Building light storage and charging integrated energy management system

Abstract

The invention relates to the technical field of data processing, in particular to a building light storage and charging integrated energy management system, which is used for solving the problems of data processing delay and resource coordination mismatch caused by insufficient multi-source heterogeneous data fusion in photovoltaic power generation, energy storage and grid connection links. The system comprises a data acquisition module, a protocol conversion module, a data fusion module, a scheduling optimization module, an instruction generation module and a feedback module. The protocol conversion module carries out Modbus, CAN bus and IEC 61850 protocol analysis on the photovoltaic output time sequence data, the energy storage charging and discharging state data and the power grid dispatching instruction to generate standardized data streams. And the safety verification module intercepts the overrun instruction and triggers an amplitude limiting strategy. Through protocol unification, dynamic fusion and closed-loop control, the system realizes improvement of photovoltaic consumption rate, reduction of energy storage circulation loss and optimization of power grid frequency modulation deviation, and significantly improves the efficiency and economy of light storage and charging cooperative operation.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular, to a building integrated photovoltaic energy storage and charging energy management system. Background Art

[0002] In the scenario of coordinated operation of new energy power generation and energy storage systems, there are time-sequence differences and coupling characteristics in the multi-dimensional data of photovoltaic power generation output volatility, energy storage charge and discharge dynamic response, and load demand. Existing energy management systems have limitations in the real-time coordinated scheduling of multi-source heterogeneous data. Due to the lack of a unified data fusion mechanism for the discrete operation data generated in the photovoltaic power generation, energy storage, and grid connection links of the booster station at the levels of transmission protocols, sampling periods, and control objectives, the system is difficult to dynamically adapt to the real-time matching of grid dispatching instructions and local operating states, resulting in data processing delays and resource coordination mismatches, affecting the overall operation efficiency and economy of the photovoltaic energy storage and charging system. Summary of the Invention

[0003] Aiming at the deficiencies of the existing technology, the present invention provides a building integrated photovoltaic energy storage and charging energy management system to solve the problems that the existing energy management system lacks a unified fusion mechanism for multi-source heterogeneous data in the photovoltaic power generation, energy storage, and grid connection links, resulting in data processing delays and resource coordination mismatches, thereby reducing the operation efficiency and economy of the photovoltaic energy storage and charging system.

[0004] To solve the above technical problems, the specific technical solutions of the present invention are as follows: A building integrated photovoltaic energy storage and charging energy management system provided by the present invention includes: A data acquisition module for acquiring multi-source heterogeneous data, where the multi-source heterogeneous data includes photovoltaic output time-series data, energy storage charge and discharge state data, charging load demand data, and grid dispatching instruction data; A protocol conversion module for performing protocol parsing on the multi-source heterogeneous data received from the data acquisition module, parsing the photovoltaic output time-series data through the Modbus protocol, the energy storage charge and discharge state data through the CAN bus protocol, and the grid dispatching instruction data through the IEC 61850 protocol to generate a standardized data stream in a unified data format; A data fusion module for performing sliding window interception on the standardized data stream received from the protocol conversion module, extracting the fluctuation feature vector in the photovoltaic output time-series data and the dynamic response vector in the energy storage charge and discharge state data, and inputting the fluctuation feature vector and the dynamic response vector into a dynamic weighted model constructed based on the photovoltaic output volatility and the energy storage response delay for time-domain convolution operation to generate a multi-dimensional coupling state matrix; A scheduling optimization module, which is used to construct a multi-objective function with the PV accommodation rate, energy storage cycle loss, and grid frequency modulation deviation as the optimization objectives according to the multi-dimensional coupling state matrix received from the data fusion module and the grid scheduling instruction data received from the data acquisition module, and uses an improved particle swarm optimization algorithm with the inertia weight dynamically adjusted according to the PV output volatility to solve and generate a real-time scheduling strategy; An instruction generation module, which is used to convert the charge and discharge power values in the real-time scheduling strategy output by the scheduling optimization module into PWM modulation signals, and distribute the PWM modulation signals to the inverters in the PV power generation units, the BMS systems in the energy storage units, and the charging pile controllers in the charging facility units through the OPC UA protocol; A feedback module, which is used to monitor the actual output data of the PV power generation unit, the real-time state of charge data of the energy storage unit, and the power consumption data of the charging facility unit, calculate the deviation between the actual output data and the predicted value in the real-time scheduling strategy, generate a feedback correction coefficient, and input the feedback correction coefficient into the scheduling optimization module to trigger the iterative optimization of the multi-objective function and adjust the real-time scheduling strategy.

[0005] Furthermore, for the building integrated PV, energy storage and charging energy management system of the present invention, the protocol conversion module is further used for: Performing Modbus protocol parsing on the PV output time series data received from the data acquisition module, and extracting the PV power generation power and voltage fluctuation characteristics; Performing CAN bus protocol parsing on the energy storage charge and discharge state data received from the data acquisition module, and extracting the state of charge and charge and discharge rate; Performing IEC 61850 protocol parsing on the grid scheduling instruction data received from the data acquisition module, and extracting the grid connection power threshold and frequency modulation instruction parameters; Taking the PV power generation power, voltage fluctuation characteristics, state of charge, charge and discharge rate, grid connection power threshold, and frequency modulation instruction parameters as the parsing results, and outputting them to the data fusion module.

[0006] Furthermore, for the building integrated PV, energy storage and charging energy management system of the present invention, the data fusion module includes: A feature extraction unit, which is used to perform sliding window interception on the standardized data stream received from the protocol conversion module, and extract the fluctuation feature vector in the PV output time series data and the dynamic response vector in the energy storage charge and discharge state data; A coupling analysis unit, which is used to input the fluctuation feature vector and the dynamic response vector into a dynamic weighting model constructed based on the PV output volatility and the energy storage response delay to generate the multi-dimensional coupling state matrix.

[0007] Furthermore, for the building integrated photovoltaic-energy storage-charging energy management system of the present invention, the coupling analysis unit further includes: The volatility of photovoltaic power output is calculated through the variance of the fluctuation feature vector; The energy storage response delay is calculated through the time delay parameter in the dynamic response vector; The dynamic weighting model dynamically adjusts the weighting coefficient according to the ratio of the volatility of photovoltaic power output to the energy storage response delay, and performs weighted superposition on the fluctuation feature vector and the dynamic response vector to generate a coupling feature tensor; The coupling feature tensor is matrix-processed to obtain the multi-dimensional coupling state matrix.

[0008] Furthermore, for the building integrated photovoltaic-energy storage-charging energy management system of the present invention, the scheduling optimization module is further used for: According to the predicted value of photovoltaic power output and the remaining capacity of energy storage in the multi-dimensional coupling state matrix received from the data fusion module, a charging and discharging power distribution plan is generated, wherein the predicted value of photovoltaic power output is calculated through the regression analysis of the fluctuation feature vector; Combined with the grid connection power threshold in the grid scheduling instruction data received from the data acquisition module, the charging and discharging power distribution plan is constrained and optimized, and the constraint optimization is based on the multi-objective optimization function to generate the real-time scheduling strategy.

[0009] Furthermore, for the building integrated photovoltaic-energy storage-charging energy management system of the present invention, it further includes: Based on the predicted value of photovoltaic power output in the multi-dimensional coupling state matrix received from the data fusion module and the weighting coefficient output from the dynamic weighting model, a multi-objective optimization function with the maximum photovoltaic power consumption rate, the minimum energy storage cycle loss, and the minimum grid frequency modulation deviation is constructed; An improved particle swarm optimization algorithm is used to solve the multi-objective optimization function, and the inertia weight of the improved particle swarm optimization algorithm is dynamically adjusted according to the volatility of photovoltaic power output to generate a Pareto optimal solution set; According to the priority weight in the grid scheduling instruction data received from the data acquisition module, the real-time scheduling strategy is selected from the Pareto optimal solution set.

[0010] Furthermore, for the building integrated photovoltaic-energy storage-charging energy management system of the present invention, the instruction generation module includes: An instruction encoding unit, which is used to convert the charging and discharging power values in the real-time scheduling strategy output by the scheduling optimization module into PWM modulation signals based on the Pareto optimal solution set generated by the scheduling optimization module; An instruction distribution unit for synchronously transmitting the PWM modulation signal to the BMS system in the energy storage unit and the charging pile controller in the charging facility unit through the OPC UA protocol.

[0011] Furthermore, in the building integrated photovoltaic-energy storage-charging energy management system of the present invention, the feedback module is further configured to: Collect the actual output data of the photovoltaic power generation unit, the real-time state of charge data of the energy storage unit, and the power consumption data of the charging facility unit; Calculate the power deviation rate according to the deviation between the actual output data and the predicted value in the real-time scheduling strategy generated by the scheduling optimization module; Generate a feedback correction coefficient based on the power deviation rate and the energy storage response delay obtained from the data fusion module; Input the feedback correction coefficient into the scheduling optimization module to trigger the parameter update of the multi-objective optimization function.

[0012] Furthermore, the building integrated photovoltaic-energy storage-charging energy management system of the present invention further includes: Perform a moving average filter on the deviation between the actual output data and the predicted value, and set the length of the filter window according to the sampling period of the photovoltaic output time series data; Calculate the power compensation amount according to the filtered deviation and the rate of change of the state of charge of the energy storage obtained by extracting from the dynamic response vector of the data fusion module; Generate the feedback correction coefficient based on the power compensation amount.

[0013] Furthermore, the building integrated photovoltaic-energy storage-charging energy management system of the present invention further includes: a safety verification module for performing overload protection verification and protocol compliance verification on the control instructions distributed by the instruction generation module; When the safety verification module detects that the power value in the control instruction exceeds the rated capacity of the device obtained by the data acquisition module, it triggers the scheduling optimization module to regenerate a limited amplitude scheduling strategy based on the multi-objective optimization function.

[0014] Advantages of the present invention; The beneficial effects of the present invention are as follows. Through the protocol unification and dynamic fusion mechanism of multi-source heterogeneous data, the problems of data heterogeneity and timing differences in the photovoltaic, energy storage, and power grid links are effectively solved. The protocol conversion module uses Modbus, CAN bus, and IEC 61850 protocols to parse photovoltaic output, energy storage status, and power grid dispatching instructions, generating standardized data streams, eliminating differences in transmission protocols and sampling periods, and improving data processing efficiency. The data fusion module generates a multi-dimensional coupling state matrix by intercepting with a sliding window and a dynamic weighting model, calculating the weight coefficient based on the photovoltaic volatility and energy storage response delay, realizing the quantitative characterization of the coordinated state of the photovoltaic and energy storage, and providing high-precision input for scheduling optimization. The scheduling optimization module constructs a multi-objective function and uses an improved particle swarm algorithm to generate a Pareto optimal solution set, dynamically generating real-time strategies in combination with the power grid priority, and optimizing the balance of the photovoltaic accommodation rate, energy storage loss, and frequency modulation deviation. The feedback module iteratively adjusts the optimization parameters through a deviation correction coefficient, and the safety verification module restricts the operating boundaries of the equipment, forming a closed-loop control link, ultimately improving the system resource coordination efficiency and economy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the present invention, the accompanying drawings required for use in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the drawings.

[0016] Figure 1 FIG. is a system architecture diagram of a building integrated photovoltaic, energy storage, and charging energy management system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. To better understand the objectives of the present invention, the present invention is further described in detail below.

[0018] Please refer to Figure 1 , a building integrated photovoltaic, energy storage, and charging energy management system provided by the present invention includes: A data acquisition module for acquiring multi-source heterogeneous data, where the multi-source heterogeneous data includes photovoltaic output time series data, energy storage charge and discharge state data, charging load demand data, and power grid dispatching instruction data; The data acquisition module collects the light intensity, component temperature, and inverter output parameters of the photovoltaic power generation unit in real time through distributed sensors and communication interfaces, forming the photovoltaic output time series data. The photovoltaic output data includes active power, reactive power, and voltage instantaneous values, which are transmitted to the protocol conversion module through the RS-485 bus. The sampling frequency is adaptively adjusted according to the light change rate, and high-frequency sampling of 1 second is used in strong fluctuation scenarios to capture the fast fluctuation characteristics.

[0019] The energy storage unit collects the single-cell voltage, total voltage, charge and discharge current, and temperature information of the battery pack through the battery management system (BMS), and generates charge and discharge state data in combination with the relay switch state. The BMS uses multi-channel parallel acquisition technology to update the state of health (SOH) and state of charge (SOC) parameters of the battery every 100 milliseconds. The data is transmitted through the CAN bus, and the transmission frame format follows the SAE J1939 standard. Redundant frames are filtered by identifiers during parsing to reduce the communication load.

[0020] The charging facility unit collects the charging demand power, user reservation information, and real-time load curve through the charging pile controller. The controller interacts with the central management system based on the Modbus TCP protocol, synchronizes the operating state of the charging pile every 5 seconds, including the connection state, charging mode, and fault code, and generates a dynamic load prediction curve in combination with the user reservation period, which is summarized into the charging load demand data.

[0021] The grid dispatching instruction data receives the grid connection power limit, frequency modulation signal, and market electricity price information issued by the superior energy management system through the communication gateway. The gateway uses the IEC 104 protocol to parse the dispatching instructions, extracts the grid-connected active power threshold, frequency regulation dead zone range, and economic dispatching coefficient. The instruction data is synchronized with the local collected data through the timestamp alignment mechanism to eliminate the timing deviation caused by communication delay. After each data source completes the acquisition, the module performs abnormal data cleaning, discards the abnormal sampling points beyond the equipment rated range, and supplements the missing data through the interpolation method to form a complete multi-source heterogeneous data set.

[0022] The protocol conversion module is used to perform protocol parsing on the multi-source heterogeneous data received from the data acquisition module, parse the photovoltaic output time series data through the Modbus protocol, the energy storage charge and discharge state data through the CAN bus protocol, and the grid dispatching instruction data through the IEC 61850 protocol to generate a standardized data stream in a unified data format; After receiving the photovoltaic output time series data transmitted by the data acquisition module, the protocol conversion module decodes the data frame using the Modbus protocol parsing mechanism. The data frame structure locates the active power, reactive power, and instantaneous voltage values output by the inverter based on the register address mapping table. Register addresses 0x0001 to 0x0003 correspond to the high-order, low-order, and checksum fields of the active power respectively. During the parsing process, the CRC checksum of the data frame is verified. If the check fails, the retransmission request mechanism is triggered, the invalid data packet is discarded, and an exception log is recorded. The voltage fluctuation characteristics are calculated by the peak-to-peak value of the continuous voltage sampling values within the sliding window. The window length matches the sampling frequency of the photovoltaic output, reflecting the short-term fluctuation amplitude characteristics.

[0023] For the energy storage charge and discharge state data, the protocol conversion module parses the multiplexed signals in the data field based on the CAN bus protocol. The state of charge (SOC) of the battery is obtained by parsing the standard data frame corresponding to the CAN ID 0x18FEF100. The data bytes are converted to the actual value according to the scaling factor and offset defined by the SAE J1939 protocol. For example, the 3rd - 4th bytes represent the SOC percentage value. The charge and discharge rate is calculated by parsing the raw data of the battery pack current sensor, combining the timestamps reported by the BMS to calculate the current difference between adjacent sampling points, and generating the current change gradient per unit time. During the parsing process, the non-destructive arbitration mechanism of the CAN bus is adapted to preferentially process high-priority data frames to avoid bus conflicts, and at the same time filter out invalid frames and noise interference data.

[0024] When parsing the power grid dispatching instruction data through the IEC 61850 protocol, the protocol conversion module calls the MMS (Manufacturing Message Specification) service to access the data set (DataSet) in the logical device model. The grid connection power threshold is extracted from the A-class data attributes of the logical node MMXU, and the floating-point value and unit identifier are parsed. For example, the VolMag attribute represents the voltage amplitude, and the HzMag represents the frequency reference value. The frequency modulation instruction parameters are obtained by parsing the FrequencySetpoint (frequency setpoint) and DeadBand (dead zone range) attributes in the logical node FSCH to obtain the frequency adjustment target value and the allowable deviation range. During the parsing process, the communication service mapping relationship described by the SCL (System Configuration Language) file is dynamically loaded to convert the abstract data type into the engineering value recognizable by the device. For example, the enumerated status quantity in the logical node is mapped to a binary control signal.

[0025] After the parsing is completed, the protocol conversion module performs data formatting processing on the photovoltaic power generation, voltage fluctuation characteristics, state of charge, charge and discharge rate, grid connection power threshold, and frequency modulation command parameters. The unified data format is represented in JSON-LD structure, and each parameter is appended with a source protocol type label, a collection timestamp, and a data confidence score label. The voltage fluctuation characteristics and the charge and discharge rate are converted into per-unit values through normalization. For example, the voltage peak-to-peak value is divided by the rated voltage value to eliminate the dimension difference. The formatted data stream is transmitted to the data fusion module through the message queue middleware. The queue adopts a priority scheduling strategy. The grid dispatching instruction data is marked as high priority to ensure real-time performance. The photovoltaic and energy storage data are arranged in timestamp order to avoid fusion errors caused by time sequence misalignment.

[0026] A data fusion module, configured to perform sliding window interception on the standardized data stream received from the protocol conversion module, extract a fluctuation feature vector in the photovoltaic output time series data and a dynamic response vector in the energy storage charge and discharge state data, and input the fluctuation feature vector and the dynamic response vector into a dynamic weighted model constructed based on the photovoltaic output volatility and the energy storage response delay for time-domain convolution operation to generate a multi-dimensional coupled state matrix; After receiving the standardized data stream output by the protocol conversion module, the data fusion module dynamically divides data segments using the sliding window interception technique. The window length is adaptively adjusted according to the fluctuation period of the photovoltaic output time series data: when it is detected that the fluctuation period is less than 10 seconds, the window length is set to 30 seconds to capture high-frequency fluctuation characteristics; when the fluctuation period exceeds 60 seconds, the window is extended to 5 minutes to analyze the trend change. The window sliding step is synchronized with the data sampling interval to avoid redundant calculations caused by data overlap and maintain the continuity of the time series.

[0027] The feature extraction unit performs discrete wavelet transform on the photovoltaic output time series data within the window, decomposing the high-frequency detail component and the low-frequency approximation component. The fundamental wave amplitude is calculated by the mean value of the low-frequency component, the harmonic distortion rate is obtained by analyzing the energy ratio of the high-frequency component, and the fluctuation frequency is counted by the zero-crossing detection algorithm for the number of fluctuations per unit time, jointly constituting the fluctuation feature vector. The energy storage charge and discharge state data are processed within the same time window, and the dynamic response vector is generated by calculating the instantaneous change rate of the state of charge (SOC), the rise time of the charge and discharge current, and the temperature gradient. Among them, the SOC change rate is calculated by the difference method for the capacity difference between adjacent timestamps, and the current rise time is obtained by measuring the delay after the step response is triggered.

[0028] The dynamic weighting model dynamically allocates weight coefficients based on the ratio of the volatility of photovoltaic power output to the response delay of energy storage. The volatility is calculated by the variance of the fundamental wave amplitude in the fluctuation eigenvector, reflecting the instability degree of power output; the response delay is evaluated by the sliding average of the current rise time in the dynamic response vector, characterizing the energy storage regulation efficiency. When the volatility exceeds 15%, the model increases the photovoltaic feature weight to 0.7 to strengthen fluctuation suppression; when the response delay exceeds 500 milliseconds, the energy storage feature weight increases to 0.6 to optimize the regulation compensation. The weighted eigenvector is fused along the feature axis direction through tensor splicing technology to form a three-dimensional coupled feature tensor containing time series, feature categories, and weight dimensions.

[0029] In the matrix processing stage, the three-dimensional tensor is sliced and expanded along the time axis, and each time slice corresponds to a two-dimensional feature matrix. The row direction of the matrix represents the parameter correlation relationship between the photovoltaic fluctuation amplitude, harmonic distortion rate, and the change rate of energy storage SOC and current response time, and the column direction records the evolution path of the same parameter in different time windows. The continuous time slices are stacked in the acquisition order to form a multi-dimensional coupled state matrix with three-dimensional structure of rows, columns, and time. This matrix provides structured input data containing device interaction characteristics and time evolution laws for the scheduling optimization module, supporting the collaborative optimization of the photovoltaic accommodation rate, energy storage loss, and frequency modulation deviation in the multi-objective optimization model.

[0030] The scheduling optimization module is used to construct a multi-objective function with the photovoltaic accommodation rate, energy storage cycle loss, and grid frequency modulation deviation as optimization objectives according to the multi-dimensional coupled state matrix received from the data fusion module and the grid scheduling instruction data received from the data acquisition module, and solve it using an improved particle swarm optimization algorithm with the inertia weight dynamically adjusted according to the volatility of photovoltaic power output to generate a real-time scheduling strategy; After receiving the multi-dimensional coupled state matrix transmitted by the data fusion module, the scheduling optimization module extracts the predicted value of photovoltaic power output and the remaining capacity parameter of energy storage from the row and column dimensions of the matrix. The predicted value of photovoltaic power output is generated through time series regression analysis of the fluctuation eigenvector, using an adaptive sliding window mechanism to match the photovoltaic fluctuation period, and training a linear model based on historical fluctuation characteristics to predict the power output trend in the future time window. The remaining capacity of energy storage is calculated by combining the real-time value of the state of charge in the dynamic response vector with the battery aging coefficient and temperature attenuation factor, and generating available energy reserve data after correcting the capacity estimation error.

[0031] During the construction of the multi-objective optimization function, the PV accommodation rate target is defined as maximizing the matching degree between PV output and local load and energy storage charging power. The matching degree is calculated through the phase consistency and amplitude coincidence degree of the output curve and the demand curve. The energy storage cycle loss target is achieved by minimizing the weighted sum of the charge-discharge depth and the current change rate. The charge-discharge depth is calculated based on the amplitude change of the state of charge, and the loss coefficient is corrected by combining the battery temperature parameter. The grid frequency regulation deviation target is achieved by constraining the root mean square error between the actual output frequency of the system and the dispatching instruction frequency. The frequency deviation data is synchronously collected from the grid-side synchronization device and dynamically updated to the optimization model.

[0032] The inertia weight adjustment mechanism of the improved particle swarm algorithm is dynamically associated with the PV output volatility. The volatility is calculated through the variance of the PV characteristic parameters in the multi-dimensional coupling state matrix. When the volatility increases by 5%, the inertia weight linearly decreases by 0.1 to narrow the global search range and accelerate the algorithm convergence; when the volatility decreases, the weight gradually recovers to enhance the global optimization ability under steady-state conditions. In the algorithm initialization stage, the non-dominated sorting strategy is used to generate the initial Pareto solution set. During the iteration process, the crowding distance is calculated to maintain the diversity of the solution set distribution and avoid the aggregation of local optimal solutions. The screening of the Pareto solution set is based on the economic dispatching coefficient, frequency regulation urgency, and market electricity price fluctuation parameters embedded in the grid dispatching instruction, generating a weighted scoring table for each optimization item in the objective function. The solution with the highest score is output as the real-time dispatching strategy.

[0033] After the real-time dispatching strategy is generated, the module executes a secondary verification process, comparing the power distribution values in the strategy with the real-time state of charge of the energy storage unit and the actual sampled values of the PV output. If it is detected that the remaining capacity of the energy storage is lower than the strategy demand value or the PV prediction value deviates from the actual value by more than 10%, the strategy backtracking mechanism is triggered, and the historical coupling state matrix is called again to iteratively generate a corrected dispatching strategy. Before the corrected strategy is distributed by the instruction generation module, the dynamic weight coefficient in the multi-dimensional coupling state matrix is synchronously updated to form a closed-loop optimization link, maintaining the dispatching stability and self-adaptability of the system under the dynamic change of multi-source data.

[0034] The instruction generation module is used to convert the charge-discharge power values in the real-time dispatching strategy output by the dispatching optimization module into PWM modulation signals, and distribute the PWM modulation signals to the inverter in the PV power generation unit, the BMS system in the energy storage unit, and the charging pile controller in the charging facility unit through the OPC UA protocol; After the instruction generation module receives the real-time scheduling strategy output by the scheduling optimization module, the instruction encoding unit normalizes the charge and discharge power values. The normalization process is based on the rated output power of the PV inverter, the maximum charge and discharge rate of the energy storage battery pack, and the load capacity of the charging pile, and converts the power distribution value into a per-unit value in the range of 0-1. The per-unit value is mapped to a duty cycle parameter through a pulse width modulation (PWM) encoder. The duty cycle has a linear relationship with the target power value. For example, a charge and discharge power of 2 MW for the energy storage corresponds to a 100% duty cycle, generating a modulation signal waveform compatible with the interfaces of the inverter, BMS, and charging pile controller. During the encoding process, a slope limit is imposed on the mutation amplitude of the duty cycle. For example, the maximum change rate per second does not exceed 5%, to avoid equipment overload or voltage fluctuations caused by power instruction jumps.

[0035] The instruction distribution unit establishes a communication link with the execution device through the OPC UA protocol and defines a unified information model. The PWM modulation signal is encapsulated as an OPC UA node attribute. The node includes signal amplitude, frequency, effective timestamp, and target device address identifier. For example, the inverter node address is marked as "PV_Inverter_01", and the energy storage BMS node is marked as "ESS_BMS_01". The distribution process adopts a publish-subscribe mode. The grid scheduling instruction data is marked as high priority, and an independent transmission time slot is allocated through the time-sensitive network (TSN) extension function to ensure the real-time performance of critical instructions. The transmission delay is controlled within 50 milliseconds, and the clock drift between devices is eliminated through periodic network clock calibration to maintain the synchronous execution of multi-device instructions.

[0036] Protocol compliance verification is performed before signal transmission to verify whether the node namespace conforms to the OPC UA specification and whether the data fields match the communication interface definition of the target device. For example, verify whether the PWM signal frequency of the charging pile controller is within the allowable range of 1 kHz - 10 kHz, and whether the duty cycle is within the valid range of 0 - 100%. The signal node that fails the verification triggers an alarm log and notifies the scheduling optimization module to regenerate the strategy, and at the same time discards the abnormal instruction. After the transmission is completed, the module receives the status confirmation message returned by the execution device. The message includes the instruction execution result and real-time operating parameters, such as the actual charge and discharge current feedback by the BMS and the output power reported by the inverter, forming a closed-loop feedback link to verify the effectiveness of the instruction execution.

[0037] The exception handling mechanism is embedded in the instruction generation process. When the deviation between the charging and discharging power fed back by the energy storage BMS and the instruction exceeds 5%, the module immediately triggers the instruction rollback process, restores the power value to the previous valid state, and sends an interrupt request to the scheduling optimization module. If the charging pile controller detects that the charging demand does not match the power instruction, for example, the user terminates the charging in advance, it starts the local priority arbitration, dynamically adjusts the power distribution according to the real-time load, and reports the corrected load parameters to the instruction generation module at the same time, supporting the system to dynamically reconstruct the scheduling strategy and maintaining the robustness of multi-device collaborative control.

[0038] The feedback module is used to monitor the actual output data of the photovoltaic power generation unit, the real-time state of charge data of the energy storage unit, and the power consumption data of the charging facility unit, calculate the deviation between the actual output data and the predicted value in the real-time scheduling strategy, generate a feedback correction coefficient, and input the feedback correction coefficient into the scheduling optimization module to trigger the iterative optimization of the multi-objective function and adjust the real-time scheduling strategy.

[0039] The feedback module periodically collects the actual output data of the photovoltaic power generation unit through distributed sensors, including the active power, reactive power, and instantaneous voltage output by the inverter. The data is transmitted through the Modbus protocol and attached with a time stamp accurate to the millisecond level. The real-time state of charge (SOC) data of the energy storage unit is extracted from the CAN bus communication frame of the BMS. When parsing, invalid frames are filtered and the actual value is converted based on the SAE J1939 protocol. The capacity estimation deviation is corrected by combining the battery temperature parameter. The power consumption data of the charging facility unit is obtained through the Modbus TCP interface of the charging pile controller, synchronizing the user reservation information and the real-time load curve to generate a dynamic load data set.

[0040] The root mean square error (RMSE) algorithm within a sliding window is used to calculate the power deviation. The window length is the same as the prediction period of the scheduling optimization module. The algorithm calculates the difference between the actual output data and the predicted value point by point, accumulates the squares, takes the average, and then takes the square root to generate a scalar index reflecting the overall deviation level. A moving average filter is introduced in the calculation process. The window length is dynamically adjusted according to the photovoltaic output sampling frequency: a 5-second short window is used to filter out instantaneous noise in the high-frequency sampling scenario, and a 30-second long window is used to retain the trend deviation characteristics in the low-frequency scenario. The filtered deviation data is combined with the energy storage response delay parameter, and the delay is extracted from the dynamic response vector, which characterizes the compensation efficiency of the system for historical deviations.

[0041] The feedback correction coefficient is calculated by multiplying the deviation rate by the time delay. The deviation rate is normalized to a per-unit value in the range of 0-1. When the energy storage response time delay exceeds 500 milliseconds, the correction coefficient is increased proportionally to enhance the compensation intensity. When the deviation rate remains high, a dynamic attenuation factor is introduced, and the factor value increases with the number of consecutive deviation cycles to suppress overcorrection caused by short-term disturbances. After the correction coefficient is input into the scheduling optimization module, it triggers the iterative update of the multi-objective function weight parameters: if the deviation is mainly caused by the fluctuation of photovoltaic power output, the weight coefficient of the photovoltaic accommodation rate target is increased; if the deviation is due to the lag of energy storage response, the optimization weight of the energy storage cycle loss target is increased.

[0042] The closed-loop control mechanism constructs an empirical database through the mapping relationship between the historical correction coefficient and the policy execution effect. The optimization module preferentially calls the weight combination of similar scenarios in subsequent iterations to accelerate convergence. If the deviation rate does not drop below the set threshold after three consecutive corrections, the module triggers an abnormal warning process and notifies the data fusion module to recalibrate the parameters of the feature extraction model, such as adjusting the sliding window length or the weight allocation logic of the dynamic weighting model, to optimize the state representation accuracy from the data source and maintain the adaptive ability of the system under complex working conditions.

[0043] The data acquisition module uses distributed sensors and communication interfaces to collect the light intensity, component temperature, and inverter output parameters of the photovoltaic power generation unit in real time to form the photovoltaic power output time series data. The energy storage unit uses the battery management system (BMS) to collect the battery pack voltage, current, and temperature information, and combines the charge and discharge switch states to generate the energy storage charge and discharge state data. The charging facility unit uses the charging pile controller to collect the charging demand power, user reservation information, and real-time load curve, and summarizes them into the charging load demand data. The grid dispatching instruction data receives the grid connection power limit, frequency modulation signal, and market electricity price information issued by the superior dispatching system through the communication gateway. Each data source aligns the data time series through the timestamp synchronization mechanism to eliminate the time series misalignment problem caused by the sampling period difference.

[0044] The protocol conversion module uses the Modbus protocol to parse the photovoltaic power output time series data and extracts the active power, reactive power, and voltage fluctuation characteristics output by the inverter. The energy storage charge and discharge state data is parsed through the CAN bus protocol to extract the state of charge (SOC), charge and discharge current rate, and state of health (SOH) parameters of the battery. The grid dispatching instruction data is parsed through the IEC 61850 protocol to extract the grid connection power threshold, amplitude, and frequency response requirements of the frequency modulation instruction. During the parsing process, the protocol conversion module performs integrity verification on the data frame, discards abnormal or verification-failed data packets, and performs standardized mapping on the data fields of different protocols to generate a standardized data stream with a unified timestamp, data label, and numerical range.

[0045] The data fusion module adopts a sliding window interception technique for the standardized data stream, and the window length is dynamically adjusted according to the fluctuation period of the photovoltaic output data. The feature extraction unit performs Fourier transform and wavelet decomposition on the photovoltaic output time series data within the window, extracts the fundamental wave component and harmonic component features, and constructs a fluctuation feature vector reflecting the fluctuation amplitude and frequency of the output. The dynamic response vector is calculated from the SOC change rate, charge and discharge current response time, and temperature change gradient in the charge and discharge state data of the energy storage. The coupling analysis unit inputs the fluctuation feature vector and the dynamic response vector into the dynamic weighting model, which dynamically allocates weight coefficients based on the variance calculation result of the photovoltaic output volatility and the reciprocal of the energy storage response delay, and generates a coupled feature tensor through weighted superposition. Matrix processing unfolds the tensor in the time dimension to form a multi-dimensional coupled state matrix containing spatio-temporal correlation features.

[0046] The scheduling optimization module constructs a multi-objective optimization function based on the photovoltaic output prediction value and the remaining capacity of the energy storage in the multi-dimensional coupled state matrix, combined with the grid connection power threshold in the grid scheduling instruction. The photovoltaic accommodation rate target is defined by maximizing the matching degree between the photovoltaic output, the load demand, and the energy storage charging power; the energy storage cycle loss target is defined by minimizing the charge and discharge depth and the current change rate; the grid frequency modulation deviation target is defined by minimizing the frequency deviation between the actual output and the frequency modulation instruction. The inertia weight of the improved particle swarm algorithm is linearly adjusted with the photovoltaic output volatility, and the global search weight is reduced when the volatility is high to enhance the local optimization ability. After the algorithm solves to generate the Pareto optimal solution set, the comprehensive optimal solution is selected according to the grid priority weight as the real-time scheduling strategy.

[0047] The instruction generation module converts the charge and discharge power distribution value in the real-time scheduling strategy into a modulation signal with adjustable duty cycle through a pulse width modulation (PWM) encoder. During the encoding process, the charge and discharge power value is normalized according to the rated capacity of the device to generate PWM waveform parameters compatible with the interfaces of the inverter, BMS system, and charging pile controller. The instruction distribution unit establishes a two-way communication link through the OPC UA protocol, synchronously sends the modulation signal to each execution unit, and receives the device status confirmation message to achieve the real-time and reliable transmission of the instruction.

[0048] The feedback module periodically obtains the actual output data of the photovoltaic power generation unit, the real-time SOC data of the energy storage unit, and the actual power consumption data of the charging facilities through the data acquisition module. The power deviation rate is calculated from the root mean square error between the actual output and the predicted value, and combined with the propagation influence of the energy storage response delay on the deviation, a feedback correction coefficient is generated. After the correction coefficient is input into the scheduling optimization module, it triggers the iterative update of the weight parameters in the multi-objective function, dynamically adjusts the optimization direction of the subsequent scheduling strategy, and forms a closed-loop control mechanism.

[0049] The safety verification module performs overload protection verification on the control instructions issued by the instruction generation module, compares the difference between the power distribution value and the rated capacity of the device, and triggers the amplitude limiting strategy if it exceeds the safety threshold. The protocol compliance verification is based on the OPC UA communication specification to verify the integrity of the instruction fields and the legality of the data format. When the verification fails, the safety verification module sends an interrupt signal to the scheduling optimization module, requesting to regenerate a scheduling strategy that conforms to the device operation boundary to avoid device damage or system failure caused by incorrect instructions.

[0050] Specifically, for the building integrated photovoltaic, energy storage and charging energy management system of the present invention, the protocol conversion module is further configured to: Perform Modbus protocol parsing on the photovoltaic output time series data received from the data acquisition module, and extract the photovoltaic power generation power and voltage fluctuation characteristics; Perform CAN bus protocol parsing on the energy storage charge and discharge state data received from the data acquisition module, and extract the state of charge and charge and discharge rate; Perform IEC 61850 protocol parsing on the grid dispatching instruction data received from the data acquisition module, and extract the grid connection power threshold and frequency modulation instruction parameters; Take the photovoltaic power generation power, voltage fluctuation characteristics, state of charge, charge and discharge rate, grid connection power threshold and frequency modulation instruction parameters as the parsing results, and output them to the data fusion module.

[0051] After receiving the photovoltaic output time series data transmitted by the data acquisition module, the protocol conversion module decodes the data frame using the Modbus protocol parsing mechanism. The data frame structure includes a register address mapping table for locating the active power, reactive power and instantaneous voltage values output by the inverter. During the parsing process, the protocol conversion module verifies the CRC checksum of the data frame, and triggers a retransmission request if the verification fails, discarding the invalid data packet. The voltage fluctuation characteristics are calculated by the peak-to-peak value and the change rate of the voltage sampling values within the sliding window, reflecting the short-term fluctuation characteristics of the photovoltaic output.

[0052] For the energy storage charge and discharge state data, the protocol conversion module parses the multiplexed signals in the data field based on the CAN bus protocol. The state of charge (SOC) of the battery is obtained by parsing the standard data frame corresponding to the CAN ID, and the data format follows the scaling factor and offset defined by the SAE J1939 protocol. The charge and discharge rate is calculated by parsing the original data of the battery pack current sensor and combining the timestamp reported by the BMS to calculate the current change gradient per unit time, generating dynamic response parameters. During the parsing process, the module adapts to the arbitration mechanism of the CAN bus to avoid parsing delays caused by data conflicts.

[0053] When the power grid dispatching instruction data is parsed through the IEC 61850 protocol, the protocol conversion module calls the MMS (Manufacturing Message Specification) service to access the logical device model. The grid connection power threshold extracts the Class A data attributes of the logical node MMXU from the DataSet (DataSet), and parses its floating-point value and unit identifier. The frequency modulation instruction parameters obtain the frequency regulation target and the allowable deviation range by parsing the FrequencySetpoint (frequency setpoint) and DeadBand (deadband range) attributes in the logical node FSCH. During the parsing process, the module dynamically loads the communication service mapping relationship described in the SCL (System Configuration Language) file to achieve accurate conversion of protocol semantics.

[0054] After the parsing is completed, the protocol conversion module performs data formatting processing on the photovoltaic power generation, voltage fluctuation characteristics, state of charge, charge and discharge rate, grid connection power threshold, and frequency modulation instruction parameters. The unified data format is represented in JSON-LD structure, and each parameter is attached with the protocol type of the data source, timestamp, and data confidence label. The voltage fluctuation characteristics and charge and discharge rate parameters are converted into per-unit values through normalization to eliminate the dimension difference. The formatted data stream is transmitted to the data fusion module through the message queue. The queue adopts a priority scheduling strategy to ensure the priority transmission of the power grid dispatching instruction data and maintain the real-time performance of the system response.

[0055] Specifically, for the building integrated photovoltaic energy storage and charging management system described in the present invention, the data fusion module includes: A feature extraction unit, configured to perform sliding window interception on the standardized data stream received from the protocol conversion module, and extract the fluctuation feature vector in the photovoltaic output time series data and the dynamic response vector in the energy storage charge and discharge state data; A coupling analysis unit, configured to input the fluctuation feature vector and the dynamic response vector into a dynamic weighted model constructed based on the photovoltaic output volatility and the energy storage response delay to generate the multi-dimensional coupling state matrix.

[0056] After the data fusion module receives the standardized data stream output by the protocol conversion module, the feature extraction unit dynamically sets the sliding window length according to the fluctuation period of the photovoltaic output data. The window length is inversely proportional to the sampling frequency of the photovoltaic output. A short window is used in the high-frequency fluctuation scenario to capture the fast-changing features, and a long window is used in the low-frequency scenario to enhance the stability of trend analysis. The window sliding step is set to an integer multiple of the sampling interval to avoid redundant calculations caused by data overlap. For the photovoltaic output time series data within the window, the feature extraction unit decomposes the high-frequency detail component and the low-frequency approximation component through discrete wavelet transform, and extracts the fundamental wave amplitude, harmonic distortion rate, and fluctuation frequency as the core parameters of the fluctuation feature vector.

[0057] The energy storage charge and discharge state data is synchronously processed within the same time window, and the dynamic response vector is generated by calculating the instantaneous change rate of the state of charge (SOC), the rise time of the charge and discharge current, and the temperature change gradient. The SOC change rate is calculated by the difference method for the capacity difference between adjacent timestamps, and the charge and discharge current response time is obtained by measuring the delay after the step signal is triggered. The feature extraction unit aligns the above parameters in the time domain, eliminates the timing deviation caused by communication delay, and generates a dynamic response vector that strictly matches the timestamp of the fluctuation feature vector.

[0058] The coupling analysis unit inputs the fluctuation feature vector and the dynamic response vector into the dynamic weighting model. The photovoltaic output volatility is calculated by the variance of the fundamental wave amplitude in the fluctuation feature vector, reflecting the degree of output instability; the energy storage response delay is evaluated by the sliding average of the current rise time in the dynamic response vector, characterizing the dynamic regulation ability of the system. The dynamic weighting model allocates weight coefficients according to the reciprocal ratio of the volatility and the response delay, increasing the photovoltaic feature weight when the volatility increases and increasing the energy storage feature weight when the response delay increases. The weighted feature vectors are fused through the tensor splicing technology to form a coupling feature tensor containing spatio-temporal correlation characteristics.

[0059] In the matrix processing stage, the coupling feature tensor is unfolded along the time dimension, and each time slice corresponds to a two-dimensional feature matrix. The rows of the matrix represent the types of characteristic parameters of the photovoltaic and the energy storage, and the columns represent the evolution sequences of the same parameter in different time windows. The multi-dimensional coupling state matrix is constructed by stacking the three-dimensional structures of consecutive time slices, and the row, column, and depth dimensions respectively represent the feature type, time series, and device coupling relationship. This matrix provides input data containing spatio-temporal evolution laws and device interaction characteristics for the scheduling optimization module, supporting the accurate solution of the multi-objective optimization model.

[0060] Specifically, for the building integrated photovoltaic-energy storage-charging energy management system described in the present invention, the coupling analysis unit further includes: The photovoltaic output volatility is calculated by the variance of the fluctuation feature vector; The energy storage response delay is calculated by the time delay parameter in the dynamic response vector; The dynamic weighting model dynamically adjusts the weight coefficients according to the ratio of the photovoltaic output volatility and the energy storage response delay, performs weighted superposition on the fluctuation feature vector and the dynamic response vector, and generates a coupling feature tensor; The coupling feature tensor is subjected to matrix processing to obtain the multi-dimensional coupling state matrix.

[0061] After receiving the fluctuation feature vector and the dynamic response vector output by the feature extraction unit, the coupling analysis unit calculates the variance of the photovoltaic output volatility through the time series data of the fundamental wave component in the fluctuation feature vector. The variance calculation uses the degree of dispersion of the sampled values within a sliding window. The window size is consistent with the analysis period set by the feature extraction unit, reflecting the stability level of the photovoltaic output in a short time domain. The calculation results of the volatility are stored as a continuous scalar sequence according to the time stamp, which is used to quantify the driving intensity of the dynamic characteristics of the photovoltaic output on the system regulation demand.

[0062] The energy storage response delay is calculated through the time delay parameter in the dynamic response vector, which is extracted from the step response curve of the energy storage charge and discharge current. After the current step is triggered, the system records the time difference required for the current value to reach 90% of the target value from the initial state as the delay measurement value for a single response. The delay measurement results of multiple consecutive time windows are processed by moving average filtering to generate a delay sequence reflecting the dynamic performance of the energy storage system. The delay sequence and the photovoltaic volatility sequence are aligned by the time stamp to form an associated mapping of the dynamic characteristics between devices.

[0063] The dynamic weighting model dynamically adjusts the weight coefficient according to the ratio of the photovoltaic output volatility to the energy storage response delay. The ratio calculation uses the quotient of the absolute value of the volatility and the normalized delay. The normalized delay is linearly scaled by the maximum allowable response delay threshold. When the volatility increases or the delay decreases, the model increases the weight coefficient of the fluctuation feature vector to enhance the dominant role of the photovoltaic output fluctuation in the coupling analysis. When the delay increases or the volatility decreases, the weight coefficient of the dynamic response vector is increased to strengthen the compensation effect of the energy storage regulation ability. The weight coefficient is updated frame by frame according to the time window to generate a dynamic weighting parameter table matching the real-time state of the system.

[0064] During the weighted superposition process, the fluctuation feature vector and the dynamic response vector are multiplied by the corresponding weight coefficients respectively to form a weighted feature subspace. Each dimension of the features in the subspace is fused along the feature axis direction through tensor splicing technology to generate a coupling feature tensor containing the photovoltaic fluctuation characteristics and the energy storage response characteristics. The tensor structure retains the time series dimension and the feature category dimension of the original vector, and at the same time adds a weight dimension to record the dynamic adjustment process, constructing a three-dimensional feature expression space.

[0065] In the matrix processing stage, the three-dimensional coupled feature tensor is sliced and expanded along the time axis. For each time slice, the rows of the corresponding two-dimensional matrix represent the feature categories, and the columns represent the feature values after weight adjustment. The matrices of consecutive time slices are stacked in chronological order to form a multi-dimensional coupled state matrix with a three-dimensional structure of rows, columns, and time. The row direction of the matrix records the correlation relationship of the characteristic parameters of the photovoltaic and energy storage systems. The column direction characterizes the evolution path of the same parameter in different time windows, and the time dimension reflects the dynamic evolution process of the system state. This matrix provides structured input data containing the device interaction characteristics and time evolution laws for the scheduling optimization module, supporting the accurate solution of the multi-objective optimization model and the generation of real-time strategies.

[0066] Specifically, for the building integrated photovoltaic-energy storage-charging integrated energy management system of the present invention, the scheduling optimization module is further configured to: Generate a charge-discharge power distribution plan based on the predicted photovoltaic output value and the remaining energy storage capacity in the multi-dimensional coupled state matrix received from the data fusion module, wherein the predicted photovoltaic output value is calculated through the regression analysis of the fluctuation feature vector; Combine the grid connection power threshold in the grid scheduling instruction data received from the data acquisition module to perform constraint optimization on the charge-discharge power distribution plan. The constraint optimization is based on the multi-objective optimization function to generate the real-time scheduling strategy.

[0067] After the scheduling optimization module receives the multi-dimensional coupled state matrix transmitted by the data fusion module, the predicted photovoltaic output value is generated through the time series regression analysis of the fluctuation feature vector. The regression analysis adopts an adaptive sliding window mechanism, and the window length matches the fluctuation period of the photovoltaic output. A linear regression model is trained based on the historical fluctuation features to predict the output trend in the future time window. The remaining energy storage capacity extracts the real-time measurement value of the state of charge (SOC) from the dynamic response vector, and combines the battery aging coefficient and the temperature attenuation factor to correct the capacity estimation accuracy to generate the available energy reserve data of the energy storage system.

[0068] The charge-discharge power distribution plan is dynamically generated based on the predicted photovoltaic output value and the remaining energy storage capacity. The plan generation process adopts a hierarchical decision-making mechanism, which first matches the predicted photovoltaic output value with the charging load demand, and the remaining power is allocated to energy storage charging or grid connection and feeding. The upper limit of the energy storage charging power is calculated based on the product of the remaining capacity and the maximum charging rate, and the lower limit of the discharge power is dynamically set according to the grid frequency regulation demand and the load deficit. During the distribution process, the module establishes a power balance equation for the photovoltaic output, energy storage charge-discharge, and grid interaction power as the basic constraint condition.

[0069] The grid connection power threshold in the grid dispatching instruction data is embedded in the optimization model as a hard constraint. The grid connection power threshold is obtained by parsing the ActivePowerLimit (active power limit) attribute in the grid dispatching instruction, and the module sets the upper and lower boundaries of the grid connection power in the charge and discharge power distribution scheme. In the constraint optimization stage, the power balance equation, grid connection threshold limit and equipment operation boundary are integrated into a set of linear inequality constraints, and a multi-objective optimization function including the photovoltaic accommodation rate, energy storage cycle loss and grid frequency modulation deviation is constructed.

[0070] In the multi-objective optimization function, the photovoltaic accommodation rate target is defined by maximizing the matching degree between the photovoltaic output and the local load and energy storage charging; the energy storage cycle loss target is achieved by minimizing the weighted sum of the charge and discharge depth and the current change rate; the grid frequency modulation deviation target is quantified by minimizing the absolute value of the frequency deviation between the actual output and the frequency modulation instruction. The inertia weight of the improved particle swarm algorithm is dynamically adjusted according to the volatility of the photovoltaic output. When the volatility is high, the global search weight is reduced to enhance the local optimization efficiency and avoid the algorithm falling into a suboptimal solution.

[0071] During the solution process, the improved particle swarm algorithm generates a Pareto optimal solution set, and each solution corresponds to a power distribution strategy. The module performs a weighted scoring on the strategies in the solution set according to the priority weights embedded in the grid dispatching instruction. The priority weights are comprehensively calculated by parsing the economic dispatching coefficient and frequency modulation emergency parameter in the grid instruction, and the strategy with the highest score is output as the real-time dispatching strategy. The strategy includes the power setting value of the photovoltaic inverter, the charge and discharge power instruction of the energy storage and the grid connection power control parameter, forming an executable control instruction set.

[0072] Before the real-time dispatching strategy is converted into a device-level control signal by the instruction generation module, the module executes a secondary verification process. The verification process compares the power distribution value in the strategy with the real-time status data of the device. If it is detected that the remaining capacity of the energy storage is insufficient or the predicted value of the photovoltaic output deviates from the actual sampled value by more than the threshold, the strategy backtracking mechanism is triggered, and the historical coupling state matrix is called again to iteratively generate a corrected dispatching strategy to maintain the robustness and self-adaptability of the system operation.

[0073] Specifically, the building integrated photovoltaic-energy storage-charging energy management system described in the present invention further includes: Construct a multi-objective optimization function with the maximum photovoltaic accommodation rate, the minimum energy storage cycle loss and the minimum grid frequency modulation deviation based on the photovoltaic output prediction value in the multi-dimensional coupling state matrix received from the data fusion module and the weight coefficient output from the dynamic weighting model; The improved particle swarm optimization algorithm is used to solve the multi-objective optimization function. The inertia weight of the improved particle swarm optimization algorithm is dynamically adjusted according to the photovoltaic output volatility to generate a Pareto optimal solution set. The inertia weight of the improved particle swarm optimization algorithm is dynamically adjusted with the photovoltaic output volatility. Specifically, when the photovoltaic output volatility increases by 5% each time, the inertia weight linearly decreases by 0.1; when the volatility decreases, the inertia weight gradually returns to the initial value of 0.9.

[0074] According to the priority weight in the grid dispatching instruction data received from the data acquisition module, the real-time dispatching strategy is selected from the Pareto optimal solution set.

[0075] The dispatching optimization module extracts the photovoltaic output prediction value and the weight coefficient output by the dynamic weighting model based on the multi-dimensional coupling state matrix provided by the data fusion module. The photovoltaic output prediction value is generated by extrapolating the time series of the fluctuation feature vector, and the exponential smoothing method is used to correct the prediction deviation to improve the short-term prediction accuracy. The weight coefficient is obtained from the dynamic weighting model, recording the real-time coupling relationship between the photovoltaic volatility and the energy storage response delay, and serving as an important adjustment parameter for the objective term in the multi-objective optimization function.

[0076] In the construction process of the multi-objective optimization function, the photovoltaic accommodation rate target is defined by maximizing the matching degree between the photovoltaic output, the energy storage charging power, and the local load demand. The cosine similarity algorithm is used to calculate the matching degree, measuring the phase and amplitude consistency between the output curve and the demand curve. The energy storage cycle loss target is achieved by minimizing the product of the charge and discharge cycle times and the single-cycle depth. The cycle depth is calculated by the amplitude of the state of charge change and the loss coefficient is corrected by combining the battery temperature decay factor. The grid frequency modulation deviation target is quantified by minimizing the mean square error between the actual output frequency of the system and the dispatching instruction frequency. The frequency deviation data is obtained in real time from the grid-side synchronization device.

[0077] The inertia weight adjustment mechanism of the improved particle swarm optimization algorithm is dynamically associated with the photovoltaic output volatility. The volatility is calculated by the variance of the photovoltaic characteristic parameters in the multi-dimensional coupling state matrix. When the volatility increases, the inertia weight linearly decreases, restricting the global search range of the particle swarm to accelerate convergence; when the volatility decreases, the inertia weight gradually recovers, enhancing the global optimization ability of the algorithm under stable working conditions. In the initialization stage of the algorithm, a non-dominated sorting strategy is set to generate the initial Pareto front solution set, and the distribution diversity of the solution set is maintained by calculating the crowding distance during the iteration process.

[0078] The screening of the Pareto optimal solution set is performed according to the priority weights embedded in the power grid dispatching instructions. The priority weights are comprehensively calculated by analyzing the economic dispatching coefficient, frequency modulation response level, and market electricity price fluctuation parameters in the power grid instructions to generate a weighted scoring table for each optimization item in the objective function. In the solution set screening stage, multi-objective weighted scoring is performed on each Pareto solution, and the solution with the highest score is output as the real-time dispatching strategy. The strategy includes the power adjustment instructions of the photovoltaic inverter, the charge and discharge power curves of the energy storage system, and the grid connection power control threshold, forming a complete set of execution parameters.

[0079] After the real-time dispatching strategy is generated, the module performs a robustness check on the strategy. During the check process, the predicted charge and discharge power in the strategy is compared with the real-time state of charge of the energy storage unit. If it is detected that the remaining capacity is insufficient to support the execution of the strategy, the prediction model backtracking mechanism is triggered, and a new charge and discharge power distribution plan is regenerated based on the historical coupling state matrix. Before the final strategy is distributed by the instruction generation module, the module synchronously updates the weight coefficients in the multi-dimensional coupling state matrix to form a closed-loop optimization link and maintain the dispatching stability of the system under the dynamic changes of multi-source data.

[0080] Specifically, for the building integrated photovoltaic-energy storage-charging energy management system of the present invention, the instruction generation module includes: An instruction encoding unit, configured to convert the charge and discharge power values in the real-time dispatching strategy output by the dispatching optimization module into PWM modulation signals based on the Pareto optimal solution set generated by the dispatching optimization module; An instruction distribution unit, configured to synchronously send the PWM modulation signals to the BMS system in the energy storage unit and the charging pile controller in the charging facility unit through the OPC UA protocol.

[0081] Based on the multi-dimensional coupling state matrix provided by the data fusion module, the dispatching optimization module extracts the predicted photovoltaic output value and the weight coefficient output by the dynamic weighting model. The predicted photovoltaic output value is generated by extrapolating the time series of the fluctuation feature vector, and the exponential smoothing method is used to correct the prediction deviation to improve the short-term prediction accuracy. The weight coefficient is obtained from the dynamic weighting model, which records the real-time coupling relationship between the photovoltaic volatility and the energy storage response delay, and serves as an important adjustment parameter for the objective items in the multi-objective optimization function.

[0082] During the construction of the multi-objective optimization function, the PV accommodation rate target is defined by maximizing the matching degree between PV output, energy storage charging power, and local load demand. The cosine similarity algorithm is used to calculate the matching degree, which measures the phase and amplitude consistency between the output curve and the demand curve. The energy storage cycle loss target is achieved by minimizing the product of the charge-discharge cycle times and the single-cycle depth. The cycle depth is calculated based on the amplitude of the state of charge change and combined with the battery temperature decay factor to correct the loss coefficient. The grid frequency regulation deviation target is quantified by minimizing the mean square error between the actual output frequency of the system and the dispatching instruction frequency. The frequency deviation data is obtained in real-time from the grid-side synchronization device.

[0083] The inertia weight adjustment mechanism of the improved particle swarm algorithm is dynamically associated with the PV output volatility. The volatility is calculated by the variance of the PV characteristic parameters in the multi-dimensional coupled state matrix. When the volatility increases, the inertia weight decreases linearly, restricting the global search range of the particle swarm to accelerate convergence; when the volatility decreases, the inertia weight gradually recovers, enhancing the global optimization ability of the algorithm under stable operating conditions. In the initialization stage of the algorithm, a non-dominated sorting strategy is set to generate the initial Pareto front solution set. During the iteration process, the diversity of the solution set is maintained by calculating the crowding distance.

[0084] The screening of the Pareto optimal solution set is executed according to the priority weights embedded in the grid dispatching instructions. The priority weights are comprehensively calculated by analyzing the economic dispatching coefficient, frequency modulation response level, and market electricity price fluctuation parameters in the grid instructions, generating a weighted scoring table for each optimization item in the objective function. In the solution set screening stage, multi-objective weighted scoring is performed on each Pareto solution, and the solution with the highest score is output as the real-time dispatching strategy. The strategy includes the power regulation instructions of the PV inverter, the charge-discharge power curve of the energy storage system, and the grid-connected power control threshold, forming a complete set of execution parameters.

[0085] After the real-time dispatching strategy is generated, the module performs a robustness check on the strategy. During the check process, the predicted charge-discharge power in the strategy is compared with the real-time state of charge of the energy storage unit. If it is detected that the remaining capacity is insufficient to support the execution of the strategy, the prediction model backtracking mechanism is triggered, and a new charge-discharge power distribution plan is generated based on the historical coupled state matrix. Before the final strategy is distributed by the instruction generation module, the module synchronously updates the weight coefficients in the multi-dimensional coupled state matrix, forming a closed-loop optimization link to maintain the dispatching stability of the system under the dynamic change of multi-source data.

[0086] Specifically, for the building integrated PV, energy storage and charging energy management system of the present invention, the feedback module is further used for: Collecting the actual output data of the PV power generation unit, the real-time state of charge data of the energy storage unit, and the power consumption data of the charging facility unit; Calculating the power deviation rate according to the deviation between the actual output data and the predicted value in the real-time dispatching strategy generated by the dispatching optimization module; Generate a feedback correction coefficient based on the power deviation rate and the energy storage response delay obtained from the data fusion module; Input the feedback correction coefficient into the scheduling optimization module to trigger the parameter update of the multi-objective optimization function.

[0087] The feedback module periodically obtains the actual output data of the photovoltaic power generation unit through the distributed data acquisition interface, and the data acquisition frequency is consistent with the prediction period of the scheduling optimization module. The actual output data includes the active power, reactive power, and instantaneous grid-connected voltage output by the inverter, and is matched with the predicted value in the real-time scheduling strategy through the timestamp alignment mechanism. The real-time state of charge (SOC) data of the energy storage unit is extracted from the CAN bus communication frame of the battery management system (BMS), and the capacity estimation error is corrected by combining the battery temperature and state of health (SOH) parameters. The power consumption data of the charging facility unit is generated by fusing the load curve reported by the charging pile controller and the user reservation information, reflecting the difference between the real-time charging demand and the planned load.

[0088] The power deviation rate is calculated using the root mean square error (RMSE) algorithm within a sliding window. The window length is the same as the backtracking period of the photovoltaic output prediction model. The algorithm calculates the point-by-point difference between the actual output data and the predicted value, accumulates the squares, takes the average, and then takes the square root to generate a scalar index reflecting the overall deviation level. A moving average filter is introduced during the deviation calculation process to filter out the instantaneous abnormal fluctuations caused by sudden changes in light or equipment noise, and retain the trend deviation characteristics. The length of the filter window is dynamically adjusted according to the sampling interval of the photovoltaic output data, and a short window is used in high-frequency sampling scenarios to reduce the calculation delay.

[0089] The generation of the feedback correction coefficient is based on the synergistic effect of the power deviation rate and the energy storage response delay. The energy storage response delay is extracted from the dynamic response vector of the data fusion module, characterizing the compensation efficiency of the system for historical deviations. The correction coefficient is calculated by multiplying the deviation rate by the delay. When the deviation rate increases or the delay rises, the coefficient value is correspondingly increased to enhance the adjustment amplitude of the optimization parameters. A dynamic attenuation factor is introduced during the calculation process, and the factor value increases with the cumulative number of consecutive deviation periods, avoiding overcorrection caused by short-term disturbances.

[0090] After the feedback correction coefficient is input into the scheduling optimization module, it triggers the parameter update process of the multi-objective optimization function. The parameter update focuses on the weight allocation of the PV accommodation rate, energy storage cycle loss, and grid frequency regulation deviation in the objective function. The correction coefficient is converted into a weight adjustment amount through normalization, and the weight adjustment direction is dynamically set according to the deviation type: if the power deviation is mainly caused by the fluctuation of PV output, the weight of the PV accommodation rate target is increased; if the deviation is due to the lag of energy storage response, the optimization weight of the energy storage cycle loss target is increased. The updated multi-objective function drives the improved particle swarm optimization algorithm to regenerate the Pareto solution set and iteratively optimize the real-time scheduling strategy.

[0091] The closed-loop control mechanism is realized through the interaction between the feedback module and the scheduling optimization module. After each parameter update, the module records the mapping relationship between the historical correction coefficient and the execution effect of the scheduling strategy to form an experience database. In the subsequent correction process, the module preferentially calls the optimized parameter combination in a similar historical scenario to accelerate convergence and improve the system's adaptability. If the deviation rate does not drop below the threshold after continuous corrections, the module triggers an abnormal warning process to notify the data fusion module to recalibrate the feature extraction model and optimize the system state representation accuracy from the data source.

[0092] Specifically, the building integrated PV energy storage and charging energy management system described in the present invention further includes: Perform a moving average filter on the deviation amount between the actual output data and the predicted value, and the length of the filter window is set according to the sampling period of the PV output time series data; Calculate the power compensation amount according to the filtered deviation amount and the rate of change of the energy storage state of charge extracted from the dynamic response vector of the data fusion module; Generate the feedback correction coefficient based on the power compensation amount.

[0093] The feedback module performs a moving average filter on the deviation amount between the actual output data and the predicted value. The length of the filter window is dynamically set according to the sampling period of the PV output time series data: when the sampling period is short, a smaller window length is used to reduce the processing delay and adapt to the high-frequency data update requirement; when the sampling period is long, the window length is increased to enhance the filtering effect and suppress the periodic noise interference. The filtering algorithm uses the exponentially weighted moving average (EWMA), which assigns a higher weight coefficient to the recent deviation data, smoothing the instantaneous fluctuations while retaining the trend deviation characteristics.

[0094] The filtered deviation and the rate of change of the energy storage state of charge (SOC) are jointly used to calculate the power compensation amount. The rate of change of SOC is extracted from the dynamic response vector of the data fusion module and generated by calculating the difference of the SOC values within adjacent time windows, reflecting the real-time charge and discharge dynamic characteristics of the energy storage system. A dynamic coupling factor is introduced in the calculation of the power compensation amount, which is generated based on the product of the rate of change of SOC and the filtered deviation. When the rate of change of SOC increases, the coupling factor increases the weight of the compensation amount calculation to strengthen the compensation effect of the energy storage regulation on the deviation; when the rate of change of SOC decreases, the weight is reduced to avoid overcompensation.

[0095] The feedback correction coefficient is generated based on the power compensation amount. The compensation amount is converted into a per-unit value in the range of 0-1 through normalization, and the per-unit value is multiplied by a preset reference correction coefficient to generate a dynamically adjusted feedback correction coefficient. The reference correction coefficient is set according to the empirical value of the deviation compensation effect in the historical operation data and is updated regularly through an offline training model. If the power compensation amount exceeds the preset threshold for multiple consecutive periods, the module triggers an adaptive adjustment mechanism to dynamically increase the reference correction coefficient based on the real-time operation scenario, enhancing the system's adjustment ability for continuous deviations.

[0096] After the correction coefficient is input to the scheduling optimization module, it triggers the dynamic relaxation or tightening of the constraint conditions in the multi-objective optimization function. For example, when the correction coefficient increases, the module relaxes the constraint boundary of the PV accommodation rate target, allowing higher energy storage charge and discharge power fluctuations to compensate for the output deviation; when the correction coefficient decreases, the constraint threshold of the grid frequency modulation deviation is tightened to give priority to maintaining the stability of the grid interaction. The correction process associates the correction coefficient with the optimization function parameters through a parameter mapping table, and the mapping table is dynamically updated according to the priority of the device operation state and the grid scheduling instruction, forming a closed-loop logic link for parameter adjustment.

[0097] The exception handling mechanism is embedded in the feedback correction process. If the filtered deviation continuously exceeds the allowable range of the device rated capacity, the module sends a recalibration request to the data fusion module, triggering the reset of the model parameters of the feature extraction unit. At the same time, an attenuation factor is introduced in the generation process of the correction coefficient, and the factor value increases with the prolongation of the exception duration, suppressing the cumulative effect of the correction amount caused by equipment failures or communication anomalies and maintaining the safety of the system operation.

[0098] Specifically, the building integrated photovoltaic energy storage charging management system described in the present invention further includes: a safety verification module for performing overload protection verification and protocol compliance verification on the control instructions distributed by the instruction generation module; When the safety verification module detects that the power value in the control instruction exceeds the device rated capacity obtained by the data acquisition module, it triggers the scheduling optimization module to regenerate a limiter scheduling strategy based on the multi-objective optimization function.

[0099] After the safety verification module receives the control instructions distributed by the instruction generation module, the overload protection verification process first obtains the rated capacity parameters of the energy storage unit, photovoltaic inverter and charging pile from the data acquisition module. The rated capacity parameters include the maximum charge and discharge power of the battery pack, the upper limit of the inverter output power and the load capacity threshold of the charging pile. The dynamic constraint boundary is generated by fusing the equipment nameplate information with the real-time status data. The verification unit compares the power allocation value in the control instruction with the dynamic constraint boundary of the corresponding equipment item by item, identifies the over-limit instruction and marks the address of the illegal equipment.

[0100] The protocol compliance check verifies the communication format and data semantics of the control instructions. The verification unit parses the OPC UA node attributes embedded in the instruction, verifies whether the node namespace is consistent with the communication interface definition of the target device, and whether the data field complies with the attribute specification of the logical device model in the IEC 61850 protocol. For instructions using PWM modulation signals, the verification unit verifies whether the signal frequency and duty cycle range are within the allowable operating range of the device driver to avoid failure of instruction execution due to protocol compatibility issues.

[0101] When it is detected that the power value exceeds the rated capacity of the equipment, the safety verification module sends an interrupt signal to the scheduling optimization module, and annotates the identifier of the over-limit equipment and the illegal power value. The interrupt signal triggers the multi-objective optimization function reconstruction process of the scheduling optimization module. The function adds a new equipment power limit constraint condition, and the constraint value is dynamically set based on the real-time status of the over-limit equipment. For example, under high temperature conditions, the upper limit of the charge and discharge power of the energy storage unit is dynamically lowered according to the temperature attenuation coefficient, generating a limit scheduling strategy that meets the safety boundary of the equipment.

[0102] In the process of generating the limit dispatching strategy, the multi-objective optimization function retains the original photovoltaic consumption rate, energy storage loss and frequency deviation targets, and adds the equipment power limit target. The limit target is achieved through the penalty function mechanism. The over-limit power value is assigned a penalty weight according to the degree of deviation, driving the improved particle swarm algorithm to preferentially generate a solution set that meets the equipment safety constraints. After the algorithm is solved, the module selects the strategy with the smallest penalty weight from the Pareto optimal solution set as the limit dispatching strategy output.

[0103] The exception handling mechanism embeds redundant design in the safety verification module. If the same device triggers overload protection multiple times in a row, the module starts the device health status assessment process, calls the battery aging parameters, inverter operation logs and charging pile fault records in the BMS system, and comprehensively calculates the device derating factor. The derating factor is updated to the multi-dimensional coupling state matrix through the data fusion module, driving the subsequent scheduling strategy to automatically adapt to the performance attenuation of the equipment to avoid repeated overloads. The verification results and processing logs are synchronized to the monitoring platform through the event bus to support the operation and maintenance personnel to trace and analyze abnormal scenarios.

[0104] The specific explanations of the technical features in the technical solution of the present invention are as follows: Multi-source heterogeneous data: refers to the photovoltaic output time-series data (such as voltage, current, power) of the photovoltaic power generation unit, the charge and discharge state data of the energy storage unit (such as state of charge, temperature, charge and discharge rate), the load demand data of the charging facility unit (such as charging power, user reservation information), and the grid dispatching instruction data (such as grid-connected power threshold, frequency modulation frequency range). Due to different source devices, these data have differences in transmission protocols, sampling frequencies, and data formats, and need to be standardized through protocol conversion.

[0105] Modbus protocol parsing: An industrial communication protocol used to parse the data frames output by the photovoltaic inverter, extract the active power, reactive power, and instantaneous voltage mapped by the register address, and verify the data integrity through CRC check. After parsing, it generates standardized fields containing timestamps and data tags, eliminating communication delays caused by protocol differences.

[0106] CAN bus protocol parsing: A communication protocol for the battery management system (BMS) of the energy storage unit. By parsing the identifier (CAN ID) in the CAN data frame, the state of charge (SOC) of the battery pack, the charge and discharge current rate, and the state of health (SOH) are extracted. The data format follows the SAE J1939 standard, and the bus arbitration mechanism is adapted during parsing to avoid data conflicts.

[0107] Sliding window interception: A technology used in the data fusion module to dynamically intercept time-series data. The window length is adaptively adjusted according to the fluctuation period of the photovoltaic output. For example, a short window (such as 30 seconds) is used in high-frequency fluctuation scenarios to capture fast-changing features, and a long window (such as 5 minutes) is used in low-frequency scenarios to analyze trend changes. The window sliding step is aligned with the data sampling interval to reduce redundant calculations.

[0108] Dynamic weighting model: Dynamically allocate weight coefficients based on the ratio of the photovoltaic output volatility (calculated by the variance of the fluctuation feature vector) to the energy storage response delay (calculated by the step response time of the charge and discharge current). When the photovoltaic volatility increases, the weight of the photovoltaic feature is increased to strengthen the compensation for output fluctuations; when the energy storage response delay increases, the weight of the energy storage feature is increased to optimize the regulation efficiency.

[0109] Pareto optimal solution set: A set of non-dominated solutions generated by the improved particle swarm algorithm in the scheduling optimization module. Each solution corresponds to a power allocation strategy and achieves a balance among the photovoltaic accommodation rate, energy storage loss, and frequency modulation deviation target. When screening the solution set, weighted scoring is combined with the priority weights of the grid dispatching instructions (such as economic coefficient, frequency modulation urgency), and the comprehensive optimal solution is selected as the real-time scheduling strategy.

[0110] OPC UA Protocol: An industrial communication protocol used to encapsulate PWM modulation signals in real-time scheduling strategies into standardized information model nodes and synchronously distribute them to energy storage BMS, photovoltaic inverters, and charging pile controllers through the publish-subscribe mode. The protocol supports the extension of Time-Sensitive Networking (TSN) and avoids communication conflicts through time slot allocation strategies, with the instruction transmission delay controlled within milliseconds.

[0111] Feedback correction coefficient: Generated by the feedback module based on the deviation between the actual output data and the predicted value (noise eliminated through moving average filtering) and the energy storage response delay. The coefficient is converted into an adjustment amount of the target weight in the optimization function through normalization. For example, when the deviation is mainly caused by photovoltaic fluctuations, the weight coefficient of the target of improving the absorption rate is increased to drive the iterative optimization of the algorithm.

[0112] Limiting scheduling strategy: A strategy reconstruction process triggered when the safety check module detects that the power instruction exceeds the rated capacity of the device. By introducing a penalty function in the multi-objective function and assigning a penalty weight to the over-limit power value, it drives the improved particle swarm algorithm to generate a power distribution plan that conforms to the device safety boundary. For example, dynamically reducing the upper limit of the energy storage charge and discharge power under high-temperature conditions.

[0113] Multi-dimensional coupling state matrix: A three-dimensional data structure output by the data fusion module. The row direction represents the characteristic parameters of photovoltaic and energy storage (such as fluctuation amplitude, response time), the column direction records the time series evolution, and the depth direction reflects the device interaction relationship. The matrix provides spatio-temporal correlated inputs for scheduling optimization and supports the accurate solution of the multi-objective model.

[0114] The above technical features eliminate data heterogeneity through the protocol, dynamically fuse and quantify the device coupling relationship, and achieve resource coordination through closed-loop optimization, ultimately solving the problems of processing delay and scheduling mismatch caused by insufficient data fusion in the existing system, and improving the operation efficiency and economy of the photovoltaic-energy storage-charging system.

[0115] Dynamic weighting model: Deployed in the data fusion module, used to quantify the dynamic coupling relationship between photovoltaic output and energy storage response. The model takes the photovoltaic output volatility (calculated by the variance of the fluctuation feature vector) and the energy storage response delay (calculated based on the step response time of the charge and discharge current) as input parameters, and dynamically allocates weight coefficients according to the ratio of volatility to delay. When the photovoltaic volatility is high, the model increases the photovoltaic feature weight to preferentially suppress output fluctuations; when the energy storage response delay is long, it increases the energy storage feature weight to strengthen the adjustment ability compensation. The output result is a weighted fusion coupled feature tensor, providing spatio-temporal correlated input data for scheduling optimization.

[0116] Improved Particle Swarm Optimization Algorithm: Applied to the scheduling optimization module for solving multi-objective optimization problems. The algorithm introduces an inertial weight dynamic adjustment mechanism in the standard particle swarm framework: the weight value changes with the volatility of photovoltaic power output. For every 5% increase in volatility, the inertial weight linearly decreases by 0.1, thereby narrowing the search range in high-volatility scenarios to accelerate convergence. The algorithm generates a Pareto optimal solution set through non-dominated sorting and combines grid priority weights (such as frequency regulation urgency and economic dispatch coefficient) to screen the comprehensive optimal solution, achieving a balance among the photovoltaic accommodation rate, energy storage loss, and frequency modulation deviation.

[0117] Sliding Window Interception Model: Used in the feature extraction stage of the data fusion module, it dynamically adjusts the window length according to the fluctuation period of photovoltaic power output time-series data. For example, a 30-second short window is used to capture fast-changing features in high-frequency fluctuations (period < 10 seconds), and a 5-minute long window is used to analyze trend changes in low-frequency scenarios (period > 60 seconds). The data within the window is used to extract fluctuation feature vectors such as fundamental wave amplitude and harmonic distortion rate through wavelet decomposition, and is time-domain aligned with dynamic response vectors such as the SOC change rate and response time of the energy storage to eliminate the time-series deviation caused by communication delay.

[0118] Feedback Correction Model: Deployed in the feedback module for generating the adjustment amount of optimization parameters. The model processes the deviation between the actual output and the predicted value through moving average filtering, filtering out instantaneous noise (such as sudden light interference) and retaining the trend deviation features. The correction coefficient is calculated in combination with the energy storage response delay (extracted from the dynamic response vector). For example, for every 200-millisecond increase in delay, the correction coefficient increases by 0.2 to enhance the compensation intensity for the lag of energy storage regulation. The correction coefficient adjusts the weight allocation in the multi-objective function through normalized mapping, driving the iterative optimization of the scheduling strategy.

[0119] Protocol Parsing Model: Includes Modbus, CAN bus, and IEC 61850 protocol parsers. The Modbus parser locates the register addresses of photovoltaic inverters and extracts features such as active power, reactive power, and voltage fluctuations; the CAN bus parser parses the energy storage BMS data frames based on the SAE J1939 standard and extracts parameters such as SOC, charge and discharge rate, and temperature; the IEC 61850 parser accesses the logical device model through MMS services and extracts grid connection power thresholds and frequency modulation command parameters. The parsed data is formatted in JSON-LD, appended with timestamps and protocol tags to form a standardized data stream.

[0120] Clipping strategy generation model: Embedded with a security verification module for equipment overload protection. When it detects that the power instruction exceeds the rated capacity of the equipment (such as the maximum charge-discharge power of energy storage), the model introduces a penalty function into the multi-objective function. For example, for every 10% excess of the over-limit power, the penalty weight increases by 0.5, forcing the improved particle swarm optimization algorithm to generate a solution set that conforms to the safety boundary. At the same time, it dynamically adjusts the clipping threshold in combination with the real-time state of the equipment (such as battery temperature). For example, for every 5°C increase in temperature, the upper limit of the energy storage power is reduced by 8% to avoid the risk of thermal runaway.

[0121] PWM modulation model: Located in the instruction generation module, it converts the power value in the scheduling strategy into a modulation signal executable by the equipment. The model normalizes the power value according to the rated capacity of the equipment. For example, the maximum energy storage power of 2MW corresponds to a duty cycle of 100%, generating a PWM waveform compatible with inverters, BMS, and charging pile controllers. The signal is encapsulated as an information model node through the OPC UA protocol, defining attributes such as amplitude, frequency, and effective timestamp, supporting multi-device synchronous subscription and real-time control.

[0122] The above models eliminate data heterogeneity through protocol unification, dynamically fuse and quantify the interactive status of devices, and achieve dynamic resource matching through closed-loop optimization, ultimately solving the data coordination problem in the photovoltaic, energy storage, and grid links, and improving the overall efficiency and economy of the system.

[0123] The present invention solves the problems of data processing delay and resource coordination mismatch in the existing system through a protocol unification and dynamic fusion mechanism for multi-source heterogeneous data. The protocol conversion module respectively parses the photovoltaic output time-series data, energy storage charge-discharge state data, and grid scheduling instructions using the Modbus, CAN bus, and IEC 61850 protocols, generating a standardized data stream in a unified data format. During the protocol parsing process, the module extracts key parameters such as photovoltaic power generation, voltage fluctuation characteristics, state of charge, and charge-discharge rate, and eliminates protocol differences and sampling period inconsistencies through timestamp synchronization and data formatting processing, providing highly consistent input for subsequent data fusion and reducing transmission and processing delays caused by data heterogeneity.

[0124] Based on the standardized data stream, the data fusion module uses a sliding window interception and dynamic weighting model to achieve spatio-temporal coupling analysis of multi-source data. The feature extraction unit dynamically intercepts the photovoltaic output fluctuation feature vector and the energy storage dynamic response vector through a sliding window. The coupling analysis unit dynamically adjusts the weight coefficient according to the ratio of the photovoltaic output volatility to the energy storage response delay, generating a multi-dimensional coupling state matrix. This matrix quantifies the correlation between the photovoltaic output fluctuation characteristics and the energy storage regulation ability, providing input data containing spatio-temporal evolution laws and device interaction characteristics for scheduling optimization, and solving the problem of resource coordination mismatch caused by insufficient multi-source data fusion.

[0125] The closed-loop control link realizes dynamic resource optimization through the cooperation of the scheduling optimization module and the feedback module. The scheduling optimization module constructs an optimization function with multiple objectives of photovoltaic accommodation rate, energy storage cycle loss, and power grid frequency modulation deviation, and uses an improved particle swarm optimization algorithm with dynamically adjusted inertia weight to generate a Pareto optimal solution set, and selects a real-time scheduling strategy in combination with the power grid priority weight. The feedback module calculates the feedback correction coefficient through the deviation between the actual output data and the predicted value, iteratively updates the parameters of the optimization function. At the same time, the safety verification module restricts the operation boundary of the equipment and triggers the amplitude limiting strategy, forming a full-process closed-loop control from data acquisition, fusion, optimization to execution, improving the operation efficiency and economy of the photovoltaic energy storage charging system.

[0126] In the specific implementation of the present invention, aiming at the problem of multi-source heterogeneous data cooperation in the photovoltaic power generation, energy storage, and grid connection links in the building photovoltaic energy storage charging scenario, the system operates efficiently through a protocol unification, dynamic fusion, and closed-loop optimization mechanism. The data acquisition module obtains the output power of photovoltaic modules, the charge and discharge current of battery packs, and the load demand of charging piles in real time through distributed sensors, and receives the grid connection threshold and frequency modulation instructions issued by the power grid scheduling system. The sampling periods of each data source are aligned through the timestamp synchronization mechanism to eliminate the time sequence misalignment between the photovoltaic output time sequence data and the energy storage state data, providing a multi-source input with consistent time for subsequent processing.

[0127] The protocol conversion module uses the Modbus protocol to parse the photovoltaic data and extracts the active power, reactive power, and voltage fluctuation amplitude output by the inverter; it uses the CAN bus protocol to parse the state of charge, charge and discharge rate, and temperature parameters of the battery for the energy storage data; and it parses the grid connection power limit and frequency modulation command frequency range based on the IEC 61850 protocol for the grid dispatching instructions. The parsed data is normalized and converted into a standardized data stream in JSON-LD format, with protocol type tags and timestamps attached, and is transmitted to the data fusion module through a priority message queue. This module uses a dynamic sliding window interception technique, and the window length is adaptively adjusted according to the fluctuation frequency of the photovoltaic output: when the fluctuation period of the photovoltaic output is less than 10 seconds, the window length is set to 30 seconds to capture the fast fluctuation characteristics; when the fluctuation period exceeds 60 seconds, the window is extended to 5 minutes to analyze the trend changes. The feature extraction unit performs wavelet decomposition on the photovoltaic data within the window, extracts the fundamental wave amplitude, harmonic distortion rate, and fluctuation frequency to construct a fluctuation feature vector; at the same time, it calculates the instantaneous change rate of the state of charge of the energy storage and the charge and discharge current response time to form a dynamic response vector. The coupling analysis unit dynamically assigns weight coefficients based on the ratio of the volatility variance to the response delay. When the volatility of the photovoltaic output exceeds 15%, the weight coefficients tilt towards the photovoltaic characteristics to strengthen the fluctuation compensation; when the response delay of the energy storage exceeds 500 milliseconds, the weight coefficients shift towards the energy storage characteristics to optimize the regulation efficiency. The weighted fusion feature tensor is unfolded into a multi-dimensional coupling state matrix in the time dimension. The row direction of the matrix represents the correlation of the characteristic parameters of the photovoltaic and the energy storage, and the column direction records the evolution law of the time series, providing input data with spatio-temporal correlation for scheduling optimization.

[0128] Based on the predicted PV output value and the remaining energy storage capacity in the multi-dimensional coupling state matrix, the scheduling optimization module constructs a multi-objective optimization function that includes PV accommodation rate, cycle loss, and frequency modulation deviation. The PV accommodation rate objective is achieved by maximizing the cosine similarity between the output curve and the local load; the energy storage loss objective is defined by minimizing the product of the charge-discharge depth and the temperature attenuation coefficient; the frequency modulation deviation objective is achieved by constraining the root mean square error between the actual output frequency and the grid command. The inertia weight of the improved particle swarm algorithm is dynamically adjusted according to the PV volatility: for every 5% increase in volatility, the inertia weight linearly decreases by 0.1, restricting the global search range to accelerate convergence. After the generated Pareto optimal solution set is weighted and scored by the grid economic dispatch coefficient and the frequency modulation emergency degree, the comprehensive optimal solution is selected as the real-time scheduling strategy. The command generation module converts the power value in the strategy into a PWM signal with adjustable duty cycle and synchronously sends it to the PV inverter, energy storage BMS, and charging pile controller through the OPC UA protocol, and the command transmission delay is controlled within 50 milliseconds to maintain control real-time. The feedback module periodically collects the deviation between the actual output and the strategy prediction value, generates a correction coefficient after moving average filtering combined with the energy storage response delay, and dynamically adjusts the objective weights in the optimization function; the safety verification module real-time verifies the matching degree between the power command and the device rated capacity, and triggers the limit strategy reconstruction process when over-limit is detected, and constrains the device operation boundary through the penalty function mechanism, forming a complete technical link from data acquisition, fusion optimization to closed-loop control, ultimately improving the resource coordination efficiency and economy of the PV-Energy Storage-Charging system.

Claims

1. An integrated energy management system for building photovoltaic energy storage and charging, characterized in that, Including: A data acquisition module for acquiring multi-source heterogeneous data, where the multi-source heterogeneous data includes photovoltaic output time-series data, energy storage charge and discharge state data, charging load demand data, and power grid dispatching instruction data; A protocol conversion module for performing protocol parsing on the multi-source heterogeneous data, parsing the photovoltaic output time-series data through the Modbus protocol, the energy storage charge and discharge state data through the CAN bus protocol, and the power grid dispatching instruction data through the IEC 61850 protocol to generate a standardized data stream; A data fusion module for performing sliding window interception on the standardized data stream, extracting the fluctuation feature vector in the photovoltaic output time-series data and the dynamic response vector in the energy storage charge and discharge state data, and inputting the fluctuation feature vector and the dynamic response vector into a dynamic weighted model constructed based on the photovoltaic output volatility and the energy storage response delay for time-domain convolution operation to generate a multi-dimensional coupled state matrix; A scheduling optimization module for constructing a multi-objective function with the photovoltaic accommodation rate, energy storage cycle loss, and power grid frequency modulation deviation as the optimization objectives according to the multi-dimensional coupled state matrix and the power grid dispatching instruction data received from the data acquisition module, and using an improved particle swarm optimization algorithm with the inertia weight dynamically adjusted according to the photovoltaic output volatility to solve and generate a real-time scheduling strategy; An instruction generation module for converting the charge and discharge power values in the real-time scheduling strategy into PWM modulation signals and distributing the PWM modulation signals to the inverters in the photovoltaic power generation unit, the BMS system in the energy storage unit, and the charging pile controllers in the charging facility unit through the OPCUA protocol; A feedback module for monitoring the actual output data of the photovoltaic power generation unit, the real-time state of charge data of the energy storage unit, and the power consumption data of the charging facility unit, calculating the deviation between the actual output data and the predicted value in the real-time scheduling strategy to generate a feedback correction coefficient, and inputting the feedback correction coefficient into the scheduling optimization module to trigger iterative optimization of the multi-objective function and adjust the real-time scheduling strategy.

2. The integrated building photovoltaic energy storage charging energy management system according to claim 1, wherein The protocol conversion module is further configured to: Perform Modbus protocol parsing on the photovoltaic output time-series data received from the data acquisition module to extract the photovoltaic power generation power and voltage fluctuation characteristics; Perform CAN bus protocol parsing on the energy storage charge and discharge state data received from the data acquisition module to extract the state of charge and charge and discharge rates; Perform IEC 61850 protocol parsing on the power grid dispatching instruction data received from the data acquisition module to extract the grid-connected power threshold and frequency modulation instruction parameters; Output the photovoltaic power generation power, voltage fluctuation characteristics, state of charge, charge and discharge rates, grid-connected power threshold, and frequency modulation instruction parameters as parsing results to the data fusion module.

3. The building integrated photovoltaic-energy storage-charging energy management system according to claim 1, characterized in that, The data fusion module includes: A feature extraction unit for performing sliding window interception on the standardized data stream received from the protocol conversion module to extract the fluctuation feature vector in the photovoltaic output time-series data and the dynamic response vector in the energy storage charge and discharge state data; A coupling analysis unit for inputting the fluctuation feature vector and the dynamic response vector into a dynamic weighting model constructed based on the volatility of photovoltaic power output and the energy storage response delay to generate the multi-dimensional coupling state matrix.

4. The integrated building photovoltaic energy storage charging energy management system according to claim 3, characterized in that The coupling analysis unit further includes: The volatility of photovoltaic power output is calculated by the variance of the fluctuation feature vector; The energy storage response delay is calculated by the time delay parameter in the dynamic response vector; The dynamic weighting model dynamically adjusts the weight coefficient according to the ratio of the volatility of photovoltaic power output to the energy storage response delay, and performs weighted superposition on the fluctuation feature vector and the dynamic response vector to generate a coupling feature tensor; The coupling feature tensor is matrix-processed to obtain the multi-dimensional coupling state matrix.

5. The integrated building photovoltaic energy storage charging energy management system according to claim 1, wherein The scheduling optimization module is further configured to: Generate a charging and discharging power distribution plan according to the predicted value of photovoltaic power output and the remaining capacity of the energy storage in the multi-dimensional coupling state matrix received from the data fusion module, wherein the predicted value of photovoltaic power output is calculated by regression analysis of the fluctuation feature vector; Combined with the grid connection power threshold in the grid scheduling instruction data received from the data acquisition module, the charging and discharging power distribution plan is constrained and optimized, and the real-time scheduling strategy is generated based on the multi-objective optimization function.

6. The integrated building photovoltaic energy storage charging energy management system according to claim 5, wherein It further includes: Based on the predicted value of photovoltaic power output in the multi-dimensional coupling state matrix received from the data fusion module and the weight coefficient output from the dynamic weighting model, a multi-objective optimization function with the maximum photovoltaic power consumption rate, the minimum energy storage cycle loss, and the minimum grid frequency modulation deviation is constructed; An improved particle swarm optimization algorithm is used to solve the multi-objective optimization function, and the inertia weight of the improved particle swarm optimization algorithm is dynamically adjusted according to the volatility of photovoltaic power output to generate a Pareto optimal solution set; According to the priority weight in the grid scheduling instruction data received from the data acquisition module, the real-time scheduling strategy is selected from the Pareto optimal solution set.

7. The integrated building photovoltaic energy storage charging energy management system according to claim 1, wherein The instruction generation module includes: An instruction encoding unit for converting the charging and discharging power value in the real-time scheduling strategy output by the scheduling optimization module into a PWM modulation signal based on the Pareto optimal solution set; An instruction distribution unit for synchronously sending the PWM modulation signal to the BMS system in the energy storage unit and the charging pile controller in the charging facility unit through the OPC UA protocol.

8. The integrated building optical storage and charging energy management system according to claim 1, characterized in that The feedback module is further configured to: Collect the actual power output data of the photovoltaic power generation unit, the real-time state of charge data of the energy storage unit, and the power consumption data of the charging facility unit; Calculate the power deviation rate according to the deviation amount between the actual power output data and the predicted value in the real-time scheduling strategy generated by the scheduling optimization module; Generate a feedback correction coefficient based on the power deviation rate and the energy storage response delay obtained from the data fusion module; Input the feedback correction coefficient into the scheduling optimization module to trigger the parameter update of the multi-objective optimization function.

9. The integrated building optical storage charging energy management system according to claim 8, wherein, It further includes: Perform moving average filtering on the deviation amount between the actual power output data and the predicted value, and the length of the filtering window is set according to the sampling period of the photovoltaic power output time series data; Calculate a power compensation amount according to the filtered deviation amount and the state of charge change rate of the energy storage extracted from the dynamic response vector of the data fusion module; Generate the feedback correction coefficient based on the power compensation amount.

10. The building integrated photovoltaic, energy storage and charging energy management system according to claim 1, characterized in that: It further includes: A safety verification module for performing overload protection verification and protocol compliance verification on the control instructions distributed by the instruction generation module; When the safety verification module detects that the power value in the control instruction exceeds the rated capacity of the device obtained by the data acquisition module, it triggers the scheduling optimization module to regenerate a clipping scheduling strategy based on a multi-objective optimization function.

Images