Research On Lithium-ion Battery Data Processing Based On Gaussian Process Regression

With many advantages such as high energy density, low self-discharge, high power endurance and long lifetime, the lithium-ion battery has gradually become a research hotspot in the field of energy storage, and is widely used in many fields. However, the performance of lithium-ion battery will gradually decline over repeated discharge cycles, an unexpected failure or malfunction of a battery will lead to huge loss and even catastrophic consequences, especially in aerospace field. Therefore, state of health estimation, remaining useful life prognostics and further guidance of battery operation and maintenance play significant roles in reliability of lithium-ion batteries. In this thesis, a data-driven approach Gaussian Process Regression (GPR) is applied to the data processing field of lithium-ion batteries, to perform lithium-ion batteries state of health estimation and remaining useful life prognostics.The thesis emphasized on several aspects as follows:1. The GPR model is established and applied to a prediction experiment of battery capacity data. Then the GPR prognostics performance is compared with the Autoregressive Integrated Moving Average model and Artificial Neural Network method.2. At present, the selection method of the kernel in GPR still lacks unified theoretical support, this thesis systematically explore the selection of kernel function in the GPR model, and analyze the different forecast distributions with single kernel and composite kernel, which provide the theoretical reference for following model. Then the GPR algorithm that based on composite kernel is applied to the offline modeling of lithium-ion battery capacity data.3. In order to solve the high computational complexity of GPR, online GPR algorithm based on the incremental learning algorithm is studied in this thesis, and this online model is compared the computational complexity with basic Gaussian process regression algorithm. Then the online GPR is applied to the lithium-ion battery voltage data processing, then the efficiency and accuracy of the prediction is analyzed.4. The conjugate gradient algorithm has the shortcomings of too strong dependence on initial value. In order to improve the prediction precision and generalization ability of GPR, Improved Gravitational Search Algorithm (IGSA) is utilized replace of conjugate gradient to search the optimal hyper-parameters then formed the IGSA-GPR algorithm. Then the IGSA-GPR algorithm is applied to state of health estimation and remaining useful life prognostics of lithium-ion batteries. And IGSA-GPR algorithm is compared with the traditional GPR method based on conjugate gradient algorithm, Genetic Algorithm GPR algorithm and Particle Swarm Optimization GPR algorithm. Experimental results confirm that the proposed method can be effectively applied to lithium-ion batteries state of health estimation and remaining useful life prognostics.