State estimation and prediction method of energy storage system based on online data fusion reduced-order mechanism

Abstract

The present invention relates to the technical field of energy storage systems, and discloses an online data fusion reduced-order mechanism energy storage system state estimation and prediction method, comprising the following steps: 1) determining the input and output parameters of the reduced-order mechanism model; 2) exporting the numerical simulation data of the cross working condition to obtain the original data of the training set of the reduced-order mechanism model; 3) obtaining the data training set of the reduced-order mechanism model; 4) training the reduced-order mechanism model to obtain an offline reduced-order mechanism model, switching to the simulation operation condition, and verifying the trained reduced-order mechanism model in combination with the simulation data; 5) the energy storage system is operated to form a digital twin of the online data fusion reduced-order mechanism model; 6) combining the online data with the digital twin, online inversion of the electrochemical parameters or equivalent circuit parameters of the internal dynamics and thermodynamic processes of the energy storage system, online estimation and prediction of the fault monomer, and obtaining evaluation and prediction data.

Description

Technical Field

[0001] The patent of this invention relates to the technical field of energy storage systems, specifically, to a state estimation and prediction method for energy storage systems using an online data fusion reduced-order mechanism. Background Art

[0002] Lithium battery energy storage technology is one of the core technologies that promotes the clean, electrified and efficient world energy, breaks the constraints of energy resources and the environment, and realizes the global energy transformation and upgrading. However, electrochemical energy storage systems are inherently unsafe, and safety issues have gradually become the primary issue of lithium battery energy storage systems. Accurate and efficient state estimation and prediction of the system is crucial for the reasonable management and safe operation of lithium battery energy storage systems.

[0003] In the existing technology, the safety management of lithium batteries mainly determines whether the battery is safe by judging whether the monitoring data such as voltage, current, and temperature exceed the threshold range. However, the number of sensors collected by the energy storage system is limited, and it is impossible to quickly obtain the global status, making it difficult to evaluate the system operation status in real time. The threshold judgment method can only trigger an alarm when the fault evolves to a certain extent or even triggers a battery thermal runaway chain reaction. At this time, it is difficult to carry out effective safety management.

[0004] At present, there are two methods for estimating the state of energy storage systems. The first method is to combine the system operation history data with the mechanism model for offline estimation. At this time, the energy storage system is in a static state and cannot play a role in fault prediction. It can only be effectively identified after the system fails. If the battery fault does not appear, it cannot be actively identified and judged. The other method is to obtain the global state of a specific problem through numerical simulation. However, each numerical simulation process is targeted at a problem and given conditions. Any change requires recalculation of the problem. There is a lot of redundant work, and it is also offline calculation, which cannot be directly applied to the system's online state estimation and prediction.

[0005] Therefore, for technicians in this field, developing a safety management and control system that can perform online real-time state estimation and prediction is a problem that needs to be solved at present. Summary of the invention

[0006] The purpose of the present invention is to provide an online data fusion reduced-order mechanism energy storage system state estimation and prediction method, aiming to solve the problem in the prior art that the energy storage system state cannot be evaluated and predicted online.

[0007] The present invention is implemented in this way: an online data fusion reduced-order mechanism energy storage system state estimation and prediction method comprises the following steps:

[0008] 1) Determine the input and output parameters of the reduced-order mechanism model;

[0009] 2) Customize the node parameters of the time interval, export the numerical simulation data of the cross-operating condition, and obtain the original data of the training set of the reduced-order mechanism model through simulation;

[0010] 3) Obtain the node status values ​​of all nodes, complete the preprocessing of the grid data, and obtain the data training set of the reduced-order mechanism model;

[0011] 4) Training the reduced-order mechanism model to obtain an offline reduced-order mechanism model, switching to a simulation operation condition, and verifying the trained reduced-order mechanism model in combination with the mechanism model simulation data;

[0012] 5) During the operation of the energy storage system, the reduced-order mechanism model is verified and updated through the online data obtained through monitoring, forming a digital twin of the reduced-order mechanism model fused with online data;

[0013] 6) The online data is transmitted to the data processing module for preprocessing, and the preprocessed online data is subjected to feature recognition. The online data is combined with the digital twin to perform online inversion of the electrochemical parameters or equivalent circuit parameters of the internal dynamics and thermodynamic processes of the energy storage system, online estimation and prediction of faulty monomers, and obtain evaluation and prediction data.

[0014] Furthermore, in step 6), the preprocessed online data and online evaluation and prediction data are transmitted to the historical database for digital twin training and updating.

[0015] Furthermore, in step 6), the evaluation and prediction data, faulty cells and system status are displayed in three-dimensional visualization.

[0016] Furthermore, in the step 2), after selecting the battery simulation model, characteristic parameters are extracted and boundary conditions are adjusted to perform mechanism model simulation, so as to obtain the original data of the training set of the reduced-order mechanism model.

[0017] Furthermore, in the step 3), multiple original nodes closest to the reduced-order nodes are found through the K-nearest neighbor algorithm, radius nearest neighbor search, and local sensitive hash search algorithm; the node state values ​​of all nodes are obtained through spline interpolation, polynomial interpolation, and radial basis function interpolation methods to complete the training data preprocessing.

[0018] Furthermore, in step 4), the preprocessed training data is combined with a neural network algorithm to train the reduced-order mechanism model.

[0019] Furthermore, in step 5), the online data includes but is not limited to the voltage, current, monitoring point temperature, fan speed of the air cooling system, and fluid flow rate of the liquid cooling system when the energy storage system is working normally.

[0020] Furthermore, in step 6), online data obtained during the operation of the energy storage system is transmitted to the data processing module via the BMS.

[0021] Furthermore, the preprocessed online data is feature-identified and used as the input of the reduced-order mechanism model to forward predict the real-time state of the system; as the output of the reduced-order mechanism model, it is combined with the particle swarm optimization algorithm, extended Kalman filter, and genetic algorithm inversion algorithm to perform online inversion estimation of the electrochemical parameters or equivalent circuit parameters of the internal dynamics and thermodynamic processes of the energy storage system.

[0022] Compared with the prior art, the online data fusion reduced-order mechanism energy storage system state estimation and prediction method provided by the present invention has the following beneficial effects:

[0023] 1) The reduced-order mechanism model formed by training is a fusion of the physical mechanism model and the data-driven model. The online data correction and update model is adopted to achieve accurate and efficient online evaluation of the energy storage system status;

[0024] 2) The speed of the reduced-order mechanism model has also been improved by thousands of times compared with numerical simulation, providing a new idea for the application of complex numerical models in online simulation of actual engineering;

[0025] 3) Driven by online data, based on the reduced-order mechanism model, and using digital twins as a means, the mutual mapping and collaborative interaction of the physical world and virtual space of the energy storage system are realized, thereby constructing a digital twin of the energy storage system based on data-driven, model-supported, virtual-real interaction;

[0026] 4) Realize intelligent evaluation of energy storage status, fault inversion analysis, and accurate trend prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart for constructing a digital twin of the online data fusion reduction mechanism model provided by the present invention;

[0028] Figure 2 It is a system architecture diagram of the online data fusion reduced-order mechanism energy storage system state estimation and prediction method provided by the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0030] The implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0031] The same or similar numbers in the drawings of this embodiment correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right" and the like indicate directions or positional relationships based on the directions or positional relationships shown in the drawings, it is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and cannot be understood as limitations on this patent. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0032] Reference Figure 1-2 The figure shows a preferred embodiment of the present invention.

[0033] 1. An online data fusion reduced-order mechanism energy storage system state estimation and prediction method, characterized in that it comprises the following steps:

[0034] 1) Determine the input and output parameters of the reduced-order mechanism model, and customize the training conditions within the parameter range to ensure that the situations that may occur in the subsequent operation process are included as much as possible.

[0035] The input and output parameters described here include parameters that can be monitored in real time during system operation (voltage, current, monitoring point temperature, fan speed of the air cooling system, fluid flow rate of the liquid cooling system, etc.) and unmonitored parameters (electrochemical parameters, equivalent circuit parameters, physical parameters, heat generation, SOC, etc.), which are flexibly selected according to the problem that the model needs to solve.

[0036] 2) Customize the node parameters of the time interval, export the simulation data of the cross-operating condition, and obtain the original data of the training set of the reduced-order mechanism model through simulation;

[0037] Change the boundary conditions, simulate the mechanism model, and export the simulation data of the cross-operating condition in step 1) to obtain the original training data of the reduced-order mechanism model. The simulation export strategy here is the module grid node temperature value, heat source term and other parameters calculated at fixed time intervals.

[0038] 3) Obtain the node state values ​​of all grid nodes, pre-process the grid data, and obtain the data training set of the reduced-order mechanism model;

[0039] Since the number of simulation grids for energy storage systems may reach millions, the time cost and computing power required for subsequent model training are very high. The preprocessing here aims to reduce the number of grid nodes and reduce the training time of subsequent models while fully reflecting the system status.

[0040] The preprocessing steps are as follows: construct a coarse grid (with a small number of nodes) for the system under consideration, find the n original simulation nodes closest to the reduced-order nodes through a search algorithm, and then obtain the node state values ​​of all coarse grids through node data interpolation.

[0041] 4) Training the reduced-order mechanism model to obtain an offline reduced-order mechanism model, switching to a simulation operation condition, and verifying the trained reduced-order mechanism model in combination with simulation data;

[0042] A neural network model is trained for each node after preprocessing to obtain an offline reduced-order mechanism model, and the trained reduced-order mechanism model is verified in combination with the simulation data of the verification condition;

[0043] The model training described here numbers each node, each node corresponds to a neural network, and finally the model parameters are centrally output and saved for subsequent system management module calls.

[0044] 5) During the operation of the energy storage system, the reduced-order mechanism model is verified and updated through the online data obtained through monitoring, forming a digital twin of the reduced-order mechanism model fused with online data;

[0045] The model is integrated into the system, and the reduced-order mechanism model is verified and updated through online data obtained through monitoring, forming a digital twin of the reduced-order mechanism model with online data fusion. The trained reduced-order mechanism model is a fusion of the physical mechanism model and the data-driven model, in which the physical mechanism model provides data type supplement for the data drive to ensure the accuracy of the data-driven model in reconstructing the physical field. At the same time, the data regularly monitored by the energy storage system provides updates and verifications for the data-driven model and the physical mechanism model, forming a digital twin model of the energy storage system.

[0046] 6) The online data is transmitted to the data processing module for preprocessing, and the preprocessed online data is subjected to feature recognition. The online data is combined with the digital twin to perform online estimation and prediction of the energy storage system status.

[0047] The constructed digital twin can achieve forward prediction, that is, taking online monitorable data as model input to estimate the future state of the system under this working condition. In addition, for the system's unmonitorable data, PSO, EKF and other algorithms are used for iterative inversion, and the electrochemical parameters or equivalent circuit parameters of the internal dynamics and thermodynamic processes of the energy storage system are inverted online through the monitorable data. Specifically, the relative error between the predicted value of the reduced-order mechanism model and the measured value of the system is used as the objective function, and the PSO algorithm is used to find the initial extreme point. Through continuous iterative optimization, new extreme points are obtained. Finally, it is determined whether the error between the predicted value and the measured value is within an acceptable range, and the inversion result is output.

[0048] The above-mentioned online data fusion reduced-order mechanism energy storage system state estimation and prediction method has the following beneficial effects:

[0049] 1) The reduced-order mechanism model formed by training is a fusion of the physical mechanism model and the data-driven model. The online data correction and update model is adopted to achieve accurate and efficient online evaluation of the energy storage system status;

[0050] 2) The speed of the reduced-order mechanism model has also been improved by thousands of times compared with numerical simulation, providing a new idea for the application of complex numerical models in online simulation of actual engineering;

[0051] 3) Driven by online data, based on the reduced-order mechanism model, and using digital twins as a means, the mutual mapping and collaborative interaction of the physical world and virtual space of the energy storage system are realized, thereby constructing a digital twin of the energy storage system based on data-driven, model-supported, virtual-real interaction;

[0052] 4) Realize intelligent evaluation of energy storage status, fault inversion analysis, and accurate trend prediction.

[0053] In this embodiment, in step 6), the preprocessed online data and the online evaluation and prediction data are transmitted to the historical database for digital twin training and updating.

[0054] The preprocessed online data is combined with the rich historical status data of the digital twin and neural network algorithms such as convolutional neural network (CNN) and long short-term memory artificial neural network (LSTM) to realize the prediction of the time scale of the energy storage system status.

[0055] In step 6), the evaluation and prediction data, faulty cells, and system status are visualized in three dimensions. Combined with the three-dimensional visualization platform, a full-time, multi-domain visualization of the energy storage system is achieved, providing information support for the reasonable management and safe operation of the energy storage system.

[0056] In step 6), the preprocessed online data is combined with the rich historical status data of the digital twin, and neural network algorithms such as convolutional neural network (CNN) and long short-term memory artificial neural network (LSTM) to realize the prediction of the time scale of the energy storage system state.

[0057] Compare the evaluation and prediction data in step 6) with the actual measurement, analyze the errors between the two, and evaluate the system operation status. Combined with the 3D visualization platform, the evaluation and prediction data, fault cells, and system status are visualized in 3D to provide information support for the reasonable management and safe operation of the energy storage system.

[0058] In step 2), after selecting the numerical simulation model, characteristic parameters are extracted and boundary conditions are adjusted to simulate the mechanism model, and the original data of the training set of the reduced-order mechanism model is obtained.

[0059] In step 2), the simulation data under all cross-operating conditions are exported, and the export strategy is the module grid node temperature value, heat source term and other parameters calculated at fixed time intervals.

[0060] If we directly train an agent model for each node to estimate and predict the system state under unknown working conditions, the time cost and computing power required for the energy storage system may reach millions, and we need to reduce the number of nodes while ensuring that the system state can be fully reflected.

[0061] In step 3), a search algorithm is used to find multiple original nodes closest to the reduced-order node, and node state values ​​of all nodes are obtained by interpolating node data.

[0062] In step 3), multiple original nodes closest to the reduced-order node are found through search algorithms such as K-NearestNeighbor, RadiusNearestNeighbor Search, and Locality Sensitive Hashing.

[0063] Through interpolation methods such as spline interpolation, polynomial interpolation, radial basis function interpolation, etc., the node state values ​​of all nodes are obtained to complete the training data preprocessing.

[0064] In step 4), the reduced-order mechanism model is trained in combination with a neural network algorithm.

[0065] In step 5), online data includes but is not limited to the voltage, current, monitoring point temperature, fan speed of the air cooling system, fluid flow rate of the liquid cooling system, etc. when the energy storage system is working normally, and can be monitored in real time.

[0066] In step 6), the online data obtained from the operation of the energy storage system is transmitted to the data processing module through the BMS.

[0067] After feature recognition of the preprocessed online data, the electrochemical parameters or equivalent circuit parameters of the internal dynamics and thermodynamic processes of the energy storage system are inverted online based on various inversion algorithms.

[0068] After feature recognition, the preprocessed online data is used as the input of the reduced-order mechanism model to forward predict the real-time state of the system; as the output of the reduced-order mechanism model, combined with inversion algorithms such as particle swarm optimization (PSO), extended Kalman filter (EKF), genetic algorithm (GA), etc., the electrochemical parameters or equivalent circuit parameters of the internal dynamics and thermodynamic processes of the energy storage system are estimated online.

[0069] In step 6), the online data is transmitted to the data processing module for preprocessing and standardization. The obtained data can be stored in the historical database for retrieval and also transmitted to the digital twin.

[0070] In step 4), the trained reduced-order mechanism model is a fusion of the physical mechanism model and the data-driven model, wherein the physical mechanism model provides data type supplement for the data driver to ensure the accuracy of the data-driven model in reconstructing the physical field. At the same time, the data regularly monitored by the energy storage system provides updates and verifications for the data-driven model and the physical mechanism model to form a digital twin model of the energy storage system.

[0071] Two model training cases are provided below. Both physical models are liquid cooling modules, which are composed of 24 100Ah lithium iron phosphate prismatic batteries placed on a liquid cooling plate with built-in liquid cooling channels to achieve liquid cooling. The system works under frequency regulation conditions, that is, the working charge / discharge current changes with time. The physical mechanism model uses an electrothermal fluid coupling model to simulate the mechanism model under different working conditions by changing the boundary conditions, and derive the training results.

[0072] Case 1 aims to achieve the goal of forward prediction of work by the reduced-order mechanism model, and Case 2 achieves internal state inversion through the reduced-order mechanism model.

[0073] Case 1

[0074] 1) Input and output parameter selection and determination. In order to explore the influence of cooling medium on the module temperature field, the cooling medium flow rate (v), inflow temperature (T in ) and working time (t) as the input of the neural network. The input parameter matrix is ​​expressed as The temperature value of each grid node is taken as output, and the output matrix is ​​represented as Y i =[T] T , T is the lattice node temperature, and i is the lattice sequence number.

[0075] 2) Training condition selection. Within the boundary value range, the cooling medium flow rate selected for the training condition is 0.5m / s, 1m / s, 1.5m / s, and the medium temperature is 288K, 293K, and 298K. All parameters form 9 different simulation conditions, and the node temperature value of the battery module is recorded every 1s during calculation.

[0076] 3) Numerical simulation and data preprocessing. After numerical simulation, 64,800 sets of original training data were obtained. The number of nodes contained in each set of data was 230,904. After data preprocessing and reduction, the number of data nodes for the reduced-order mechanism model training was only 89,204, which greatly accelerated the training speed.

[0077] 4) Reduced-order mechanism model training and verification. After numbering each node, the neural network model is trained and the model training parameters are saved for subsequent calls. The verification process changes the boundary conditions to the verification conditions, and compares the numerical simulation results with the calculation results of the reduced-order mechanism model. In the subsequent calling process of the model trained by this method, if the node number is used as the basis, the single-point neural network is called to realize the temperature prediction of the characteristic point time scale. If all node neural networks are called, the three-dimensional temperature field evolution prediction of the system can be realized.

[0078] 5) Digital twin construction and model application. Integrate the model into the management system, combine it with online monitoring data, and make forward predictions by giving cooling medium parameters to predict the spatiotemporal evolution of the temperature field during system operation; combine it with the current system state, give the temperature threshold, and control the maximum temperature and maximum temperature difference as the goal to determine the best system thermal management strategy and realize dynamic online response of active thermal safety management.

[0079] 6) Data fusion and abnormality analysis. Combine the predicted temperature value in step 5) with the actual monitoring value of the system, and judge whether the system operation status is abnormal by analyzing the relative temperature error at a fixed time point and the error change rate within the time interval. If the relative error between the two is large or the error change rate increases, the system is abnormal.

[0080] Case 2

[0081] 1) Input and output parameter selection and determination. Different from case 1, the model input here is the heat source item of the corresponding battery, and the output is the temperature change within a fixed time period, that is, the model input is the internal state parameter of the battery, and the output is the system monitorable parameter.

[0082] 2) Training condition selection: In order to simplify the problem, the model cooling medium parameters are fixed. By changing different heat source items, the node temperature within 3s is simulated and derived, and the temperature change within this time period is calculated.

[0083] 3) The data preprocessing and model training of this case are the same as steps 3) and 4) of case 1.

[0084] 4) Digital twin construction and model application. The model is integrated into the management system, and the heat source term is corrected by actually monitoring the temperature change, combined with inversion algorithms such as PSO and EKF, and finally the heat source term of the battery under the corresponding temperature change is obtained, realizing the inversion mapping of the monitored parameters to the internal state.

[0085] 5) Data fusion and anomaly analysis. Combine the heat source term inverted in step 4) with the Bernardi heat generation model Compare the calculation results and determine whether the battery is generating abnormal heat based on the deviation value.

[0086] Finally, combined with the 3D visualization platform, the evaluation and prediction data, fault units, and system status in the two cases were visualized in 3D, providing information support for the rational management and safe operation of the energy storage system.

[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An online data fusion reduced-order mechanism energy storage system state estimation and prediction method, characterized in that: The following steps are involved: 1) Determine the input and output parameters of the reduced-order mechanism model; 2) Customize the node parameters of the time interval, export the numerical simulation data of the cross-operating condition, and obtain the original data of the training set of the reduced-order mechanism model through simulation; 3) Obtain the node status values ​​of all nodes, complete the preprocessing of the grid data, and obtain the data training set of the reduced-order mechanism model; 4) Training the reduced-order mechanism model to obtain an offline reduced-order mechanism model, switching to a simulation operation condition, and verifying the trained reduced-order mechanism model in combination with the mechanism model simulation data; 5) During the operation of the energy storage system, the reduced-order mechanism model is verified and updated through the online data obtained through monitoring, forming a digital twin of the reduced-order mechanism model fused with online data; 6) The online data is transmitted to the data processing module for preprocessing. The preprocessed online data is subjected to feature recognition. The online data is combined with the digital twin to perform online inversion of the electrochemical parameters or equivalent circuit parameters of the internal dynamics and thermodynamic processes of the energy storage system, and online estimation and prediction of faulty monomers are performed to obtain evaluation and prediction data. In the step 2), after selecting the battery simulation model, characteristic parameters are extracted and boundary conditions are adjusted to simulate the mechanism model, so as to obtain the original data of the training set of the reduced-order mechanism model; In the step 3), multiple original nodes closest to the reduced-order nodes are found through the K-nearest neighbor algorithm, radius nearest neighbor search, and local sensitive hash search algorithm; node state values ​​of all nodes are obtained through spline interpolation, polynomial interpolation, and radial basis function interpolation methods to complete training data preprocessing; In the step 4), the preprocessed training data is combined with a neural network algorithm to train the reduced-order mechanism model, wherein a neural network model is trained for each preprocessed node; In step 4), the trained reduced-order mechanism model is a fusion of the physical mechanism model and the data-driven model, in which the physical mechanism model provides data type supplement for the data drive to ensure the accuracy of the data-driven model in reconstructing the physical field. At the same time, the data regularly monitored by the energy storage system provides updates and verifications for the data-driven model and the physical mechanism model to form a digital twin model of the energy storage system.

2. The method for state estimation and prediction of energy storage system based on online data fusion reduced-order mechanism as claimed in claim 1, characterized in that: In step 6), the preprocessed online data and the online evaluation and prediction data are transmitted to the historical database for digital twin training and updating.

3. The online data fusion reduced-order mechanism energy storage system state estimation and prediction method according to claim 1, characterized in that: In the step 6), the evaluation and prediction data, faulty cells and system status are displayed in three-dimensional visualization.

4. The method for state estimation and prediction of an energy storage system using online data fusion reduced-order mechanism according to any one of claims 1 to 3, characterized in that: In step 5), the online data includes but is not limited to the voltage, current, monitoring point temperature, fan speed of the air cooling system, and fluid flow rate of the liquid cooling system when the energy storage system is working normally.

5. The method for state estimation and prediction of energy storage system using online data fusion reduced-order mechanism according to any one of claims 1 to 3, characterized in that: In step 6), the online data obtained during the operation of the energy storage system is transmitted to the data processing module via the BMS.

6. The method for state estimation and prediction of energy storage system using online data fusion reduced-order mechanism according to any one of claims 1 to 3, characterized in that: After feature recognition, the preprocessed online data is used as the input of the reduced-order mechanism model to forward predict the real-time state of the system; as the output of the reduced-order mechanism model, combined with the particle swarm optimization algorithm, extended Kalman filter, and genetic algorithm inversion algorithm, the electrochemical parameters or equivalent circuit parameters of the internal dynamics and thermodynamic processes of the energy storage system are estimated online.

Images