CN-121980896-A - Liquid metal battery capacity prediction method, system, equipment and medium based on Stacking model

CN121980896ACN 121980896 ACN121980896 ACN 121980896ACN-121980896-A

Abstract

The invention relates to the technical field of energy storage battery capacity, and discloses a liquid metal battery capacity prediction method, a system, equipment and a medium based on a Stacking model, wherein the method comprises the steps of selecting a gradient lifting decision tree, a random forest and a support vector regression as a basic learner, and using a linear regression as a meta learner; the method comprises the steps of constructing a Stacking model through Stacking integrated learning, training the model by adopting a cross-validation method to prevent overfitting, carrying out a liquid metal battery aging experiment to obtain historical capacity data, inputting the historical capacity data in a battery circulation process into the trained Stacking model, and predicting future capacity change. The method fully plays the advantages of the selected basic model, effectively fuses the sensitivity of different models to the aging characteristics of the liquid metal battery, and comprehensively improves the accuracy of the capacity prediction of the liquid metal battery through the comprehensive capture of the aging characteristics.

Inventors

FAN LEI
LIU XI
Long Junxu
Fu Ninglong
JIANG GAOFENG
WANG MINGWEI
PANG LINGRONG
GUO ZHENGWEI
HU RONGJUN
TANG XUEYONG
LI AONAN
TIAN TAO
TENG YANG

Assignees

贵州电网有限责任公司

Dates

Publication Date: 20260505
Application Date: 20251128

Claims (10)

1. The liquid metal battery capacity prediction method based on the Stacking model is characterized by comprising the following steps of, Selecting a gradient lifting decision tree, a random forest and a support vector regression as a basic learner, and selecting a linear regression as a basic learner; Based on a Stacking integrated learning method, taking the prediction output of the base learner as new characteristic input, training the meta learner, and constructing a layering Stacking prediction model; Carrying out a liquid metal battery aging experiment, and acquiring historical capacity attenuation data of the battery in a cyclic charge and discharge process to form a data set for model training and testing; Dividing the historical capacity data training set into a plurality of data subsets by adopting a data cross-dividing method, sequentially training the base learner by utilizing the data subsets, and combining the prediction results of the base learner on the corresponding test subsets into a new training data set for training the meta learner; And inputting the historical capacity data of the liquid metal battery to be predicted into a trained stacking model, generating a primary prediction result through the base learner, integrating the primary prediction result through the element learner, and outputting a predicted value of the future capacity of the battery.
2. The method for predicting the capacity of the liquid metal battery based on the Stacking model of claim 1, wherein the cross-partitioning data comprises partitioning the whole training set into k data subsets with the same data size; Sequentially taking one data subset as a test set of the base learner, taking the rest k-1 subsets as training sets, and training each base learner to obtain k models of each base learner; and inputting each test subset into a corresponding basic learner model, and combining the prediction results into a new characteristic data set for training the meta learner.
3. The method for predicting the capacity of the liquid metal battery based on the Stacking model of claim 2, wherein the Stacking model comprises the steps of respectively inputting characteristic data of a test set into k models corresponding to each base learner to obtain k groups of prediction results of each base learner; Carrying out average processing on k groups of prediction results of the same base learner to obtain final primary prediction output of each base learner; And combining the final primary prediction output of each base learner into a feature vector, and inputting the feature vector into the trained meta learner to obtain a final capacity prediction result.
4. The method for predicting the capacity of the liquid metal battery based on the Stacking model of claim 3, wherein the gradient boosting decision tree comprises the steps of constructing a decision tree in a sequential manner by adopting a gradient boosting idea, wherein each subsequent tree is trained based on a prediction residual of a previous tree, gradually minimizing a loss function, and capturing complex nonlinear trends in capacity fading data; the random forest comprises the steps of constructing decision trees through a Bagging parallelization integration mode, training each tree by using a returned random sampling sample and a feature subset of a training set, and averaging prediction results of all trees; the support vector regression includes processing a high-dimensional nonlinear capacity data relationship by mapping input data to a high-dimensional feature space and constructing a regression function that balances model complexity and prediction error during optimization.
5. The method for predicting the capacity of a liquid metal battery based on a Stacking model of claim 4, wherein selecting the linear regression as a meta-learner comprises learning and linearly combining the predictive dominance of each base learner by receiving a feature vector composed of the final primary predictive output of the base learner and solving a weight parameter and a bias term using a least squares method to generate a final stacked model predictive output.
6. The method for predicting the capacity of the liquid metal battery based on the Stacking model of claim 5, wherein inputting the historical capacity data of the liquid metal battery to be predicted into the Stacking model after training comprises a continuous capacity data sequence from one starting cycle number q to another cycle number m+q, wherein the sequence defines a time window for predicting the future capacity; the future capacity data is output in the form of a sequence of consecutive capacity data starting from the cyclic sequence number m+q+1 and ending with the cyclic sequence number m+q+n, the sequence defining a future time span to be predicted.
7. The method for predicting capacity of liquid metal battery based on Stacking model as set forth in claim 6, wherein in the model prediction stage, when the k sets of prediction results of each base learner are averaged, the primary prediction output of the base learner is obtained by summing k output values and dividing the sum by k by adopting an arithmetic average method.
8. The liquid metal battery capacity prediction system based on the Stacking model is characterized by comprising a data management module, a model construction module, a training engine and a prediction engine, wherein the method for predicting the liquid metal battery capacity based on the Stacking model is disclosed in any one of claims 1-7; The data management module is used for receiving, storing and preprocessing historical capacity data obtained from a liquid metal battery aging experiment, dividing the historical capacity data into a training set and a testing set and providing data support for model training and prediction; The model construction module is used for configuring the structure of the Stacking model, comprising the steps of designating gradient lifting decision trees, random forests and support vector regression as a basic learner, designating linear regression as a meta learner and defining a logic flow of Stacking integrated learning; The training engine is connected with the data management module and the model construction module and is used for executing a training process of the model, training the base learner by calling a method of cross division data, constructing a new training feature set by utilizing the output of the base learner to train the element learner, and finally generating an available stack prediction model; The prediction engine is connected with the data management module and the training engine and is used for loading a final stacking model generated by the training engine, receiving historical capacity data of the battery to be predicted, coordinating the base learner to perform primary prediction, calling the element learner to integrate results, and finally outputting a future capacity prediction sequence.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the Stacking model based liquid metal battery capacity prediction method of any one of claims 1 to 7.
10. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor implements the steps of the Stacking stack model based liquid metal cell capacity prediction method of any one of claims 1 to 7.

Description

Liquid metal battery capacity prediction method, system, equipment and medium based on Stacking model Technical Field The invention relates to the technical field of energy storage battery capacity prediction, in particular to a liquid metal battery capacity prediction method, system, equipment and medium based on a Stacking model. Background In recent years, with the increase of environmental pollution and the increasing exhaustion of fossil energy, the development of clean energy and research and application of efficient energy storage technology have become particularly important. The energy storage system is used as a key hub for connecting energy production and consumption, and the performance of the energy storage system directly determines the feasibility and economy of large-scale access of renewable energy sources to a power grid. Among the energy storage technologies, the liquid metal battery is gradually becoming a powerful competitor of the new generation of electrochemical energy storage technology due to the unique structural design and excellent performance, and has a wide application prospect. The liquid metal battery is generally composed of a liquid metal cathode, a liquid alloy anode and an intermediate molten salt electrolyte, wherein the three components are kept in a liquid state at the working temperature and naturally layered due to the density difference and the characteristic of mutual incompatibility, so that a stable liquid-liquid three-layer structure is formed. The full-liquid structure not only enables the battery interface to be smoother and the reaction kinetics to be faster, but also avoids the common structure degradation mechanisms of the traditional solid-state electrode, such as the problems of active material falling, volume expansion, lithium dendrite growth and the like, thereby remarkably improving the cycle life and the safety performance of the battery. Furthermore, liquid metal batteries also have significant advantages in cell manufacturing and modular assembly due to the high degree of structural adaptability to the manufacturing process. In the actual running process of the battery, the capacity is used as a core index for measuring the performance and the health state of the battery, and the change of the capacity reflects various internal mechanisms such as electrochemical reaction efficiency, interface stability, material loss and the like. Particularly in the long-term operation of liquid metal batteries, although a microscopic damage mechanism of a solid electrode does not exist, the liquid metal battery is still influenced by factors such as material interdiffusion, corrosion, electrolyte component evolution and the like, so that the capacity is slowly attenuated. Therefore, the method for accurately predicting the capacity change of the battery in the future has important significance for guaranteeing the stable service and improving the system operation efficiency. The capacity prediction not only can realize the dynamic evaluation of the battery health state and the early warning of fault trend, but also can provide key support for intelligent scheduling, operation maintenance and service life management of the energy storage system. Along with the wide application of the liquid metal battery in the fields of power peak regulation, renewable energy source access and the like, the construction of a capacity prediction model with high precision and strong robustness becomes one of important research directions in the engineering application process. Capacity prediction methods are mainly classified into model-based methods and data-driven-based methods. The traditional model-based method mainly obtains aging characteristic parameters of the battery through an empirical model, an electrochemical model and an equivalent circuit model, and predicts the capacity of the battery by combining Kalman filtering, particle filtering and other methods. The health state estimation method based on the model has the advantages of strong physical interpretation, lower data demand, complex modeling, high calculation cost, difficult model parameter acquisition, limited adaptability of actual working conditions and possibility of increasing estimation errors in dynamic or extreme environments. The data-driven based methods are largely divided into predictions based on health factors and predictions based on historical capacity trajectories. The first type of method is to extract health factors with aging characteristics by analyzing experimental data of battery circulation, and then input machine learning models such as Gaussian process regression and the like to obtain a capacity prediction result of the battery. The method needs to acquire experimental data of the battery in real time, has higher data requirements and is not easy to realize. The second type of method can realize future capacity prediction by only using historical capacity data of the battery and combini