at the end of the optimization period (when the rolling window ends) to make it approach the target SOC value at that moment in the original long-term optimization plan, thereby achieving trajectory correction.

[0054] In step S6, the adaptive adjustment specifically includes: Based on the complete charge-discharge cycle data of the energy storage battery cell within a preset period, the relationship between its capacity decay and cumulative throughput and average depth of discharge is fitted, and the model parameters of the electrochemical age model are updated online. Based on user feedback ratings regarding past charging completion time delays, dynamically adjust the weighting coefficient w_ev within the overall dissatisfaction index D_unsat; Based on the Pareto front distribution of historical optimization results, dynamically adjust the weighting coefficients when transforming a multi-objective optimization problem into a single-objective problem in a multi-objective optimization model, or adjust the preference rules for selecting the optimal compromise solution.

[0055] The battery health status model can be updated monthly or quarterly, using complete cyclical data within that period to refit the parameters of the lifespan degradation model. User habit models can be continuously updated. For example, the system records the actual completion time of each electric vehicle charge and the user's expected time. When a user rates the charging delay (e.g., 1-5 stars) via the app, the system dynamically adjusts the weighting coefficient w_ev in the comprehensive dissatisfaction index D_unsat based on the recent (e.g., the average of the last 10 times) ratings (the lower the rating, the higher the w_ev value, indicating greater emphasis on timeliness). Adjustments to optimization weights or preference rules can be based on the distribution of historical Pareto front solutions. If the historical solution set is too concentrated on one objective (e.g., cost) while performing poorly on another objective (e.g., comfort), the weighting coefficients can be automatically fine-tuned to promote a more balanced optimization direction.

[0056] Step S5 (rolling optimization) and step S6 (adaptive adjustment) together constitute the self-correction and self-evolution layer of the present invention, ensuring the robustness, real-time performance and continuous optimization of the system in long-term operation.

[0057] The rolling optimization in step S5 is a feedback correction mechanism for short-term uncertainties. Since the prediction in step S1 inevitably contains errors, or sudden disturbances (such as abrupt weather changes or a sudden increase in load) may occur, the actual power will deviate from the optimized plan. When the deviation exceeds a threshold, the prediction and optimization are immediately re-executed within a shorter rolling window, starting from the current moment. Crucially, the rolling optimization corrects the planned trajectory of the energy storage battery's state of charge (SOC), ensuring it returns to near its long-term optimal trajectory by the end of the window. This is equivalent to adding a closed-loop feedback to the system, enabling the optimization plan to 'keep pace with the times,' dynamically correcting deviations and avoiding control performance degradation or safety risks caused by model mismatch.

[0058] The adaptive adjustment in step S6 is a self-learning evolutionary mechanism designed for long-term changes. The system periodically updates two core models using operational data: 1) Battery health status model: By fitting battery aging parameters with actual charge and discharge data, the prediction of lifespan degradation costs is made closer to the current actual state of the battery; 2) User habit model: Based on user feedback on charging delays, etc., the weights in the optimization objectives (such as w_ev in D_unsat) are dynamically adjusted. This means that the system's optimization objectives are not static, but dynamically adjusted as the battery ages and user preferences change. For example, when battery health declines, the system will automatically prioritize battery lifespan in optimization; when users show greater sensitivity to charging punctuality, the system will correspondingly increase the weight of comfort objectives.

[0059] In summary, the combination of rolling optimization (short-term feedback) and adaptive adjustment (long-term learning) enables this invention to evolve from a static optimization program into an intelligent system with continuous learning, self-updating, and dynamic adaptation capabilities. This allows it to maintain an efficient, safe, and economical optimal operating state over the long term, overcoming the fundamental defects of existing technologies such as rigid strategies and lack of adaptability. Assume a residential photovoltaic-storage-charging system, comprising 20kW photovoltaic power, 30kWh / 10kW energy storage, two 7kW AC charging piles, and adjustable loads such as air conditioners and water heaters. The grid connection point power limit is 15kW.

[0060] Step S1: The system starts at a fixed time each day (e.g., 00:00). It collects weather forecasts, time-of-use electricity prices, user-defined demand for vehicle pickup and full charging by 8 PM the following day, historical usage data for each load, and the battery's current SOC (State of Charge) of 50%. Using a predictive model, it generates a 24-hour photovoltaic output curve, an adjustable charging demand curve, and power demand curves for each load. Based on the current battery SOH (health state of 95%), it calculates the lifespan degradation cost of different charging and discharging strategies.

[0061] Steps S2 & S3: Establish an optimization model with 96 time slots, using the next 24 hours as a cycle and 15-minute intervals. The goal is to minimize total electricity costs, user dissatisfaction, and battery lifespan degradation costs. Constraints include a total power output not exceeding 15kW and battery SOC maintained between 20% and 90%. An improved NSGA-II algorithm is used to solve the problem, yielding a Pareto optimal solution set. The system selects a compromise solution from this set based on a preset strategy (prioritizing economy) and generates the first 15-minute power control command for the following day: Photovoltaics generate power at full capacity, energy storage charges at 5kW, charging pile A charges at 3kW, charging pile B is suspended, and air conditioning is set to energy-saving mode.

[0062] Step S4: The system enters optimization mode to execute the command. At 1 PM, the photovoltaic output reaches its peak, and the total power is close to but does not exceed the limit. The system strictly allocates power according to the optimization plan. At 6 PM, due to a sudden change in weather, the actual photovoltaic output is lower than predicted, and all loads start simultaneously. The total power is detected to instantly reach 14.5kW (>15kW 95%), and the system immediately switches to emergency mode. First, the power of the two charging piles is proportionally reduced from 7kW and 5kW to 3.5kW and 2.5kW respectively. The power is reduced to 13.5kW, restoring to a safe level. The system sends an alarm: "Power too high detected, charging power temporarily reduced." After the power stabilizes, it automatically switches back to optimization mode*.

[0063] Step S5: Due to the afternoon photovoltaic forecast deviation, rolling optimization was triggered. The system re-predicts and optimizes for the next 4 hours, starting from the current time (6:15 pm), and ensures that the battery SOC can be adjusted back to near the SOC value of that time in the original long-term plan by 10:15 pm.

[0064] Step S6: Monthly, the system analyzes battery cycle data and updates the lifespan degradation model parameters. Based on user ratings of charging completion time over the past month, w_ev is automatically adjusted from 0.5 to 0.7, indicating a greater focus on charging timeliness in subsequent optimizations.

[0065] In summary, the advantages of this invention are as follows: 1. Enhance the system's multi-objective dynamic collaborative optimization capability under power security constraints by constructing a closed-loop decision-making framework of "prediction-optimization-execution-monitoring-adaptation": First, predict photovoltaic and flexible load demand and battery life degradation costs based on multi-dimensional data. Then, with the comprehensive goal of minimizing the total system operating cost, user comfort, and battery life degradation costs, establish and solve a global optimization model covering energy storage, charging piles, and all schedulable loads to generate and execute a collaborative power allocation plan. Ensure real-time power security by introducing rolling optimization triggered by actual deviations and a working mode that dynamically switches according to the power margin of the grid connection point (including emergency rapid reduction strategy). At the same time, regularly update the battery health and user habit models and adaptively adjust optimization parameters so that the entire system can continuously and dynamically balance multiple objectives such as economy, experience, and equipment health while meeting strict power constraints.

[0066] 2. Establish a fully intelligent closed-loop management and control architecture of "prediction-optimization-control-rolling-self-learning" with strong systematicity and adaptability: Construct a complete intelligent management and control closed loop, from multi-source data collection and high-precision prediction, to optimization decision-making based on multi-objective models, to precise execution and dynamic mode switching, and finally to deviation correction through rolling optimization and self-evolution through regular model updates. This closed-loop architecture ensures that the system can not only make optimal decisions based on the current state, but also cope with uncertainties in actual operation (rolling optimization), and continuously learn from historical operating data to optimize its own model parameters (adaptive adjustment), so that the system can maintain a high-efficiency and reliable operating state in the long term, significantly improving the robustness and intelligence level of the overall solution.

[0067] 3. A multi-objective optimization model that comprehensively considers economy, comfort, and equipment lifespan is proposed to achieve refined coordination and optimal balance of benefits from multiple parties: "lowest total system operating cost", "highest user energy comfort" and "minimum cost of energy storage battery cell lifespan degradation" are taken as optimization objectives. By constructing a precise quantitative model (such as using the comprehensive dissatisfaction index D_unsat to quantify comfort), the multi-objective problem is transformed into a solvable optimization problem. Under the premise of meeting the hard constraints of the power grid and equipment, Pareto optimal solutions are found in multiple dimensions such as reducing user electricity costs, ensuring user electricity / vehicle experience, and extending the lifespan of expensive energy storage equipment. This achieves synergistic optimization of multiple stakeholders, and the comprehensive benefits far exceed those of single-objective optimization, making it more commercially valuable.

[0068] 4. Advanced, fine-grained, and targeted forecasting methods are employed to provide a highly reliable data foundation for optimized decision-making: Advanced forecasting models are used for different objects, significantly improving forecast accuracy; for example, an attention-based LSTM network is used to predict photovoltaic power, which can effectively capture the complex spatiotemporal correlation between meteorological sequences and power generation; an "adjustable charging demand curve" and flexible time window are constructed for electric vehicle charging demand, which respects users' vehicle pickup time and provides flexible space for power scheduling; a hybrid "physical + data-driven" forecasting model is established for loads such as air conditioners and water heaters, taking into account both physical laws and the randomness of user behavior; these high-precision forecasting results are the prerequisite for multi-objective optimization models to generate effective plans, ensuring the feasibility and superiority of the optimization scheme.

[0069] 5. The design incorporates a dynamic priority-based safety anti-conflict and rapid response mechanism, greatly enhancing the system's operational safety and stability: Based on the real-time power margin at the grid connection point, the system dynamically enters three operating modes: "sufficient," "optimized," and "emergency." In emergency mode, a rapid reduction strategy based on dynamic priority is initiated. This mechanism effectively prevents the total system power from exceeding limits from both "prevention" (strict execution of the plan in optimized mode) and "emergency" (rapid intervention in emergency mode), ensuring grid safety. The rapid reduction strategy follows a progressive logic of "proportionally reducing charging pile power → shutting down adjustable loads according to priority," balancing response speed and differentiated handling of the impact on different loads. It can quickly pull the system back to a safe range in extreme situations, preventing system-wide power outages caused by protection device activation, demonstrating high reliability.

[0070] 6. Integrating rolling optimization and adaptive adjustment mechanisms for model parameters enables the system to continuously optimize and adapt over the long term: Through rolling optimization and adaptive adjustment, the system is equipped with the ability to dynamically respond to uncertainties and improve itself; the rolling optimization process is triggered when the deviation between actual and planned deviations exceeds the limit, and short-cycle optimization is re-executed with the latest data, which can promptly correct plan failures caused by prediction errors or sudden disturbances, ensuring real-time optimal control; the battery health status model and user habit model are updated regularly, and the internal parameters of the optimization model (such as weight coefficients) are adjusted accordingly, so that the optimization target can be dynamically adjusted with battery aging and changes in user behavior, avoiding the performance degradation of the model over time due to "fixation", and ensuring the long-term efficient operation of the system throughout its entire life cycle.

[0071] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A power intelligent distribution and anti-collision method for optical storage and charging system, characterized in that: The steps include the following: Step S1: Collect system operation data, environmental data, user data and market data of the photovoltaic-storage-charging system in real time, and predict the output power curve of the photovoltaic unit, the adjustable charging demand curve of the charging pile, the power demand curve of each schedulable load, and the life decay cost of the energy storage battery unit under different charging and discharging strategies within a preset period in the future. Step S2: Establish a multi-objective optimization model with a future preset time period as the optimization cycle. The multi-objective optimization model takes the lowest total system operating cost, the highest user energy comfort, and the lowest energy storage battery unit life decay cost as the comprehensive optimization objectives, and takes the following constraints as the total power of the grid connection point not exceeding the limit in real time, the charging and discharging power and state of charge of the energy storage battery unit not exceeding the limit, and meeting the basic requirements of charging piles and dispatchable loads. Step S3: Based on the output power curve, adjustable charging demand curve, power demand curve, and lifetime degradation cost, solve the multi-objective optimization model to obtain the optimized power allocation plan for energy storage battery units, charging piles, and various schedulable loads in the future preset time period, and then generate the power control command for the current control cycle. Step S4: Execute the power control command to control the charging and discharging power of the energy storage battery unit, the charging power of the charging pile, and the activation status and power of the schedulable load in real time. Step S5: During the control cycle, continuously monitor the deviation between the actual power and the optimized power allocation plan. If the deviation exceeds the set deviation threshold, trigger the rolling optimization process. Step S6: Regularly update the health status model and user habit model of the energy storage battery unit, and adaptively adjust the relevant parameters of the multi-objective optimization model.

2. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, wherein: Step S1 specifically involves: Real-time acquisition of system operation data, environmental data, user data, and market data of the photovoltaic energy storage and charging system; The system operation data includes: the actual power generation of the photovoltaic unit, the real-time state of charge, health parameters, internal resistance, temperature and historical cycle data of the energy storage battery unit, the real-time charging power, connection status and vehicle battery information of the charging pile, the real-time power and switching status of each schedulable load, and the total power limit and real-time total power of the system grid connection point. The environmental data includes: real-time meteorological data and weather forecast information for a future preset period; The user data includes: historical charging records of the charging pile, the user's expected vehicle retrieval time and target battery level, and historical usage records and user preference settings for each schedulable load. The market data includes: current and future time-of-use electricity prices and feed-in electricity prices; Based on the weather forecast information and the historical actual power generation of the photovoltaic unit, the output power curve of the photovoltaic unit is predicted within a future preset period. Based on the user data, the adjustable charging demand curve of the charging pile and the power demand curve of each schedulable load are predicted within a preset time period in the future. Based on the real-time state of charge, health parameters, internal resistance, temperature, and historical cycle data of the energy storage battery cell, the lifetime degradation cost under different charge and discharge strategies is predicted.

3. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, characterized by: In step S1, the output power curve is predicted based on a photovoltaic power output prediction model. The photovoltaic power output prediction model takes irradiance, temperature and cloud cover from weather forecast information as input and is trained with the historical actual power generation of the photovoltaic unit and the corresponding historical meteorological data. The photovoltaic power output prediction model is a long short-term memory network model based on an attention mechanism, and its training process specifically includes: Time alignment and cleaning of historical actual power generation data with historical meteorological data were performed to construct a training dataset; The training dataset is input into the photovoltaic power output prediction model, and the attention mechanism automatically learns and weights the historical meteorological features and power change patterns most relevant to the prediction of a specific future period. The network parameters of the photovoltaic power output prediction model are optimized by backpropagation algorithm to minimize the error between the output power curve predicted by the photovoltaic power output prediction model and the historical actual power generation.

4. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, wherein: In step S1, the prediction process of the adjustable charging demand curve is specifically as follows: Based on the user data, obtain the user's expected vehicle retrieval time T_target, the vehicle's current battery charge SoC_current, and the total battery capacity C; Based on the maximum allowable charging power P_max_charge of the charging pile, calculate the shortest time T_min required to fully charge at constant power. The charging demand time window is defined as [T_target-T_min-ΔT_buffer,T_target], where ΔT_buffer is a time buffer amount preset based on the uncertainty of user behavior; Within the charging demand time window, a segmented charging power demand function is constructed with the charging completion time T_finish as the variable. The segmented charging power demand function satisfies that the cumulative charging amount from the current time to T_finish is equal to C*(1-SoC_current), and the charging power is continuously adjustable within the range of [0, P_max_charge] at any time.

5. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, characterized in that: In step S1, the prediction of the power demand curve specifically includes: For air conditioners and water heaters, a hybrid prediction model based on physical models and data-driven approaches is established: using a building thermodynamic model and combining future temperature and sunshine forecasts, the basic heat load is calculated; at the same time, a neural network is trained using historical user adjustment data to predict the random adjustment behavior of user-set temperatures, and the power demand curve is output by combining the basic heat load and random adjustment behavior. For washing machines and dishwashers, based on the historical operation time patterns of users, a hidden Markov model is used to predict their start-up probability distribution during future optimization periods, and then converted into a power demand curve. The specific process for predicting the lifetime degradation cost is as follows: Based on the real-time state of charge, health parameters, internal resistance, temperature and historical cycle data of the energy storage battery cell, the current health status of the energy storage battery cell is calculated. Based on the electrochemical age model, the equivalent influence factor K_degrade of different charge-discharge depths and average charge-discharge rates on the battery cycle life degradation under the current health state is calculated. Quantify the cost of lifespan decay as follows: Lifetime degradation cost = K_degrade * |P_bss(t)| * Δt * C_battery; Where P_bss(t) is the charging and discharging power of the energy storage battery unit at time t; Δt is the control period; and C_battery is the purchase cost conversion factor corresponding to the unit battery capacity.

6. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, wherein: In step S2, the formula for calculating the total operating cost of the system is: Total system operating cost = Electricity purchase cost - Electricity sales revenue + Demand response penalty cost + Lifetime degradation cost; Among them, the cost of purchasing electricity is calculated based on the time-of-use price and the power absorbed from the grid; the revenue from selling electricity is calculated based on the grid connection price and the power fed back to the grid; the demand response penalty cost is activated when the total demand power of the system exceeds the demand response threshold agreed with the grid. The user's energy comfort level is quantitatively represented by constructing a comprehensive dissatisfaction index D_unsat: D_unsat=Σ_i(w_i*(ΔE_i / E_i_nom)^2)+ w_ev*((T_finish_ev-T_target_ev) / T_target_ev)^2; Where i represents each schedulable load; E_i_nom represents the rated energy consumption of the schedulable load; ΔE_i represents the deviation between the actual energy consumption of the schedulable load and E_i_nom within an optimization cycle; w_i represents the weighting coefficient; T_finish_ev represents the actual charging completion time of the electric vehicle; T_target_ev represents the user's expected vehicle pickup time; and w_ev represents the weighting coefficient for charging delay.

7. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, wherein: In step S3, the solution to the multi-objective optimization model employs an improved non-dominated sorting genetic algorithm, specifically including: The optimization variables are encoded, including the power of the energy storage battery unit in each control time slot, the power of the charging pile in each control time slot, and the start-stop time sequence of each schedulable load. Initialize the population and introduce initial individuals generated based on historical best solutions and heuristic rules to improve the quality of the initial population; In each generation of evolution, the total system operating cost, user energy comfort, and lifespan degradation cost of individual energy storage battery units are calculated, and fast non-dominated sorting and congestion calculation are performed. The offspring population is generated using tournament selection, simulated binary crossover, and polynomial mutation operators. An elite retention strategy is introduced, merging the parent and offspring populations, and selecting the next generation of population based on non-dominance level and crowding. After reaching the maximum number of iterations, the final optimal compromise solution is determined from the first non-dominated front solution set, based on the preset target weight or the compromise point selected by user interaction, as the optimized power allocation plan.

8. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, wherein: In step S4, different operating modes are dynamically entered based on the real-time total power margin of the system's grid connection points, including: Ample Mode: When the real-time total power of the system grid connection point is detected to be lower than 60% of the total power limit, the charging piles and dispatchable loads operate at the maximum power required by the user, and the surplus photovoltaic power and / or grid power charge the energy storage battery unit. The charging power adopts the planned value of the optimized power allocation plan or the current maximum value. Optimization mode: When the real-time total power of the system's grid connection point is between 60% and 95% of the total power limit, the power control command of the current control cycle is strictly executed; Emergency Mode: When the real-time total power of the system's grid connection points exceeds 95% of the total power limit, the power control command is immediately interrupted, and a rapid reduction strategy based on dynamic priority is initiated: First, the power of all charging piles is reduced to a safe level proportionally according to a preset ratio. If the limit is still exceeded, the power is sequentially delayed and shut down according to the real-time interruptibility level of the schedulable load, and an alarm is sent to the user. After the real-time total power recovers to below 85% of the total power limit and stabilizes for more than a preset time, the system automatically switches back to optimization mode.

9. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, characterized in that: In step S5, the rolling optimization process specifically includes: Record the deviation between the current actual power and the optimized power allocation plan, including photovoltaic output deviation ΔP_pv and load demand deviation ΔP_load; Starting from the current moment, the prediction step is re-executed with a rolling optimization window shorter than the optimization period; Within the rolling optimization window, the planned trajectory of the state of charge of the energy storage battery cells is corrected to ensure that it returns to near the target trajectory of the optimized power allocation plan at the end of the rolling optimization window.

10. A power intelligent distribution and anti-collision method for optical storage and charging system as claimed in claim 1, characterized in that: In step S6, the adaptive adjustment specifically includes: Based on the complete charge-discharge cycle data of the energy storage battery cell within a preset period, the relationship between its capacity decay and cumulative throughput and average depth of discharge is fitted, and the model parameters of the electrochemical age model are updated online. Based on user feedback ratings regarding past charging completion time delays, dynamically adjust the weighting coefficient w_ev within the overall dissatisfaction index D_unsat; Based on the Pareto front distribution of historical optimization results, dynamically adjust the weighting coefficients when transforming a multi-objective optimization problem into a single-objective problem in a multi-objective optimization model, or adjust the preference rules for selecting the optimal compromise solution.

Images