Research And Implementation Of Remaining Battery Power Detection System Based On ANFIS

This paper obtained the sponsorship of “a system of monitoring SOC(state of charge) for batteries” from China Aerospace Science & Industry Corporation.A model based on adaptive neural network for predicting is studied and established. The system can be applied to systems that need accurate control of battery power.In the preliminary section, the model of Adaptive-Network-based Fuzzy Inference System(ANFIS) for remaining power prediction for batteries is establish. Initially, the theories of ANFIS is introduced, the reason of choosing the algorithm is presented. Then, the analysis of main factors that related to battery SOC is conduct, aiming to ascertain the input parameters for the model, which followed with fuzzy set partitioning and subordinating degree function calculation for each parameter. Finally, we work out the structure of ANFIS and set up the BP Neural Network Learning Algorithm.In the experiment section, tests have been conducted to modify the model relying on the test data. The simulation program is built together with a detailed execution plan. We generate several initiations to train the ANFIS model, and comparing the values between ANFIS model and the real experimental values, which reveal that the model can mimic the ANFIS well, especially the one obtain subtractive clustering method. It has the least number of nodes after training. Later on, we adjust the perimeters and do contrast experiment again.In the system structure section, the design of hardware and software is executed. As for hardware, modules for temperature measure, internal resistance measure, voltage measure, current measure, and communication are integrated in a singlechip in the measuring unit. Here, the difficulty lies in measure of battery internal resistance, an AC injection method is obtained in this paper to deal with it. As for the software, the system is a combination of a display control unit and a measure unit, for which the former mainly conduct data collection and transmission and the latter runs the ANFIS algorithm to provide results for user. Adding that the display control unit needs to carry out the design of communication protocol stack and low-power dissipation control while the measure unit have to manage user UI design. Further, a MODBUS-RTU based communication protocol is formulated in this section.The actual application test of the system is execute and present in this section. According to the result, measurement unit measured data errors within 5% and the maximum error of battery SOC predicted values is 0.0046, which satisfy the requirement of application. However, the system has a limited speed and consume too much resources of display and control unit, which leads to the decline of performance in the display and control unit. Due to this, two improved methods are proposed and a deliberate predict of the future of the problem scope.