Diabetes is one of the most common diseases that endanger human health. Effective management and control of blood sugar is very important for patients. Traditional blood glucose prediction models are mostly single deep learning models， which have the defect of insufficient accuracy or low efficiency， restricting their effect in practical application. Therefore， a hybrid neural network model based on sparrow search is proposed and applied to blood glucose prediction. The proposed model combines a temporal convolutional network （TCN） and a gated recurrent unit （GRU）， and it is a sequential neural network trained in an end-to-end manner to predict blood glucose based on a patient’s blood glucose level history. In order to ensure the generalization ability of the model， data sets from two different sources are used for validation. Firstly， the feature sampling frequency of multi-source timing monitoring data is set at a time interval of 5 min， the data is smooth-processed and standardized， and the timing pattern and dependency characteristics are captured by TCN. Then， by constructing a GRU model based on the attention mechanism （GRU-Attention）， features are further extracted and modeled. Finally， the sparrow search algorithm is used to optimize the hyperparameters of the TCN and GRU-attention models to realize the blood glucose prediction model. To prove the validity of the proposed model， its prediction results are compared with those of other models， including LSTM， ARIMA， RNN， etc. The results show that the proposed TCN and GRU-Attention models based on the sparrow search algorithm perform well in the task of predicting blood glucose value. The root mean square error （RMSE） and mean absolute error （MAE） of the two datasets are 0.552 and 0.402， 0.531 and 0.388， respectively， which are all better than other models.