Abstract: This paper proposes a new method for motor fault prediction based on fuzzy neural network. This method combines time series and fuzzy neural network, and also introduces time difference method to predict the motor state, thereby improving the prediction accuracy and reducing the system error. Simulation results show that the error of this prediction method is significantly smaller, making it a practical prediction method. Keywords: Fuzzy neural network; Time series; Fault prediction 1 Introduction System state prediction is an essential part of fault diagnosis and one of the important goals of diagnostic technology. Prediction based on fuzzy neural network is a nonparametric model prediction. In fuzzy neural network prediction, existing methods typically involve building a model once using a large amount of existing observation data (sample data), and then not learning again during prediction, meaning the network parameters remain unchanged. Over time, this model built using historical data cannot fully reflect the recent and present characteristics of the time series, requiring continuous adjustment of the network model parameters as new data accumulates to improve the model. Therefore, this paper proposes a new fuzzy neural network algorithm, which introduces a time difference method based on the original algorithm. This method can, to some extent, continuously adjust the model parameters in real time based on observation data and prediction results, making the model as perfect as possible and thus improving prediction accuracy. 2. Fuzzy Neural Network A fuzzy neural network is a combination of a fuzzy system and a neural network. It implements fuzzy logic through the neural network and utilizes the self-learning ability of the neural network to dynamically adjust the membership function and optimize control rules online. This fusion compensates for the shortcomings of neural networks in fuzzy data processing and the learning deficiencies of pure fuzzy systems. We use a multi-layer feedforward fuzzy neural network with a serial structure, as shown in Figure 1. This model has four layers: an input layer, a membership function generation layer (fuzzification layer), an inference layer, and an anti-fuzzification layer. Twelve adjacent peak-to-peak data points were taken as a training sample, and the 13th data point was used as the training target. A total of 10 sets were used to train the network. The closest data was used as the training sample, and the training sample was continuously updated as the system sampled, so as to obtain more accurate prediction results using the closest data. Simulation was performed using the neural network toolbox in Matlab. Through testing, the input was set to 12, the maximum number of epochs was 50, and the expected minimum error was 0.001. The prediction process is shown in Figure 4, and the prediction analysis is shown in Table 1. Compared with the actual results, the maximum absolute error of the prediction results is 0.06. Considering the measurement error of the field sensors, it can be considered that these data basically meet the prediction requirements for the normal operation of the motor. 5. Conclusion This paper combines fuzzy neural networks with time series analysis and introduces the time difference method to establish a new prediction model. Based on the peak-to-peak value of the vibration voltage during motor operation, the operating state of the motor was predicted. The detection results show that the prediction model has high prediction accuracy and small error, making it a practical and feasible method. References [1] Niu Yongsheng, Zhao Xinmin, Sun Jinwei. A new method for sensor fault diagnosis using time series predictor based on neural network [J]. Journal of Instrumentation. 1998, 19 (4): 383-388. [2] Zhao Xiang, Xiao Deyun. Fault diagnosis of nonlinear system based on multi-step time series prediction of neural network [J]. Control Theory and Applications. 2000, 17 (6): 803-808. [3] Hu Shousong, Zhang Zhengdao. Fault prediction of nonlinear time series based on neural network [J]. Acta Automatica Sinica. 2007, 33 (7): 744-748. [4] Lin CT, Lee CC GN Neuranl network based on logic control and decision system [J]. IEEE Trans. On Computers, 1991, 40 (12): 1320 [5] Chen Guo. Automatic extraction technology of dynamic chaotic features of nonlinear time series [J]. Journal of Aerospace Power. 2007, 22 (1): 1-6 [6] Shi Huiyu, Meng Fanrong. Application of BP neural network in prediction of coal mine monitoring data [J]. Microcomputer Information, 2008, 24 (71): 26-27 [7] Yang Lu, Huang Tiyun. A time series adaptive modeling and prediction method based on neural network [J]. Decision and Decision Support Systems, 1996, 6 (2): 69-74.