Research On Methods About Estimating State-of-health Of VRLA Battery In Data Centers

Valve regulated lead acid(VRLA)batteries are widely used in data centers because of their good safety performance and price advantage.State-of-health(SOH)is used to evaluate the aging degree of batteries and is a key factor affecting the power supply performance and safe operation of batteries.At present,the recognized SOH measurement method is full capacity verification test method,but the battery discharge costs too much time and is too deep to meet operation needs.Therefore,it is very necessary to study the SOH estimation method of VRLA batteries in data centers.The SOH estimation method based on equivalent circuit model and the SOH estimation method based on black box model are studied respectively,and the constant current discharge data with discharge depth of 50% is used to realize the accurate estimation of battery SOH.In the research of SOH estimation method based on equivalent circuit model,firstly,the dual Kalman filter algorithm with different innovation matrices is used to realize SOH estimation combined with the inverse process of Ampere-hour integration method.The algorithm has better estimation accuracy.However,the robustness is weak.After parameter disturbance,the percentage error of SOH estimation rises by 256.64%.Then,the algorithm structure is studied and improved,the coupling structure of parameter identification and state estimation is established,and a dual extended Kalman filter algorithm with the same innovation matrix is proposed.The robustness of the algorithm is significantly improved compared with that before the improvement.After parameter disturbance,the percentage error of SOH estimation rises by 10.75%.Finally,a multitime scale method is introduced to improve the estimation accuracy of the algorithm,and a multi-time scale dual extended Kalman filter algorithm with innovation matrix is proposed.The application results show that the algorithm obtains the best estimation accuracy.The estimation percentage errors on the battery discharge data with SOH of200 Ah,175Ah and 150 Ah are 1.09%,1.22% and 2.59% respectively.In the research of SOH estimation method based on black box model,the long shortterm memory network shows strong adaptability to high-dimensional,multivariable and nonlinear time series data.The mean absolute percentage error(MAPE)of the estimation index on the data set is 0.64%,which is better than 0.72% of particle swarm optimization support vector regression model.Combined with the difference of the contribution of battery characteristics to SOH,a long short-term memory network based on spatiotemporal attention is designed.The application results show that the proposed design method can effectively improve the estimation accuracy and convergence speed of SOH.The estimated index MAPE on the battery samples with SOH of 200 Ah,175Ah and 150 Ah is 0.66%,0.89% and 1.31% respectively,and the estimated index MAPE on the test set is 0.43%.The method based on equivalent circuit model does not need a large number of sample data,and the observation is easy to obtain;The method based on black box model does not need to consider battery mechanism,and is simple to implement and easy to understand.In practical application,the appropriate SOH estimation method can be selected according to the situation of the use scenario.