Modeling of Soft Sensor Based on DBN-ELM and Its Application in Measurement of Nutrient Solution Composition for Soilless Culture

Abstract

At present, the detection of important components of nutrient solution in soilless culture is high cost, difficult and low precision. A soft measurement method of nutrient solution component based on deep belief network and extreme learning machine (DBN-ELM) is proposed. The component concentration in nutrient solution was selected as the dominant variable, and the variables that were easy to be measured and correlated with the ion concentration were the auxiliary variables, including PH value, conductivity, nutrient solution circulation speed and temperature. The deep belief network is used to extract the features of the auxiliary variables, then the extracted features are input into the ultimate learning machine for training, and the soft measurement model is obtained. Finally, the data of soil - free tomato culture nutrient solution was used to verify the experiment. The results show that this method has higher comprehensive measurement accuracy than the method using the extreme learning machine or the least square method, and is of great significance for improving the yield and quality of the soilless cultivated crops.