Energy management and intelligent distribution method and system for optical storage and charging integrated system

Abstract

The invention discloses an energy management and intelligent distribution method and system for an optical storage and charging integrated system. The method comprises the following steps: collecting whole system data from equipment of a perception execution layer in real time through an edge energy manager in an edge calculation layer; the cloud platform layer performs photovoltaic and load ultra-short-term power prediction based on an AI ultra-short-term prediction model, and generates a prediction curve of photovoltaic power generation power and load demand in a future period of time; dynamically calibrating the digital twin model in the cloud platform layer based on real-time full-system data; in the calibrated digital twin model, target optimization is carried out in a rolling time window mode, and a multi-target dynamic permission decision mechanism is carried out; generating an optimal power distribution scheme; after optimization solution, outputting an optimal power instruction set in a time window in the future; and according to the optimal power instruction set, performing execution and scheduling based on an elastic power pool scheduler. Therefore, the system has strong environment adaptive capability, dynamically tracks the optimal operation point, and maximizes the utilization rate of system resources.

Description

Technical Field

[0001] This invention relates to an energy management and intelligent allocation method and system, and more particularly to an energy management and intelligent allocation method and system for an integrated photovoltaic-storage-charging system. Background Technology

[0002] With the advancement of the "dual carbon" target and the deepening of power market reform, the deployment of integrated "photovoltaic + energy storage + charging pile" systems in the industrial and commercial sectors has become the mainstream solution for reducing electricity costs, improving energy stability, and achieving green transformation.

[0003] However, traditional, isolated control strategies struggle to coordinate multiple conflicting objectives, specifically due to the following shortcomings: 1. The strategy is rigid and has poor adaptability; it relies only on preset and fixed thresholds and fixed peak and valley charging and discharging time periods, and cannot effectively respond to real-time fluctuations in photovoltaic power generation, random changes in load, and dynamic signals of grid electricity prices; this results in the system operating in a suboptimal state for a long time and failing to capture the best energy allocation opportunity in a complex and ever-changing environment. Second, the optimization objectives are singular and the coordination is insufficient. Most system solutions only aim to maximize economic benefits, ignoring the coordination of various components within the system and the impact on the external power grid. That is, photovoltaic, energy storage, charging piles and loads are treated as independent units, and there is no connection management. This can easily lead to problems such as photovoltaic curtailment, shortened lifespan of energy storage batteries, and disorderly charging of charging piles impacting the system demand. Third, the lack of predictive planning and weak real-time optimization capabilities; the existing system focuses on passive response to the current state or only makes plans based on insufficient day-ahead forecasts; due to the lack of high-precision ultra-short-term photovoltaic and load forecasts, the system cannot perform real-time dynamic optimization in a rolling manner, which makes the system unable to anticipate future changes, resulting in decision lag and reduced efficiency. Fourth, the resource allocation mechanism is rigid and lacks flexibility; the simple first-come-first-served or fixed power allocation mode is used for adjustable loads such as charging piles, which cannot realize the flexible allocation and reuse of system power resources; for example, when a vehicle finishes charging, the power resources it releases cannot be intelligently and quickly redistributed to other waiting charging piles or loads, which easily leads to the coexistence of idle resources and extended user waiting time, limiting the overall throughput and efficiency of the system and resulting in a poor user experience. Summary of the Invention

[0004] To address the shortcomings of the aforementioned technologies, this invention provides a method and system for energy management and intelligent allocation in an integrated photovoltaic, energy storage, and charging system.

[0005] To solve the above technical problems, the technical solution adopted by this invention is: an energy management and intelligent allocation method for an integrated photovoltaic-storage-charging system, specifically including the following steps: Step S1: The edge energy manager in the edge computing layer collects system-wide data in real time from the devices in the perception and execution layer through various communication protocols; Step S2: The cloud platform layer performs ultra-short-term power prediction of photovoltaic power and load based on the AI ​​ultra-short-term prediction model, and generates prediction curves of photovoltaic power generation and load demand in the future period. Step S3: Based on the real-time full-system data in Step S1, dynamically calibrate the digital twin model in the cloud platform layer; Step S4: In the calibrated digital twin model, target optimization is performed using a rolling time window, and a multi-objective dynamic permission decision-making mechanism is implemented; Step S5: Generate the optimal power allocation scheme; after optimization, output the optimal power command set for the next time window; Step S6: Execute and schedule according to the optimal power instruction set and based on the elastic power pool scheduler; Step S7: Feed back the operating data of each device to the AI ​​ultra-short-term prediction model and store it for the AI ​​ultra-short-term prediction model to perform self-learning and retrain the AI ​​ultra-short-term prediction model periodically.

[0006] Furthermore, step S2 involves performing ultra-short-term power prediction for photovoltaic power and load based on an AI ultra-short-term prediction model, specifically including the following steps: Step S21: Collect raw data and clean and organize it; the raw data includes photovoltaic data, load data, meteorological data, time characteristics and external data; Step S22: Select a model and define its inputs and outputs; Step S23: Train and optimize the model; divide the dataset, selecting 70% of the data as the training set for learning patterns, 15% as the validation set for adjusting parameters, and 15% as the test set for final evaluation; use the mean squared error as the loss function and apply the Adam optimizer to automatically adjust the learning rate; calculate the evaluation metric to measure the magnitude of the prediction error. Step S24: Perform online predictions and continuously optimize the AI ​​ultra-short-term prediction model.

[0007] Furthermore, step S21 involves cleaning and organizing the raw data, specifically including the following steps: Step S21-1: Complete missing values; Step S21-2: Remove outliers; Step S21-3: Standardize the timestamps to 15-minute intervals;

[0008] Step S21-4: Normalize the data to the range [0,1].

[0009] Furthermore, step S3 involves dynamically calibrating the digital twin model in the cloud platform layer, specifically including the following steps: Step S31: Perform physical system modeling; define system boundaries and components, and construct a complete energy balance equation from the photovoltaic DC side, energy storage DC side, PCS DC side, AC side to the grid, and define the voltage, current and power constraints of each connection point; Step S32: Parameter initialization and calibration; Obtain initial parameters, including nameplate parameters, measured data and empirical values; After the system is put into operation for the first time or after a major overhaul, collect full-condition operation data for a period of time, and use parameter identification methods to optimize model parameters to minimize the error between the model simulation output and the actual measured values; Step S33: Set up a dynamic calibration mechanism; determine standard parameters, establish a mapping relationship between standard parameters and running data, and update the mechanism model. When the deviation between the measured value and the simulated value exceeds the threshold, adjust the model parameters in reverse. After adjustment, use a Kalman filter to smooth the parameter evaluation; verify the accuracy of the calibrated model, establish a model performance evaluation mechanism, and recalibrate offline according to the cycle.

[0010] Furthermore, the multi-objective dynamic optimization algorithm for the multi-objective dynamic permission decision-making mechanism in step S4 is as follows: Minimize [ α·Cost - β·Stability + γ·Green ]; Wherein, α is the economic weighting factor; β is the stability weighting factor; γ is the green energy consumption weighting factor; Cost is the total electricity cost, including electricity cost and demand cost; Stability is the system stability index, including voltage deviation, frequency deviation and battery life loss factor; Green is the green energy index, namely photovoltaic self-consumption rate; Specifically, the weighting factors of each objective are adjusted by a dynamic weighting adjuster; when a high risk of grid voltage or frequency exceeding limits is detected, the stability weight β is automatically increased to enable the system to prioritize grid stability; during peak hours of time-of-use pricing, the economic weight α is automatically increased to enable the system to more actively implement peak-valley arbitrage; when the user sets a mode that maximizes green electricity preference, the green energy consumption weight γ is automatically increased to enable the system to prioritize photovoltaic power consumption.

[0011] Furthermore, step S6 involves execution and scheduling based on the elastic power pool scheduler, specifically including the following steps: Step S61: The Edge Energy Manager (EMU) in the edge computing layer receives and executes instructions from the cloud platform layer; Step S62: Through PCS and photovoltaic inverter control, the power command is converted into the device communication protocol and sent out; Step S63: Scheduling is performed based on the elastic power pool scheduler, and the edge energy manager (EMU) calculates the elastic power pool capacity in real time; Step S64: The charging pile controller requests power from the flexible power pool, and the system receives the power requests from each charging pile; Step S65: The elastic power pool scheduler dynamically allocates the capacity in the elastic power pool to each charging pile based on the charging urgency, user priority, and real-time electricity price.

[0012] Furthermore, in step S63, the value of the flexible power pool capacity is the difference between the sum of the photovoltaic adjustable power and the energy storage adjustable power and the critical load power.

[0013] The integrated photovoltaic-storage-charging energy management and intelligent allocation system includes a cloud platform layer, comprising: a prediction unit for minute-level predictions of photovoltaic power generation and load based on an AI ultra-short-term prediction model; a digital twin unit for building a virtual simulation system using a system-level digital twin model to perform real-time modeling and simulation prediction of the physical system; an optimization decision-making unit for achieving multi-objective energy trade-offs based on a multi-objective dynamic optimization algorithm to ensure economic efficiency, stability, and minimize carbon emissions; an instruction generation unit for generating power allocation instructions; a digital storage unit for storing various types of data; and a self-learning unit for iterative updates of the AI ​​ultra-short-term prediction model. The edge computing layer includes a built-in edge power manager (EMU). The edge power manager (EMU) includes a data collection and preprocessing unit, which collects, cleans, and structures data from the perception and execution layer; an elastic power pool scheduler, which flexibly schedules various controllable resources to improve system elasticity; and a cloud command execution control unit, which receives and executes scheduling commands issued by the elastic power pool in the cloud platform layer to achieve power allocation and scheduling. The perception and execution layer includes sensing devices for perceiving the physical world and execution devices for controlling the execution devices, responsible for energy conversion and control, and executing control commands issued from the upper layer. The physical layer includes energy storage systems, photovoltaic power systems, and electric vehicle charging stations.

[0014] Furthermore, the Edge Energy Manager (EMU) also includes a local fast control logic unit for enabling rapid local control in case of network instability or emergencies, ensuring power supply to critical loads.

[0015] Furthermore, the sensing devices include smartwatches for collecting relevant parameters at the grid inlet; weather stations for monitoring environmental variables and providing data support for photovoltaic forecasting; and EMS battery management systems for monitoring the SOC state of charge, SOH state of health, battery voltage, and current of energy storage batteries. The execution equipment includes PCS energy storage converters, photovoltaic inverters, and charging pile controllers.

[0016] This invention discloses a method and system for energy management and intelligent allocation in an integrated photovoltaic-storage-charging system, which has the following beneficial effects: 1. By setting up a perception and execution layer that can perceive the state within the system in real time, and making minute-level rolling optimization decisions based on this, the system has a strong environmental adaptability and dynamically tracks the optimal running point. Second, by setting up a multi-objective dynamic weight decision-making mechanism, the system can intelligently weigh and switch between multiple objectives such as economic benefits, system stability, green electricity consumption ratio and grid friendliness, ensuring that the system achieves global optimization under complex constraints. Third, the system is equipped with an AI ultra-short-term prediction model and a digital twin model. The high-precision prediction provides the system with accurate prediction capabilities, and through simulation calculations in the digital twin, it realizes the transformation from passive response to active planning, thereby improving the accuracy and robustness of control. Fourth, the management method of scheduling by setting up a flexible buffer pool virtualizes the available power of photovoltaic and energy storage into a shared resource pool, and performs flexible, dynamic and reusable intelligent scheduling of charging pile power according to the urgency of charging demand, user priority and real-time electricity price, thereby maximizing the utilization rate of system resources while ensuring user experience. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the system framework of the present invention.

[0018] Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 and 2 The energy management and intelligent allocation method and system for the integrated photovoltaic-storage-charging system are shown. The energy management and intelligent allocation method for the integrated photovoltaic-storage-charging system specifically includes the following steps: Step S1: The edge energy manager in the edge computing layer collects system-wide data in real time from the devices in the perception and execution layer through various communication protocols; In this embodiment, the system data includes grid-side data, photovoltaic-side data, energy storage-side data, load-side data, and environmental-side data. Grid-side data includes entry point voltage, frequency, active power, reactive power, and time-of-use electricity price signal. Photovoltaic-side data includes DC-side voltage, DC-side current, and AC-side output power. Energy storage-side data includes SOC (State of Charge), SOH (State of Health), voltage, and temperature. Load-side data includes power of important loads, real-time power, status, and charging requests of all charging piles. Environmental-side data includes irradiance and ambient temperature.

[0021] Step S2: The cloud platform layer performs ultra-short-term power prediction of photovoltaic power and load based on the AI ​​ultra-short-term prediction model; the cloud platform layer receives real-time and historical data uploaded from the edge computing layer, runs the built-in AI ultra-short-term prediction model, and generates prediction curves for photovoltaic power generation and load demand for the next 15 to 30 minutes; in this embodiment, the AI ​​ultra-short-term prediction model predicts photovoltaic power generation and load demand for the next 30 minutes; it should be noted that this AI ultra-short-term prediction model continuously optimizes its prediction accuracy by learning from historical data and weather information; the prediction of photovoltaic power generation and load demand through the AI ​​ultra-short-term prediction model specifically includes the following steps: Step S21: Collect raw data and clean and organize the raw data; the raw data includes photovoltaic data, i.e., historical power generation; load data, i.e., historical power consumption; meteorological data, i.e., real-time irradiance, temperature, humidity and cloud cover; time characteristics, i.e. time, week, season and holiday; external data, i.e. radiation and temperature forecasts in weather forecasts. The cleaning and organization of the raw data includes the following steps: Step S21-1: Complete missing values; Step S21-2: Remove outliers, such as non-zero power generation data at night; Step S21-3: Standardize the timestamps to 15-minute intervals; Step S21-4: Normalize the data to the range [0,1].

[0022] Step S22: Select a model and define its input and output; in this embodiment, LSTM or Transformer is selected; LSTM is good at processing time series and capturing long-term dependencies, while Transformer is suitable for multivariate prediction and has high parallel computing efficiency; use the raw data from the past 4 to 6 hours as the model input and the power prediction values ​​for the next 15 to 60 minutes as the model output. Step S23: Perform model training and optimization; specifically including the following steps: Step S23-1: Divide the dataset; the dataset consists of continuous data with a duration longer than a full year, including seasonal changes, holiday patterns, and all typical weather types; 70% of the data will be selected as the training set for learning patterns; 15% will be used as the validation set for adjusting parameters; and 15% will be used as the test set for final evaluation; it should be noted that the training set is a continuous time period, not randomly selected time points. Step S23-2: Use the mean squared error as the loss function and apply the Adam optimizer to automatically adjust the learning rate; Step S23-3: Calculate the evaluation metrics; the evaluation metrics include RMSE, which measures the magnitude of the prediction error; MAE (mean absolute error), R0 2 The confidence level used to predict the outcome.

[0023] Step S24: Perform online predictions and continuously optimize the AI ​​ultra-short-term prediction model; receive the latest data in 5-15 minute cycles and run the model to generate prediction curves; retrain the model regularly with new data, use A / B testing to compare the effects of the new and old models, and automatically update the online model version.

[0024] Combining AI ultra-short-term prediction models with historical data is not only used for front-end optimization input, but also for back-end continuous iterative optimization of AI ultra-short-term prediction models and strategy parameters, ultimately effectively improving the overall energy efficiency, economic benefits, and grid friendliness of the system.

[0025] Step S3: Based on the real-time full-system data from Step S1, update the model parameters, i.e., dynamically calibrate the digital twin model. It should be noted that the cloud platform maintains a high-fidelity system digital twin model, which simulates the entire energy flow from the DC side to the AC side. Dynamically calibrating the model parameters based on the real-time full-system data collected in S1 ensures a high degree of consistency between the model and the physical system, providing a reliable simulation environment for subsequent optimization calculations. Specifically, this includes the following steps: Step S31: Perform physical system modeling; specifically including the following steps: Step S31-1: Define the system boundary and components; the components include a photovoltaic array, used to establish the PV model, with irradiance and temperature as inputs and DC power as output; Energy storage system, used to establish battery equivalent circuit model, including SOC-OCV relationship, internal resistance and capacity decay; where SOC is the percentage of battery remaining capacity, and OCV is the open circuit voltage of battery in rest state; PCS energy storage converters or photovoltaic inverters are used to establish efficiency curve models, power limiting, and response characteristics. Power grid load is used to establish the power grid equivalent impedance model and the load power characteristic model.

[0026] Step S31-2: Construct a complete energy balance equation from the photovoltaic DC side, energy storage DC side, PCS DC side, AC side to the grid, and define the voltage, current and power constraints at each connection point. Step S32: Parameter initialization and calibration; specifically including the following steps: Step S32-1: Obtain initial parameters; initial parameters include nameplate parameters, measured data, and empirical values; nameplate parameters are the rated power, voltage range, and efficiency obtained from the equipment; measured data are the key curves obtained through factory testing or on-site commissioning; empirical values ​​are used to initialize unknown parameters using industry-typical values. Step S32-2: Perform offline calibration; after the system is put into operation for the first time or after a major overhaul, collect full-condition operation data for a period of time, and use parameter identification methods to optimize model parameters so as to minimize the error between the model simulation output and the actual measurement value; parameter identification methods include least squares method and genetic algorithm.

[0027] Step S33: Set up the dynamic calibration mechanism; specifically including the following steps: Step S33-1: Determine calibration parameters; calibration parameters include slow time-varying parameters, fast time-varying parameters, and environmental parameters; slow time-varying parameters include battery internal resistance and capacity decay coefficient, wherein the capacity decay coefficient is calibrated daily or weekly; fast time-varying parameters include inverter efficiency and photovoltaic array contamination coefficient, wherein the photovoltaic array contamination coefficient is calibrated hourly; environmental parameters include thermal resistance coefficient and heat dissipation efficiency, wherein the heat dissipation efficiency varies with temperature; Step S33-2: Design a calibration algorithm; establish a mapping relationship between standard parameters and running data, and update the mechanism model. When the deviation between the measured value and the simulated value exceeds the threshold, adjust the model parameters in reverse; after adjustment, use Kalman filters and other methods to smooth the parameter estimates. Step S33-3: Verify the accuracy of the calibrated model under typical working conditions, establish a model confidence assessment mechanism, and recalibrate offline according to the cycle.

[0028] Step S4: In the calibrated digital twin model, multi-objective optimization is performed using a rolling time window, and a multi-objective dynamic weight decision-making mechanism is implemented; among which, the multi-objectives include economic objectives, stability objectives, and green energy consumption objectives; The multi-objective dynamic optimization algorithm is as follows: Minimize [ α·Cost - β·Stability + γ·Green ]; Wherein, α is the economic weighting factor; β is the stability weighting factor; γ is the green energy consumption weighting factor; Cost is the total electricity cost, including electricity cost and demand cost; Stability is the system stability index, including voltage deviation, frequency deviation and battery life loss factor; Green is the green energy index, namely the photovoltaic self-consumption rate. The system adjusts the weighting factors of each objective through a dynamic weight adjuster. When a high risk of grid voltage or frequency exceeding limits is detected, the stability weight β is automatically increased to prioritize grid stability. During peak hours of time-of-use pricing, the economic weight α is automatically increased to encourage more proactive peak-valley arbitrage. When users set a mode that maximizes green energy preference, the green energy consumption weight γ is automatically increased to prioritize photovoltaic energy consumption. A multi-objective optimization decision-making module is used to dynamically decide on the weights, intelligently balancing economic efficiency, safety, and green energy consumption objectives.

[0029] It should be noted that the variables for solving multi-objective dynamic optimization decisions are subject to certain constraints; these include: upper and lower limits of energy storage SOC; maximum charging and discharging power of the energy storage converter PCS and inverter; and the allowable range of grid voltage and frequency. These constraints limit the optimization variables to a feasible domain that ensures the safe operation of physical equipment and compliance with grid connection regulations, thereby guaranteeing the executability of the optimization decisions.

[0030] Step S5: Generate the optimal power allocation scheme; after optimization, output the optimal power instruction set for the next time window. The optimal power instruction set includes the target charging and discharging power of the PCS, the power limit of the photovoltaic inverter, and the overall power allocation instruction of the charging pile group. Step S6: Execute and schedule the power allocation instructions, i.e., the optimal power instruction set, based on the elastic power pool scheduler; specifically, this includes the following steps: Step S61: The Edge Energy Manager (EMU) in the edge computing layer receives and executes instructions from the cloud platform layer; Step S62: Through PCS and photovoltaic inverter control, the power command is converted into the device communication protocol and sent out; Step S63: Scheduling is performed based on the elastic power pool scheduler; the capacity of the elastic power pool is calculated in real time through the EMU. The value of the elastic power pool capacity is the difference between the sum of the adjustable power of photovoltaic and the adjustable power of energy storage and the power of the critical load; through the elastic power pool, the interaction between photovoltaic, energy storage and load is abstracted into the elastic power pool. The charging pile, as a "consumer", requests power from the pool. The system dynamically allocates power according to multiple strategies, realizing flexible scheduling and resource reuse of charging power. This changes the traditional extensive mode of plug-and-play and first-come-first-served charging piles, and greatly improves the overall efficiency of the system.

[0031] Step S64: The charging pile controller requests power from the flexible power pool, and the system receives the power requests from each charging pile; Step S65: The flexible power pool scheduler dynamically allocates power, i.e., dynamically allocates the capacity in the flexible power pool to each charging pile based on factors such as charging urgency, user priority, and real-time electricity price. For example, for non-urgent charging tasks, high power is allocated during peak photovoltaic periods, and low power is allocated when the system power is tight, so as to carry out power reuse and intelligent orderly charging.

[0032] Step S7: Feed back the operating data of each device to the AI ​​ultra-short-term prediction model and store it for the AI ​​ultra-short-term prediction model to perform self-learning and retrain the AI ​​ultra-short-term prediction model regularly; the system feeds back the actual operating data to the cloud platform database; the cloud platform regularly uses new data to retrain the AI ​​ultra-short-term prediction model and evaluates the effectiveness of the optimization strategy to achieve continuous evolution of the entire system.

[0033] The system will trigger the collection of the latest data every five minutes, that is, it will automatically start a complete work cycle of "data collection → model prediction → optimization decision → execution" every five minutes.

[0034] The integrated photovoltaic, energy storage, and charging system's energy management and intelligent allocation system includes a cloud platform layer, which is used for global optimization decision-making and intelligent analysis based on AI ultra-short-term prediction models, multi-objective dynamic optimization algorithms, and system-level digital twin models. It includes a forecasting unit for making minute-level forecasts of photovoltaic power generation and load based on an AI ultra-short-term forecasting model, thereby improving the foresight of the system response; Digital twin units are used to build virtual simulation systems from system-level digital twin models, enabling real-time modeling and simulation prediction of physical systems. The optimized decision-making unit is used to achieve multi-objective energy trade-offs based on multi-objective dynamic optimization algorithms, ensuring economic efficiency, stability, and minimization of carbon emissions; The instruction generation unit is used to generate power allocation instructions. It also includes a data storage unit for storing various types of data; and a self-learning unit for iterative updates of the AI ​​ultra-short-term prediction model.

[0035] The cloud platform layer communicates with the edge computing layer via 4G, 5G or Ethernet to send control policies and receive field data.

[0036] The edge computing layer has a built-in edge power manager (EMU), which uses various communication protocols to collect system-wide data from devices in the perception and execution layer in real time. The edge energy manager includes a data collection and preprocessing unit for collecting, cleaning, and structuring data from the perception and execution layer; A flexible power pool scheduler is used to flexibly schedule various controllable resources, improving system resilience. In this embodiment, controllable resources include energy storage and charging piles; The cloud-based command execution control unit is used to receive and execute scheduling commands issued by the elastic power pool in the cloud platform layer to realize power allocation and scheduling. It should be noted that it contains an elastic power pool scheduling submodule, which is used for dynamic power allocation of charging piles. It also includes a local fast control logic unit, which is used to achieve fast local control in case of network instability or emergency, and ensure power supply to critical loads; The edge computing layer communicates with the sensing and execution layer devices through industrial communication protocols such as CAN, RS485, or Modbus-TCP.

[0037] The perception and execution layer includes sensing devices for perceiving the physical world and execution devices for controlling the execution devices, responsible for energy conversion and control, and executing control commands issued from the upper layer. Sensing devices include smart meters, which are used to collect parameters such as voltage, current and energy at the power grid inlet; Weather stations are used to monitor environmental variables such as solar radiation and temperature, providing data support for photovoltaic forecasting. The BMS (Battery Management System) is used to monitor information such as the SOC (State of Charge), SOH (State of Health), battery voltage, and current of energy storage batteries. The execution devices include PCS energy storage converters, photovoltaic inverters, and charging pile controllers. The sensing and execution layer communicates with the physical layer via industrial communication protocols such as CAN, RS485, and Modbus-TCP.

[0038] The physical layer, including energy storage systems (Battery), photovoltaic power generation systems (PV), and electric vehicle charging stations (EV Chargers), forms the basic resource layer of the system.

[0039] This invention utilizes a rolling optimization technique combining a digital twin model and a multi-objective optimization decision module, employing real-time data for dynamic calibration. It executes optimization calculations within minute-level rolling time windows, achieving a leap from static planning to dynamic adaptation. Simultaneously, it adjusts the weight decision mechanism in real-time based on system mode, power grid status, and user preferences, realizing an adaptive intelligent trade-off between economic efficiency, safety, and environmental friendliness. This achieves a leap from static to dynamic adaptive system strategy, resolving the problems of rigid strategies, insufficient coordination, and low resource utilization efficiency in existing solutions.

[0040] The above embodiments are not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the technical solution of the present invention are also within the protection scope of the present invention.

Claims

1. A method for energy management and intelligent allocation in an integrated photovoltaic-storage-charging system, characterized in that, Specifically, the following steps are included: Step S1: The edge energy manager in the edge computing layer collects system-wide data in real time from the devices in the perception and execution layer through various communication protocols; Step S2: The cloud platform layer performs ultra-short-term power prediction of photovoltaic power and load based on the AI ​​ultra-short-term prediction model, and generates prediction curves of photovoltaic power generation and load demand in the future period. Step S3: Based on the real-time full-system data in Step S1, dynamically calibrate the digital twin model in the cloud platform layer; Step S4: In the calibrated digital twin model, target optimization is performed using a rolling time window, and a multi-objective dynamic permission decision-making mechanism is implemented; Step S5: Generate the optimal power allocation scheme; after optimization, output the optimal power command set for the next time window; Step S6: Execute and schedule according to the optimal power instruction set and based on the elastic power pool scheduler; Step S7: Feed back the operating data of each device to the AI ​​ultra-short-term prediction model and store it for the AI ​​ultra-short-term prediction model to perform self-learning and retrain the AI ​​ultra-short-term prediction model periodically.

2. The energy management and intelligent allocation method for the integrated photovoltaic-storage-charging system according to claim 1, characterized in that: Step S2, which involves forecasting ultra-short-term power of photovoltaic power and load based on an AI ultra-short-term prediction model, specifically includes the following steps: Step S21: Collect raw data and clean and organize it; the raw data includes photovoltaic data, load data, meteorological data, time characteristics and external data; Step S22: Select a model and define its inputs and outputs; Step S23: Train and optimize the model; divide the dataset, selecting 70% of the data as the training set for learning patterns, 15% as the validation set for adjusting parameters, and 15% as the test set for final evaluation; use the mean squared error as the loss function and apply the Adam optimizer to automatically adjust the learning rate; calculate the evaluation metric to measure the magnitude of the prediction error. Step S24: Perform online predictions and continuously optimize the AI ​​ultra-short-term prediction model.

3. The energy management and intelligent allocation method for the integrated photovoltaic-storage-charging system according to claim 2, characterized in that: Step S21 involves cleaning and organizing the raw data, specifically including the following steps: Step S21-1: Complete missing values; Step S21-2: Remove outliers; Step S21-3: Standardize the timestamps to 15-minute intervals; Step S21-4: Normalize the data to the range [0,1].

4. The energy management and intelligent allocation method for the integrated photovoltaic-storage-charging system according to claim 1, characterized in that: Step S3 involves dynamically calibrating the digital twin model in the cloud platform layer, specifically including the following steps: Step S31: Perform physical system modeling; define system boundaries and components, and construct a complete energy balance equation from the photovoltaic DC side, energy storage DC side, PCS DC side, AC side to the grid, and define the voltage, current and power constraints of each connection point; Step S32: Parameter initialization and calibration; Obtain initial parameters, including nameplate parameters, measured data and empirical values; After the system is put into operation for the first time or after a major overhaul, collect full-condition operation data for a period of time, and use parameter identification methods to optimize model parameters to minimize the error between the model simulation output and the actual measured values; Step S33: Set up a dynamic calibration mechanism; determine standard parameters, establish a mapping relationship between standard parameters and running data, and update the mechanism model. When the deviation between the measured value and the simulated value exceeds the threshold, adjust the model parameters in reverse. After adjustment, use a Kalman filter to smooth the parameter evaluation; verify the accuracy of the calibrated model, establish a model performance evaluation mechanism, and recalibrate offline according to the cycle.

5. The energy management and intelligent allocation method for the integrated photovoltaic-storage-charging system according to claim 1, characterized in that: The multi-objective dynamic optimization algorithm for the multi-objective dynamic permission decision-making mechanism in step S4 is as follows: Minimize [ α·Cost - β·Stability + γ·Green ]; Wherein, α is the economic weighting factor; β is the stability weighting factor; γ is the green energy consumption weighting factor; Cost is the total electricity cost, including electricity cost and demand cost; Stability is the system stability index, including voltage deviation, frequency deviation and battery life loss factor; Green is the green energy index, namely photovoltaic self-consumption rate; Specifically, the weighting factors of each objective are adjusted by a dynamic weighting adjuster; when a high risk of grid voltage or frequency exceeding limits is detected, the stability weight β is automatically increased to enable the system to prioritize grid stability; during peak hours of time-of-use pricing, the economic weight α is automatically increased to enable the system to more actively implement peak-valley arbitrage; when the user sets a mode that maximizes green electricity preference, the green energy consumption weight γ is automatically increased to enable the system to prioritize photovoltaic power consumption.

6. The energy management and intelligent allocation method for the integrated photovoltaic-storage-charging system according to claim 1, characterized in that: Step S6 involves execution and scheduling based on the elastic power pool scheduler, specifically including the following steps: Step S61: The Edge Energy Manager (EMU) in the edge computing layer receives and executes instructions from the cloud platform layer; Step S62: Through PCS and photovoltaic inverter control, the power command is converted into the device communication protocol and sent out; Step S63: Scheduling is performed based on the elastic power pool scheduler, and the edge energy manager (EMU) calculates the elastic power pool capacity in real time; Step S64: The charging pile controller requests power from the flexible power pool, and the system receives the power requests from each charging pile; Step S65: The elastic power pool scheduler dynamically allocates the capacity in the elastic power pool to each charging pile based on the charging urgency, user priority, and real-time electricity price.

7. The energy management and intelligent allocation method for the integrated photovoltaic-storage-charging system according to claim 6, characterized in that: In step S63, the value of the flexible power pool capacity is the difference between the sum of the photovoltaic adjustable power and the energy storage adjustable power and the critical load power.

8. An integrated photovoltaic-storage-charging system energy management and intelligent allocation system, applied to the integrated photovoltaic-storage-charging system energy management and intelligent allocation method as described in any one of claims 1-7, characterized in that, include: The cloud platform layer includes a forecasting unit, which is used to make minute-level forecasts of photovoltaic power generation and load based on an AI ultra-short-term forecasting model; Digital twin units are used to build virtual simulation systems from system-level digital twin models, enabling real-time modeling and simulation prediction of physical systems. The optimization decision-making unit is used to achieve multi-objective energy trade-offs based on multi-objective dynamic optimization algorithms, ensuring economy, stability and minimization of carbon emissions; the instruction generation unit is used to generate power allocation instructions. Digital storage units are used to store various types of data; Self-learning units are used to enable iterative updates of AI ultra-short-term prediction models; The edge computing layer includes a built-in edge power manager (EMU). The edge power manager (EMU) includes a data collection and preprocessing unit for collecting, cleaning, and structuring data from the perception and execution layer; and an elastic power pool scheduler for flexibly scheduling various controllable resources to improve system elasticity. The cloud-based command execution control unit is used to receive and execute scheduling commands issued by the elastic power pool in the cloud platform layer to realize power allocation and scheduling. The perception and execution layer includes sensing devices for perceiving the physical world and execution devices for controlling the execution devices, responsible for energy conversion and control, and executing control commands issued from the upper layer. The physical layer includes energy storage systems, photovoltaic power systems, and electric vehicle charging stations.

9. The energy management and intelligent distribution system for the integrated photovoltaic-storage-charging system according to claim 8, characterized in that: The edge power manager (EMU) also includes a local fast control logic unit, which enables rapid local control in case of network instability or emergencies, ensuring power supply to critical loads.

10. The energy management and intelligent distribution system for the integrated photovoltaic-storage-charging system according to claim 9, characterized in that: The sensing devices include a smartwatch for collecting relevant parameters from the power grid inlet; a weather station for monitoring environmental variables and providing data support for photovoltaic forecasting; and an EMS battery management system for monitoring the SOC state of charge, SOH state of health, battery voltage, and current of the energy storage battery. The execution equipment includes PCS energy storage converters, photovoltaic inverters, and charging pile controllers.

Images