Study on Ammonia Concentration Prediction Model of Pigsty Based on LTSM Neural Network

In the large-scale farming, the environment of Pigsty has a direct impact on the health of pig and production capacity. Pigsty major environmental factors such as wind speed, temperature, humidity and ammonia concentration data were collected for one hundred and sixty-seven consecutive days, the acquisition interval was set to 60 seconds. Since the predicted ammonia concentration is 1 hour later, 3975 groups of hourly environmental data were obtained from the original data through calculation. Then, The LTSM neural network model was built to predict the Ammonia concentration. According to the time sequence, the data collected in the first 120 days were taken as training samples, and the rest of the data collected were test samples. It is shown in simulation experiment that network reaches the target error after 40 steps with each batch size of 72, the model has the characteristics of fast network convergence and high efficiency by using adam optimizer, the root mean square error between the actual environmental quality of 2880 groups of test data and the network prediction value is only 1.6%, the accuracy and timeliness of the ammonia concentration prediction of Pigsty is greatly improved. The ammonia concentration prediction model established in the paper can provide support for the Pigsty environment early warning and control.