Integrated scheduling system for realizing PCS, EMS and BMS

Abstract

The invention discloses an integrated scheduling system for realizing a PCS, an EMS and a BMS, and relates to the technical field of power control, and the system comprises a multi-dimensional performance evaluation module which constructs a battery aging dynamic model, carries out the training, carries out the health state pre-judgment through the battery aging dynamic model based on a standardized state vector, and generates a multi-dimensional performance evaluation index; the multi-objective optimization module is used for generating a collaborative scheduling strategy set by combining a fuzzy analytic hierarchy process with a multi-objective optimization solver of an improved genetic algorithm based on the multi-dimensional performance evaluation indexes; the dynamic derating module is used for generating an executable instruction queue with security constraints by combining an industrial internet of things protocol stack with a dynamic derating coefficient algorithm based on the collaborative scheduling strategy set; according to the invention, through the physical driving characteristic layer and the dynamic parameter calibration layer, the nonlinear coupling modeling of the cyclic attenuation and calendar aging mechanism in the battery aging dynamic model is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power control, and in particular to an integrated dispatching system for realizing PCS, EMS and BMS. Background Art

[0002] With the increasing penetration of renewable energy and the large-scale deployment of energy storage systems, the coordinated scheduling of power conversion systems, energy management systems, and battery management systems has become a research hotspot in power system automation. Existing technologies primarily use EMSs to generate economically optimized scheduling strategies based on load forecasts and electricity price signals, such as implementing peak-valley arbitrage through linear programming or dynamic programming algorithms. BMSs estimate the state of charge and health (SOC) based on battery equivalent circuit or electrochemical models, and ensure battery safety through charge and discharge current limiting strategies. PCSs, as the execution unit, employ PI control or direct power control to achieve grid interaction.

[0003] In recent years, digital twin technology has been introduced to the energy storage system sector. Offline simulation models are used to predict battery aging trends or the impact of grid fluctuations. For example, a temperature-dependent lifespan model based on the Arrhenius equation is combined with a cyclic decay model based on the rainflow counting method for battery lifespan prediction. Furthermore, multi-objective optimization algorithms are used to balance economic efficiency, battery lifespan, and grid stability. However, these weighting coefficients are typically statically set, making them difficult to adapt to dynamic operating conditions. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an integrated scheduling system for realizing PCS, EMS and BMS to solve the problem of insufficient multi-objective collaborative optimization of energy storage systems.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides an integrated scheduling system for realizing PCS, EMS and BMS, which includes: a data preprocessing module, which collects equipment operation status data in real time, performs preprocessing, and generates a standardized state vector; a multidimensional performance evaluation module, which constructs and trains a battery aging dynamic model, and performs dynamic correction and digital twin mapping based on the standardized state vector through the battery aging dynamic model to generate a multidimensional performance evaluation index; a multi-objective optimization module, which generates a collaborative scheduling strategy set based on the multi-dimensional performance evaluation index through a fuzzy hierarchical analysis method combined with a multi-objective optimization solver of an improved genetic algorithm; a dynamic derating module, which generates an executable instruction queue with safety constraints based on the collaborative scheduling strategy set through an industrial Internet of Things protocol stack combined with a dynamic derating coefficient algorithm; a fuse monitoring module, which generates an execution feature data set with abnormal coding based on the executable instruction queue through a real-time fuse mechanism and a power quality monitoring algorithm; a closed-loop optimization module, which updates the battery aging model parameters through an incremental online learning algorithm based on the execution feature data set, and generates a power grid fluctuation feature library through an Arrhenius equation parameter optimizer.

[0008] As a preferred solution for realizing the integrated scheduling system of PCS, EMS and BMS described in the present invention, the equipment operation status data includes PCS power and voltage, EMS load forecast data and BMS charge state and health status parameters.

[0009] As a preferred solution for realizing the integrated scheduling system of PCS, EMS and BMS described in the present invention, the preprocessing includes denoising, normalization, time alignment of multi-source heterogeneous data and feature fusion.

[0010] As a preferred solution for realizing the integrated scheduling system of PCS, EMS and BMS according to the present invention, the specific steps of constructing the battery aging dynamic model and training it are as follows:

[0011] The input layer is constructed based on the multi-source heterogeneous data fusion method to convert the standardized state vector into a time series tensor;

[0012] The Arrhenius multi-mechanism coupling analysis method is used to extract the joint characteristics of cyclic decay and calendar aging from the input data, and a shared physical driving feature layer is constructed.

[0013] By improving the particle swarm optimization algorithm, nonlinear parameter fitting is performed on shared features to construct a parameter calibration layer;

[0014] The calibrated cyclic attenuation factor is fused with real-time data through the extended Kalman filter algorithm to construct a dynamic state estimation layer;

[0015] A battery aging dynamic model is constructed based on the input layer, physical driving feature layer, parameter calibration layer and dynamic state estimation layer and trained using a hybrid collaborative training method.

[0016] As a preferred solution for realizing the integrated scheduling system of PCS, EMS and BMS described in the present invention, wherein: based on the standardized state vector, dynamic correction and digital twin mapping are performed through the battery aging dynamic model to generate multi-dimensional performance evaluation indicators, the specific steps are as follows:

[0017] Based on the standardized state vector, the health status is predicted through the battery aging dynamic model to generate the health status indicator;

[0018] Based on the health status indicators, dynamic state correction is performed through the extended Kalman filter algorithm to generate real-time health status estimation values;

[0019] Based on the real-time health status estimation value, multi-dimensional performance quantification is performed through the digital twin mapping engine to generate multi-dimensional performance evaluation indicators.

[0020] As a preferred solution for realizing the integrated scheduling system of PCS, EMS and BMS described in the present invention, wherein: based on the multi-dimensional performance evaluation index, the collaborative scheduling strategy set is generated by combining the fuzzy analytic hierarchy process with the multi-objective optimization solver of the improved genetic algorithm. The specific steps are as follows:

[0021] Based on multi-dimensional performance evaluation indicators, multi-objective weight allocation is performed through fuzzy analytic hierarchy process to generate priority weight vectors;

[0022] Based on the priority weight vector, the multi-objective optimization solution is solved by using the improved non-dominated sorting genetic algorithm to generate the Pareto optimal solution set;

[0023] Based on the Pareto optimal solution set, the TOPSIS decision method is used to perform strategy optimization and generate the final collaborative scheduling strategy set.

[0024] As a preferred solution for realizing the integrated scheduling system of PCS, EMS and BMS described in the present invention, wherein: based on the collaborative scheduling strategy set, the industrial Internet of Things protocol stack is combined with the dynamic derating coefficient algorithm to generate an executable instruction queue with security constraints. The specific steps are as follows:

[0025] Based on the collaborative scheduling strategy set, the command is securely encapsulated through the industrial Internet of Things protocol stack to generate protocol-compatible command frames;

[0026] Based on the protocol-compatible instruction frame, security constraints are injected through the dynamic derating coefficient algorithm to generate risk response instructions;

[0027] Based on the risk response instructions, an executable instruction queue with security constraints is generated through an instruction priority sorting algorithm.

[0028] As a preferred solution for realizing the integrated dispatching system of PCS, EMS and BMS described in the present invention, wherein: based on the executable instruction queue, the execution feature data set with abnormal coding is generated through the real-time fuse mechanism and the power quality monitoring algorithm. The specific steps are as follows:

[0029] Based on the executable instruction queue, overload risk monitoring is performed through a real-time circuit breaker mechanism, generating circuit breaker events and exception tags;

[0030] Based on the execution process data of the executable instruction queue, dynamic parameter analysis is performed through the power quality monitoring algorithm to generate quality abnormality waveforms;

[0031] Based on the circuit-breaking events and quality abnormal waveforms, feature fusion is performed through the abnormal coding rule library to generate an execution feature dataset with security labels.

[0032] As a preferred solution for realizing the integrated scheduling system of PCS, EMS and BMS according to the present invention, the battery aging model parameters are updated by an incremental online learning algorithm based on the execution feature data set. The specific steps are as follows:

[0033] Based on the execution feature dataset, dynamic parameter updates are performed through an online sequential extreme learning machine to generate preliminary calibration parameters.

[0034] Based on the preliminary calibration parameters, physical range correction is performed through a regularized constrained optimizer to generate updated battery aging model parameters.

[0035] As a preferred solution for realizing the integrated dispatching system of PCS, EMS and BMS described in the present invention, wherein: the power grid fluctuation characteristic library is generated by the Arrhenius equation parameter optimizer, and the specific steps are as follows:

[0036] Based on the execution feature dataset, the temperature-dependent aging features are generated by wavelet transform-Arrhenius temperature feature extractor;

[0037] Based on the temperature-dependent aging characteristics, a spectrum clustering analyzer is used to perform multi-dimensional feature clustering to generate a power grid fluctuation feature library.

[0038] The beneficial effects of the present invention are: through the physical driving feature layer and the dynamic parameter calibration layer, nonlinear coupling modeling of the cycle attenuation and calendar aging mechanisms in the battery aging dynamic model is realized, thereby improving the health status prediction accuracy and model adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 Schematic diagram of an integrated scheduling system for implementing PCS, EMS, and BMS.

[0041] Figure 2 The figure is a flow chart for realizing the integrated scheduling of PCS, EMS and BMS.

[0042] Figure 3 Flowchart constructed for the dynamic model of battery aging.

[0043] Figure 4 Flowchart of dual-channel processing for the closed-loop optimization module. DETAILED DESCRIPTION

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0045] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0047] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides an integrated scheduling system for implementing PCS, EMS and BMS, including the following steps:

[0048] Data preprocessing module collects equipment operation status data in real time, performs preprocessing, and generates a standardized state vector;

[0049] Equipment operating status data includes PCS power and voltage, EMS load forecast data, and BMS state of charge and health parameters;

[0050] Preprocessing includes denoising, normalization, time alignment of multi-source heterogeneous data and feature fusion.

[0051] Furthermore, during the denoising process, a wavelet threshold denoising method is used to address high-frequency noise interference contained in PCS power parameters, EMS load forecast data, and BMS state of charge and health parameters. Specifically, this method performs a multi-scale decomposition of transient power fluctuations in PCS power parameters based on Daubechies wavelet basis functions, filtering out abnormal pulse components in high-frequency detail coefficients using a soft threshold function; suppresses non-physical fluctuations in low-frequency approximation coefficients of random drift noise in BMS state of charge and health parameters using a fixed threshold rule; and eliminates periodic background noise in EMS load forecast data using translation-invariant wavelet denoising, retaining the main components of the load trend.

[0052] Furthermore, during the normalization process, the Z-score normalization method is used to eliminate data distribution offsets based on the dimensional differences of the PCS power parameters, the EMS load forecast data, and the BMS state of charge and health status parameters. Specifically, the PCS power parameter series is statistically averaged based on the mean and standard deviation of the sliding window, converting the instantaneous power value into a zero-mean unit variance distribution; the electricity price time series in the EMS load forecast data is linearly scaled using the global mean and standard deviation to ensure the comparability of electricity price weights at different time scales; the BMS state of charge parameters are dynamically normalized based on the range of the charge and discharge cycle, and the health status parameters are offset corrected by fitting the baseline value based on the historical capacity decay trajectory.

[0053] Furthermore, in the process of time alignment of multi-source heterogeneous data, a timestamp alignment method based on a precise clock synchronization protocol is used to achieve microsecond-level synchronization for the acquisition frequency and timestamp differences of the PCS power parameters, the EMS load forecast data, and the BMS state of charge and health status parameters. Specifically, it includes: linear interpolation resampling of the PCS power parameters to generate a power parameter sequence that is timestamp-aligned with the BMS state of charge parameters; the EMS load forecast data is aligned to the PCS power parameter time axis through a sliding window mean aggregation method; and the BMS health status parameters use a forward filling method to fill the timestamp gap. After time alignment, the PCS power parameters, EMS load forecast data, and BMS state of charge and health status parameters form a time-synchronized multi-source data stream, providing time-consistent input for feature fusion.

[0054] Furthermore, during the feature fusion process, principal component analysis is used to achieve cross-domain feature association and dimensionality reduction for the time-aligned PCS power parameters, EMS load forecast data, and BMS charge and health status parameters. Specifically, the power fluctuation rate, charge and discharge efficiency, and voltage deviation characteristics are extracted from the PCS power parameters; the electricity price sensitivity coefficient and load change gradient characteristics are extracted from the EMS load forecast data; and the capacity attenuation slope and internal resistance temperature correlation coefficient characteristics are extracted from the BMS charge and health status parameters. The covariance matrix of the above features is calculated and orthogonal transformed through principal component analysis, and the principal component vector with the highest cumulative contribution rate is retained to generate a three-dimensional standardized state vector that integrates the PCS dynamic response characteristics, EMS economic preferences, and BMS health status characteristics.

[0055] The multi-dimensional performance evaluation module builds and trains a dynamic battery aging model. Based on the standardized state vector, it dynamically corrects the battery aging dynamic model and maps it to a digital twin to generate multi-dimensional performance evaluation indicators.

[0056] The input layer is constructed based on the multi-source heterogeneous data fusion method to convert the standardized state vector into a time series tensor;

[0057] It should be noted that in the process of constructing the input layer based on the multi-source heterogeneous data fusion method, the process of converting the standardized state vector into a time series tensor is as follows: the time-aligned PCS power parameters, EMS load forecast data, and BMS charge state and health state parameters are sampled in a sliding window to extract the standardized state vector within the continuous time window; the sampling results are organized through a three-dimensional tensor structure, where the first dimension is the time step (representing the historical time series dependency), the second dimension is the number of features (corresponding to the principal component vectors of PCS dynamic response characteristics, EMS economic preferences, and BMS health state characteristics), and the third dimension is the number of devices (distinguishing independent data sources for PCS, EMS, and BMS). The resulting time series-feature-device three-dimensional tensor serves as the input layer of the battery aging dynamic model, preserving the time evolution characteristics and cross-device correlation of multi-source data.

[0058] The Arrhenius multi-mechanism coupling analysis method is used to extract the joint characteristics of cyclic decay and calendar aging from the input data, and a shared physical driving feature layer is constructed.

[0059] It should be noted that cross-mechanism feature extraction is performed on the PCS power parameters, EMS load forecast data, and BMS state-of-charge and health parameters contained in the time series tensor. Specifically, this involves calculating the calendar aging rate of the BMS state-of-charge parameters at different temperature gradients based on the temperature dependence of the Arrhenius equation, generating a temperature-time-dependent aging feature vector. The depth and number of charge and discharge cycles in the PCS power parameters are counted using the rainflow counting method to extract the capacity decay feature vector associated with cyclic stress. The temperature-time-dependent aging feature vector and the cyclic stress capacity decay feature vector are jointly modeled using a physically driven coupling function (formally, the product of calendar aging rate and cyclic decay factor) to generate a shared physically driven feature layer that integrates temperature, cyclic stress, and time evolution. This feature layer preserves the coupled influence of the PCS dynamic power behavior on the BMS health state, providing cross-domain-dependent input features for the parameter calibration layer.

[0060] By improving the particle swarm optimization algorithm, nonlinear parameter fitting is performed on shared features to construct a parameter calibration layer;

[0061] It should be noted that during the nonlinear parameter fitting of the temperature-time correlation aging eigenvector and the cyclic stress capacity decay eigenvector in the shared physical driving feature layer using an improved particle swarm optimization algorithm, a dynamic inertia weight adjustment strategy is used to optimize the particle search process. Specifically, the following steps are performed: initializing the particle swarm position and velocity, where the particle position vector corresponds to the combination of the cyclic decay factor and calendar aging coefficient parameters in the battery aging dynamic model; constructing a fitness function based on the temperature gradient, cycle depth, and capacity decay observation data in the shared physical driving feature layer, and calculating the fitting error of each individual in the particle swarm; during the iteration process, dynamically adjusting the inertia weight based on the particle's historical optimal position and the global optimal position to balance local search and global exploration capabilities; and imposing constraint penalties on out-of-bounds cyclic decay factor and calendar aging coefficient parameters to ensure the physical meaning of the parameters is reasonable. By improving the optimal solution set of cyclic decay factor and calendar aging coefficient output by the particle swarm optimization algorithm, a parameter calibration layer is constructed to map the multi-mechanism coupling characteristics in the shared physical driving feature layer to the physical parameter space of the battery aging dynamic model, providing calibrated cyclic decay factor and calendar aging coefficient inputs for the dynamic state estimation layer.

[0062] The calibrated cyclic attenuation factor is fused with real-time data through the extended Kalman filter algorithm to construct a dynamic state estimation layer;

[0063] It should be noted that in constructing the dynamic state estimation layer by fusing the calibrated cyclic decay factor with real-time data via the extended Kalman filter algorithm, a state vector is defined containing the capacity retention rate (BMS health state parameter), the calibrated cyclic decay factor, and the charge-discharge efficiency deviation (PCS power parameter). A state equation is established to describe the dynamic relationship between the capacity retention rate and cycle number and temperature gradient, with the cyclic decay factor acting as a time-varying parameter in the state transition matrix calculation. The observation equation generates a capacity decay observation sequence based on the measured AC output power of the PCS and the real-time state-of-charge parameters of the BMS. Through the iterative calculation of the prediction and update steps of the extended Kalman filter, the estimated capacity retention rate and its covariance matrix in the state vector are dynamically corrected by combining the calibrated cyclic decay factor with the real-time power fluctuations of the PCS and the BMS temperature parameters. The final output of the dynamic state estimation layer contains the real-time updated BMS health state parameter estimates, reflecting the coupled impact of PCS operating mode switching and EMS scheduling strategies on the battery aging process.

[0064] A battery aging dynamic model is constructed based on the input layer, physical driving feature layer, parameter calibration layer and dynamic state estimation layer and trained using a hybrid collaborative training method.

[0065] It should be noted that based on the number of PCS charge and discharge cycles, BMS temperature gradient and duration, and EMS load forecast data in the input layer, a standardized state vector is generated and converted into a time series tensor. The physical driving feature layer extracts the cycle attenuation characteristics and calendar aging characteristics through a multi-mechanism coupling analysis method. The cycle attenuation characteristic is based on the rain flow counting method to count the equivalent cycle number and combines the joint influence function of discharge depth and temperature to quantify the attenuation rate. The calendar aging characteristic calculates the aging rate based on temperature data. The parameter calibration layer uses an improved particle swarm optimization algorithm to perform offline global parameter identification of the cycle attenuation factor and calendar aging coefficient. The constraints include the range of activation energy values ​​and the physical upper limit of the cycle attenuation factor. The dynamic state estimation layer uses the extended Kalman filter algorithm to fuse the calibrated cycle attenuation factor with the real-time collected capacity deviation and temperature data to iteratively update the health state estimate. The hybrid collaborative training method uses the improved particle swarm optimization algorithm to optimize the physical driving parameters offline. The extended Kalman filter algorithm then online fuses the PCS real-time charge and discharge data, BMS temperature monitoring values, and EMS load duration information to achieve continuous adaptive calibration of model parameters and dynamic tracking of health status.

[0066] Based on the standardized state vector, the health status is predicted through the battery aging dynamic model to generate the health status indicator;

[0067]

[0068] Where H is the normalized state vector, α is the cycle attenuation factor, ΔQ is the single cycle capacity throughput, D is the depth of discharge, T is the battery operating temperature, β is the calendar aging coefficient, E is the activation energy, R is the gas constant, t is the number of cycles, f(D,T) is a function that represents the joint influence of the depth of discharge D and temperature T on the battery cycle aging rate to correct the actual capacity loss degree of each cycle, K(ΔQ,T) is an error correction term that represents the dynamic calibration of the battery state of health (SOH) model based on the observation value of the capacity change ΔQ and temperature T based on the extended Kalman filter algorithm, and S(t) is the health status indicator of the battery at time t;

[0069] It should be noted that the cyclic attenuation characteristics and calendar aging characteristics are modeled as the cyclic attenuation term α∑ΔQ and the calendar aging term respectively through the physical driving feature layer. The discharge depth D and temperature T are obtained by rain flow counting method and time series sampling. The joint influence function f(D,T) quantifies the nonlinear superposition effect of different discharge depths and temperatures on cyclic attenuation. The Arrhenius equation term Reflects the accelerating effect of temperature on calendar aging; the parameter calibration layer dynamically updates the calibrated cycle attenuation factor α and calendar aging coefficient β based on the execution feature data set through the online sequence extreme learning machine, and uses the extended Kalman filter algorithm to generate the error correction function K(ΔQ,T) according to the observed residual of the real-time capacity deviation ΔQ and the temperature T to compensate for the unmodeled dynamics of the model; the dynamic state estimation layer takes the initial health state H as the benchmark, and uses the extended Kalman filter algorithm to fuse the cycle attenuation term, calendar aging term and error correction term to iteratively calculate the real-time health state estimate S(t). The calculation process is as follows: the number of PCS charge and discharge cycles ΔQ is weighted and accumulated according to the joint influence function f(D,T) of the discharge depth D and the temperature T, and then multiplied by the calibrated cycle attenuation factor α to generate the cycle attenuation; the calendar aging is calculated by combining the BMS temperature parameter T, activation energy E and EMS running time t After superimposing the error term K(ΔQ,T) dynamically corrected by the extended Kalman filter, the cycle attenuation and calendar aging are deducted from the initial health state H, and the real-time health state indicator S(t) is output, realizing multi-source data-driven dynamic modeling of battery aging and accurate state estimation.

[0070] Based on the health status indicators, dynamic state correction is performed through the extended Kalman filter algorithm to generate real-time health status estimation values;

[0071] It should be noted that during the dynamic state correction process using the extended Kalman filter algorithm based on health status indicators (including capacity retention offset and calendar aging loss), the state vector is defined as the estimated capacity retention rate, the calibrated cycle decay factor, and the calendar aging coefficient of the battery health status. A state transition equation is established to describe the dynamic evolution of capacity retention rate with the number of PCS charge and discharge cycles and BMS temperature parameters. The calibrated cycle decay factor is used as a time-varying parameter in the state transition matrix calculation. The observation equation generates a capacity decay observation sequence based on the real-time measurements of the BMS state of charge parameters and the AC output power of the PCS. Through the prediction step (updating the prior estimate and covariance matrix based on the state transition equation) and the update step (calculating the Kalman gain and correcting the posterior estimate based on the observation equation) of the extended Kalman filter, the real-time PCS charge and discharge power fluctuations, BMS temperature gradient changes, and EMS accumulated operating time data are iteratively integrated to dynamically adjust the confidence interval of the capacity retention rate estimate and generate a real-time health status estimate. The real-time health status estimate reflects the synergistic effect of the PCS charge and discharge strategy and the BMS temperature control, providing dynamic correction input for the digital twin mapping engine.

[0072] Based on the real-time health status estimation value, multi-dimensional performance quantification is performed through the digital twin mapping engine to generate multi-dimensional performance evaluation indicators.

[0073] It should be noted that, based on the real-time health state estimate, the PCS's real-time charge and discharge power data and the EMS's time-of-use electricity price weight coefficient are first input into the economic quantification model. The economic benefit within the current dispatch window is calculated by multiplying the charge and discharge power and the electricity price to generate an economic score. Subsequently, the capacity retention rate and BMS temperature parameters from the real-time health state estimate are input into the life loss model. The temperature-dependent aging rate is calculated based on the Arrhenius equation. The life decay rate index is generated by combining the number of charge and discharge cycles calculated using the rain flow counting method and the calibrated cycle decay factor. Simultaneously, the PCS's AC voltage RMS deviation, the EMS's load tracking error rate, and the BMS's state-of-charge fluctuation gradient are input into the stability assessment model. A grid stability index is generated by weighted fusion of the RMS voltage deviation, the cumulative absolute value of the load error, and the slope of the state-of-charge fluctuation. The resulting multi-dimensional performance evaluation indicators (economic score, life decay rate, and grid stability index) fully characterize the energy storage system's comprehensive performance in terms of revenue, battery life, and grid interaction, providing a quantitative decision-making basis for multi-objective collaborative optimization.

[0074] The multi-objective optimization module generates a collaborative scheduling strategy set based on multi-dimensional performance evaluation indicators through a multi-objective optimization solver combining fuzzy analytic hierarchy process with an improved genetic algorithm;

[0075] Based on multi-dimensional performance evaluation indicators, multi-objective weight allocation is performed through fuzzy analytic hierarchy process to generate priority weight vectors;

[0076] It should be noted that the target weight allocation logic framework is constructed through the fuzzy hierarchical analysis method, and its criterion layer is composed of the economic dimension, life loss dimension and grid stability dimension. The data-driven fuzzy rule base is dynamically triggered by the real-time electricity price range, BMS temperature parameters and PCS voltage deviation in the EMS load forecast data: when the real-time electricity price is higher than the historical average, the fuzzy importance level of the economic score compared to the life decay rate is set to "strong correlation"; when the BMS temperature parameter exceeds the preset safety threshold, the fuzzy importance level of the life decay rate compared to the grid stability index is set to "extremely strong correlation"; when the PCS AC voltage effective value deviation exceeds the allowable range of the grid, the fuzzy importance level of the grid stability index compared to the economic score is set to "medium correlation". The above fuzzy rules are mapped into triangular fuzzy numbers to construct a criterion-level fuzzy judgment matrix. The fuzzy weight intervals of each dimension are calculated by the geometric mean method, and the defuzzification processing (center of gravity method) is used to convert the fuzzy weight intervals into deterministic scalar weight values. The priority weight vectors (economic weight, life weight, stability weight) are generated, which provide a dynamic quantitative decision-making basis for the multi-objective optimization of the improved non-dominated sorting genetic algorithm and ensure that the weight distribution strictly matches the real-time operating status.

[0077] Based on the priority weight vector, the multi-objective optimization solution is solved by using the improved non-dominated sorting genetic algorithm to generate the Pareto optimal solution set;

[0078] It should be noted that the improved non-dominated sorting genetic algorithm achieves performance improvement based on the traditional non-dominated sorting genetic algorithm through dynamic crossover and mutation probability adjustment strategies and an elite retention strategy based on crowding. The dynamic crossover probability is adaptively adjusted according to the genetic diversity of the population (based on the difference in gene sequence). When the similarity of individuals in the population is too high, the crossover probability is automatically increased to enhance the global search capability; the mutation probability is reversely adjusted according to the individual fitness ranking, and individuals with lower fitness are given a higher mutation probability to avoid local convergence. After non-dominated sorting, the elite retention strategy based on crowding prioritizes individuals in sparsely distributed areas in the target space. By calculating the neighborhood density of individuals in the target space, solutions in high-crowding areas are retained to improve the distribution uniformity of the Pareto front. The above-mentioned improved strategy enhances the algorithm's ability to collaboratively solve multiple objectives such as PCS operating mode, EMS scheduling requirements, and BMS health status constraints through dynamic parameter adjustment and solution set distribution optimization.

[0079] Furthermore, based on the priority weight vector, the process of performing multi-objective optimization solution through an improved non-dominated sorting genetic algorithm is integrated as follows: First, in the population initialization stage, the PCS charging and discharging power setpoint value sequence, the EMS scheduling period division strategy, and the BMS state of charge operating range constraints are encoded as individual genes, covering all feasible scheduling schemes. Subsequently, the multi-dimensional performance evaluation indicators are weighted and summed using the priority weight vector to generate a fitness function, which quantifies the comprehensive performance of the individuals in multi-objective collaborative optimization. In this process, the individuals in the population are divided into Pareto levels through non-dominated sorting, and highly distributed individuals are selected as parents by combining the crowding distance calculation. Multi-point crossover and boundary mutation operations are performed on the parent gene sequence based on the dynamically adjusted crossover and mutation probabilities to generate the offspring population. After merging the parent and offspring populations, the next generation of individuals are selected based on the crowding elite retention strategy to ensure that the solution set is evenly distributed in the target space and close to the Pareto frontier. Finally, a set of non-inferior solutions that meet the PCS power constraints, EMS economic requirements, and BMS health status protection are output, providing multi-dimensional optimization solution input for the TOPSIS decision method.

[0080] Based on the Pareto optimal solution set, the TOPSIS decision method is used to perform strategy optimization and generate the final collaborative scheduling strategy set.

[0081] It should be noted that when the TOPSIS decision method is used to optimize the strategy based on the Pareto optimal solution set, a decision matrix is ​​first constructed to perform individual solutions in the corresponding Pareto solution set, and corresponding multi-dimensional performance evaluation indicators are listed, such as economic score, life decay rate, and power grid stability index; the decision matrix is ​​normalized to eliminate dimensional differences, and the normalized matrix is ​​weighted based on the priority weight vector to generate a weighted decision matrix; the positive ideal solution of each solution is calculated, such as: the maximum economic score, the minimum life decay rate, and the maximum power grid stability index, and the negative ideal solution is calculated, such as: the minimum economic score, the maximum life decay rate, and the minimum power grid stability index; the distance of each solution to the positive and negative ideal solutions is calculated using the Euclidean distance formula to generate a relative closeness score; the closeness score is sorted in descending order, and the solution with the highest score is selected as the final collaborative scheduling strategy set to ensure the optimal balance between PCS power output, EMS scheduling plan and BMS health status protection.

[0082] The dynamic derating module, based on the collaborative scheduling strategy set, generates an executable instruction queue with security constraints through the industrial Internet of Things protocol stack combined with the dynamic derating coefficient algorithm;

[0083] Based on the collaborative scheduling strategy set, the command is securely encapsulated through the industrial Internet of Things protocol stack to generate protocol-compatible command frames;

[0084] It should be noted that when securely encapsulating instructions through the Industrial Internet of Things protocol stack based on the coordinated scheduling strategy set's scheduling period division strategy and the BMS state of charge, the PCS charge and discharge power setting value sequence in the coordinated scheduling strategy set is first parsed into a power instruction code, the EMS scheduling period division strategy is parsed into a timestamp control code, and the BMS state of charge operating range constraint is parsed into a threshold check code. The power instruction code, timestamp control code, and threshold check code are converted into a protocol data unit (PDU) using the message encapsulation rules of the Industrial Internet of Things protocol stack. Security fields are added to the PDU, such as an encrypted message header, hash check code, and device identity identifier based on the TLS protocol, to generate a protocol-compatible instruction frame that complies with Industrial Internet of Things security standards. The protocol-compatible instruction frame is adapted through the physical layer and data link layer interfaces of the Industrial Internet of Things protocol stack to ensure that PCS, EMS, and BMS instructions can be securely parsed and executed by the target device.

[0085] Based on the protocol-compatible instruction frame, security constraints are injected through the dynamic derating coefficient algorithm to generate risk response instructions;

[0086] It should be noted that when injecting safety constraints using a dynamic derating factor algorithm based on a protocol-compatible instruction frame, the PCS charge and discharge power instruction code in the protocol-compatible instruction frame is first parsed to obtain the initial power setting value. The BMS temperature parameters, the PCS DC-side current RMS value, and the EMS load forecast error rate are monitored in real time. Based on the overload risk level, such as temperature exceeding the limit, current exceeding the limit, or load deviation exceeding the threshold, the derating factor is dynamically calculated. The derating factor is multiplied by the initial power setting value to generate a derating power instruction code. Simultaneously, a logical AND operation is performed on the derating factor and the BMS state of charge threshold check code to generate a derating constraint check code. The derating power instruction code, the EMS scheduling period division strategy, and the derating constraint check code are integrated and repackaged through the security field (hash check code and device identity identifier) ​​of the Industrial Internet of Things protocol stack to generate a risk response instruction. The risk response instruction ensures that the PCS charge and discharge power dynamically adapts to the BMS health status and real-time operational risks, achieving coordinated scheduling under safety constraints.

[0087] Based on the risk response instructions, an executable instruction queue with security constraints is generated through an instruction priority sorting algorithm.

[0088] The specific process for generating a security-constrained executable instruction queue based on risk response instructions using an instruction priority sorting algorithm is as follows: The PCS derated power setting value sequence, EMS scheduling time period division strategy, and BMS derated constraint check code within the risk response instructions are parsed. The security level identifier (based on the match between the hash check code integrity verification result and the device identity identifier), timestamp control code urgency, and constraint check code compliance of each instruction are extracted. Using multi-level priority determination rules, PCS charge and discharge power instructions are assigned a basic priority based on their security level identifier (high, medium, or low risk). EMS scheduling instructions are prioritized based on the time period urgency of the timestamp control code. BMS constraint check codes are dynamically adjusted based on the degree to which the state of charge (SOC) deviates from the threshold. A comprehensive priority score is generated by integrating the security level, urgency, and offset. The instruction sequence is sorted in descending order by score, forming a security-constrained executable instruction queue. The final queue is re-encapsulated using the security fields (hash check code and device identity identifier) ​​of the Industrial Internet of Things protocol stack, ensuring the secure implementation of PCS charge and discharge operations, EMS scheduling execution, and BMS health status protection in prioritized order.

[0089] The fuse monitoring module, based on the executable instruction queue, generates an execution feature data set with abnormal coding through real-time fuse mechanism and power quality monitoring algorithm;

[0090] Based on the executable instruction queue, overload risk monitoring is performed through a real-time circuit breaker mechanism, generating circuit breaker events and exception tags;

[0091] It should be noted that when overload risk monitoring is performed through a real-time fusing mechanism based on an executable instruction queue with safety constraints, the effective value of the PCS DC side current, the BMS temperature parameters, and the EMS load forecast error rate are collected in real time and dynamically compared with the safety constraint thresholds in the executable instruction queue (PCS charge and discharge power upper limit, BMS temperature protection threshold, and EMS load deviation allowable range). When the effective value of the PCS DC side current exceeds the charge and discharge power upper limit, it is marked as an overcurrent fusing event and an overcurrent anomaly tag is generated. When the BMS temperature parameter exceeds the temperature protection threshold, it is marked as an overtemperature fusing event and an overtemperature anomaly tag is generated. When the EMS load forecast error rate exceeds the allowable range, it is marked as a load deviation fusing event and a load anomaly tag is generated. The fusing event and the anomaly tag are associated with the timestamp, device identifier, and the type of the exceeded parameter to form a structured anomaly record, which provides risk monitoring result input for the execution feature dataset.

[0092] Based on the execution process data of the executable instruction queue, dynamic parameter analysis is performed through the power quality monitoring algorithm to generate quality abnormality waveforms;

[0093] Based on the execution process data of the executable instruction queue, when performing dynamic parameter analysis through the power quality monitoring algorithm, the PCS AC voltage effective value deviation, voltage harmonic distortion rate and frequency offset are first extracted. Combined with the active / reactive power fluctuation gradient in the EMS load tracking response data, the high-frequency noise and transient disturbance components of the voltage / current waveform are decomposed through the wavelet transform method; a time-frequency joint analysis is performed on the transient fluctuation of the BMS state of charge to detect the charge state jump characteristics caused by abnormal charging and discharging; the waveform segment with the voltage harmonic distortion rate exceeding the preset harmonic distortion rate threshold is marked as a harmonic distortion anomaly, the waveform segment with the voltage effective value deviation exceeding the preset voltage deviation threshold is marked as a voltage sag anomaly, and the waveform segment with the frequency offset exceeding the preset frequency offset threshold is marked as a frequency fluctuation anomaly; based on the wavelet coefficient energy distribution and transient feature matching rules, comprehensive quality abnormal waveform data including abnormality type, timestamp and waveform segment is generated to provide abnormal feature input of the power quality dimension for the execution feature dataset.

[0094] Based on the circuit-breaking events and quality abnormal waveforms, feature fusion is performed through the abnormal coding rule library to generate an execution feature dataset with security labels.

[0095] It should be noted that, based on the fuse event and the quality abnormal waveform, when the feature fusion is performed through the abnormal coding rule library, the type of the exceeding parameter of the fuse event, such as overcurrent, overtemperature, and load deviation, is first mapped to the safety risk level code, such as overcurrent is coded as "OL", overtemperature is coded as "OT", and load deviation is coded as "LD". At the same time, the abnormal type of the quality abnormal waveform, such as harmonic distortion, voltage sag, and frequency fluctuation, is mapped to the quality defect level code, such as harmonic coding as "HD", voltage sag coding as "VS", and frequency fluctuation coding as "FF". The fuse event and the quality abnormal waveform are aligned based on the timestamp. During the period of occurrence of normal waveforms, the PCS AC voltage / current waveforms, EMS load tracking response data, and synchronous operating status data of BMS temperature parameters are extracted; through the association rules of the abnormal coding rule base, such as "OL+HD" mapping to "high safety risk" and "OT+VS" mapping to "medium safety risk", the safety risk level code and the quality defect level code are combined to generate a composite safety label; the timestamp, device identity identifier, composite safety label and synchronous operating status data are integrated to generate an execution feature data set with safety labels, providing multi-dimensional abnormal feature input for the battery aging dynamic model.

[0096] The closed-loop optimization module updates the battery aging model parameters through an incremental online learning algorithm based on the execution feature dataset, and generates a power grid fluctuation feature library through the Arrhenius equation parameter optimizer.

[0097] Based on the execution feature dataset, dynamic parameter updates are performed through an online sequential extreme learning machine to generate preliminary calibration parameters.

[0098] It should be noted that when dynamic parameter updates are performed using an online sequence extreme learning machine based on the health status parameters of the execution feature dataset, the PCS AC voltage RMS deviation, EMS load tracking error rate, and BMS temperature parameters in the execution feature dataset are first used as input feature vectors, and the safety label is used as a supervisory signal. The weight matrix of the hidden layer nodes of the online sequence extreme learning machine is initialized, and the time series samples of the execution feature dataset are read batch by batch based on the incremental learning model. The connection weights from the hidden layer to the output layer are updated using the recursive least squares method. In response to the dynamic offset of the BMS health status parameters, the PCS power correction coefficient and the EMS scheduling compensation factor are adjusted through the error feedback mechanism to generate preliminary calibration parameters (including PCS charge and discharge efficiency compensation, EMS load forecast error correction weight, and BMS health status estimation offset). After the preliminary calibration parameters are verified for data integrity using a hash checksum, they are input into the battery aging dynamic model to complete the parameter iterative update.

[0099] Based on the preliminary calibration parameters, a regularized constrained optimizer is used to perform physical range correction to generate updated battery aging model parameters.

[0100] It should be noted that, based on the preliminary calibration parameters, when performing physical range correction through the regularization constrained optimizer, the physically feasible interval constraints of the battery aging dynamic model parameters are first defined: the PCS charging and discharging efficiency compensation amount is limited to the preset percentage fluctuation range of the rated efficiency, the EMS load forecast error correction weight is limited to the quantile interval of the historical error statistical distribution, and the BMS health status estimation offset is limited to the coupling correlation threshold between the state of charge and the state of health; the parameter update amplitude is constrained by adding the L2 norm regularization term to prevent overfitting; a loss function with a regularization term is constructed, and the optimal parameter combination that meets the physically feasible interval constraints is iteratively solved based on the gradient descent algorithm; the updated battery aging model parameters are generated to ensure that the parameters achieve a balance between physical interpretability and model accuracy.

[0101] Based on the execution feature dataset, the temperature-dependent aging features are generated by wavelet transform-Arrhenius temperature feature extractor;

[0102] It should be noted that the specific process of generating temperature-dependent aging features through the wavelet transform-Arrhenius temperature feature extractor based on the execution feature data set is as follows: perform wavelet transform multi-scale decomposition on the BMS temperature parameter time series record, extract the low-frequency trend component and high-frequency fluctuation component of the temperature waveform, and identify temperature transient impact events, such as rapid heating or cooling segments and steady-state temperature rise periods; input the low-frequency trend component into the Arrhenius equation to calculate the average aging rate coefficient under the steady-state temperature rise period, and calculate the cumulative contribution of the high-frequency fluctuation component to the aging rate through time-frequency energy integration; fuse the ambient temperature gradient in the PCS operating temperature waveform and the temperature distribution difference inside the BMS to generate a temperature spatial non-uniformity compensation factor; combine the steady-state aging rate coefficient, transient impact contribution and temperature non-uniformity compensation factor, and generate a temperature-dependent aging feature vector through a weighted fusion formula to characterize the coupled influence of temperatures of different time scales and spatial dimensions on the battery aging dynamic model, and provide physical driving feature input for the parameter calibration layer.

[0103] Based on the temperature-dependent aging characteristics, a spectrum clustering analyzer is used to perform multi-dimensional feature clustering to generate a power grid fluctuation feature library.

[0104] It should be explained that the specific process of generating a power grid fluctuation feature library by performing multi-dimensional feature clustering through a spectrum clustering analyzer based on temperature-dependent aging characteristics is as follows: extracting the time series of the steady-state aging rate coefficient, the frequency domain energy distribution of the transient impact contribution, and the spatial gradient parameters of the temperature non-uniformity compensation factor in the temperature-dependent aging characteristics to construct a multi-dimensional feature matrix; calculating the similarity matrix of the feature matrix through the Gaussian kernel function, generating a Laplace matrix and performing feature decomposition to obtain a low-dimensional embedding vector representation; based on the correlation constraint of the PCS AC voltage harmonic distortion rate and the EMS frequency offset, performing K-means clustering on the low-dimensional embedding vector to divide the high-frequency harmonic fluctuation mode, low-frequency oscillation fluctuation mode, and transient impact fluctuation mode; associating the clustering results with the PCS voltage waveform segment, the EMS load scheduling period, and the BMS temperature transient event to generate a power grid fluctuation feature library containing fluctuation mode labels, feature vectors, and timestamps, providing temperature-grid interaction coupling feature input for the battery aging dynamic model.

[0105] In summary, the present invention realizes the nonlinear coupling modeling of cyclic attenuation and calendar aging mechanisms in the battery aging dynamic model through: the physical driving feature layer and the dynamic parameter calibration layer, thereby improving the health status prediction accuracy and model adaptability.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An integrated scheduling system for PCS, EMS and BMS, characterized by: include, Data preprocessing module collects equipment operation status data in real time, performs preprocessing, and generates a standardized state vector; The multi-dimensional performance evaluation module builds and trains a dynamic battery aging model. Based on the standardized state vector, it dynamically corrects the battery aging dynamic model and maps it to a digital twin to generate multi-dimensional performance evaluation indicators. The multi-objective optimization module generates a collaborative scheduling strategy set based on multi-dimensional performance evaluation indicators through a multi-objective optimization solver combining fuzzy analytic hierarchy process with an improved genetic algorithm; The dynamic derating module, based on the collaborative scheduling strategy set, generates an executable instruction queue with security constraints through the industrial Internet of Things protocol stack combined with the dynamic derating coefficient algorithm; The fuse monitoring module, based on the executable instruction queue, generates an execution feature data set with abnormal coding through real-time fuse mechanism and power quality monitoring algorithm; The closed-loop optimization module updates the battery aging model parameters through an incremental online learning algorithm based on the execution feature dataset, and generates a power grid fluctuation feature library through the Arrhenius equation parameter optimizer.

2. The integrated scheduling system for implementing PCS, EMS and BMS according to claim 1, characterized in that: The equipment operation status data includes PCS power and voltage, EMS load forecast data and BMS charge state and health status parameters.

3. The integrated scheduling system for PCS, EMS and BMS according to claim 2, characterized in that: The preprocessing includes denoising, normalization, multi-source heterogeneous data time alignment and feature fusion.

4. The integrated scheduling system for implementing PCS, EMS and BMS according to claim 3, characterized in that: The specific steps of building a dynamic battery aging model and training it are as follows: The input layer is constructed based on the multi-source heterogeneous data fusion method to convert the standardized state vector into a time series tensor; The Arrhenius multi-mechanism coupling analysis method is used to extract the joint characteristics of cyclic decay and calendar aging from the input data, and a shared physical driving feature layer is constructed. By improving the particle swarm optimization algorithm, nonlinear parameter fitting is performed on shared features to construct a parameter calibration layer; The calibrated cyclic attenuation factor is fused with real-time data through the extended Kalman filter algorithm to construct a dynamic state estimation layer; A battery aging dynamic model is constructed based on the input layer, physical driving feature layer, parameter calibration layer and dynamic state estimation layer and trained using a hybrid collaborative training method.

5. The integrated scheduling system for implementing PCS, EMS and BMS according to claim 4, characterized in that: Based on the standardized state vector, dynamic correction and digital twin mapping are performed through the battery aging dynamic model to generate multi-dimensional performance evaluation indicators. The specific steps are as follows: Based on the standardized state vector, the health status is predicted through the battery aging dynamic model to generate the health status indicator; Based on the health status indicators, dynamic state correction is performed through the extended Kalman filter algorithm to generate real-time health status estimation values; Based on the real-time health status estimation value, multi-dimensional performance quantification is performed through the digital twin mapping engine to generate multi-dimensional performance evaluation indicators.

6. The integrated scheduling system for implementing PCS, EMS and BMS according to claim 5, characterized in that: The collaborative scheduling strategy set is generated based on the multi-dimensional performance evaluation index by combining the fuzzy hierarchical analysis method with the multi-objective optimization solver of the improved genetic algorithm. The specific steps are as follows: Based on multi-dimensional performance evaluation indicators, multi-objective weight allocation is performed through fuzzy analytic hierarchy process to generate priority weight vectors; Based on the priority weight vector, the multi-objective optimization solution is solved by using the improved non-dominated sorting genetic algorithm to generate the Pareto optimal solution set; Based on the Pareto optimal solution set, the TOPSIS decision method is used to perform strategy optimization and generate the final collaborative scheduling strategy set.

7. The integrated scheduling system for implementing PCS, EMS and BMS according to claim 6, characterized in that: Based on the collaborative scheduling strategy set, the industrial Internet of Things protocol stack is combined with the dynamic derating coefficient algorithm to generate an executable instruction queue with security constraints. The specific steps are as follows: Based on the collaborative scheduling strategy set, the command is securely encapsulated through the industrial Internet of Things protocol stack to generate protocol-compatible command frames; Based on the protocol-compatible instruction frame, security constraints are injected through the dynamic derating coefficient algorithm to generate risk response instructions; Based on the risk response instructions, an executable instruction queue with security constraints is generated through an instruction priority sorting algorithm.

8. The integrated scheduling system for implementing PCS, EMS and BMS according to claim 7, characterized in that: Based on the executable instruction queue, the execution feature data set with abnormal coding is generated through the real-time fuse mechanism and power quality monitoring algorithm. The specific steps are as follows: Based on the executable instruction queue, overload risk monitoring is performed through a real-time circuit breaker mechanism, generating circuit breaker events and exception tags; Based on the execution process data of the executable instruction queue, dynamic parameter analysis is performed through the power quality monitoring algorithm to generate quality abnormality waveforms; Based on the circuit-breaking events and quality abnormal waveforms, feature fusion is performed through the abnormal coding rule library to generate an execution feature dataset with security labels.

9. The integrated scheduling system for implementing PCS, EMS and BMS according to claim 8, characterized in that: The battery aging model parameters are updated through an incremental online learning algorithm based on the execution feature data set. The specific steps are as follows: Based on the execution feature dataset, dynamic parameter updates are performed through an online sequential extreme learning machine to generate preliminary calibration parameters. Based on the preliminary calibration parameters, physical range correction is performed through a regularized constrained optimizer to generate updated battery aging model parameters.

10. The integrated scheduling system for implementing PCS, EMS and BMS according to claim 9, characterized in that: The specific steps of generating the power grid fluctuation feature library by the Arrhenius equation parameter optimizer are as follows: Based on the execution feature dataset, the temperature-dependent aging features are generated by wavelet transform-Arrhenius temperature feature extractor; Based on the temperature-dependent aging characteristics, a spectrum clustering analyzer is used to perform multi-dimensional feature clustering to generate a power grid fluctuation feature library.

Images