Development of an intelligent control system for the process of preparation and water transfer in the cooling circuit of an ammonia station

Introduction. In the modern socio-economic and geopolitical development of Russia, the development of industry comes to the fore. Among the many industries, ammonia stations play the most important role. The main regularities of the process of pumping and preparing water. The process consists of six stages, this article discusses the automation of stages 1 and 2: for water treatment and pumping it out with pumps H1 and H2 from the tank P2. Products in the form of purified water are the most important criteria for subsequent production at an ammonia plant, therefore, increased requirements are imposed on the quality of finished products, including the quality of purification of the water used with the help of nanofilters. The required quality cannot be achieved without control the process in an automated mode. Development of a neural network. To control the converters frequency values during the preparation and pumping of water, an artificial neural network must be used. Its development was carried out in the Matlab environment in the Neural Network Toolbox package, input and output data were defined for this, data processing and preparation were performed, as well as the choice of the type and architecture of the neural network. The architecture of the Layer Recurrent neural network, the process of its construction and training in Matlab is described. Testing of neural networks. During testing of the Layer Recurrent network for the degree of their training, the smallest error was obtained for 30 neurons in the hidden layer. The proximity to the set values indicates the applicability of the network for controlling the parameters of frequency converters. Development of the neural network controller model in the Simulink package. The simulation of the control system in the Simulink package using a neural network controller with the Layer Recurrent architecture is performed. Checking the frequencies of the frequency converters H1 and H2 in Simulink for the level parameters in the tanks and in the tank LT1_вх, LT2_вх, LT3_вх showed that the object model works correctly, thus, the simulation of the neural network showed that the training was successful. Conclusion. As a result of the conducted research, an artificial neural network was developed to control the process of preparing and pumping water in the Matlab environment and a simulation of a neural network in the Simulink package.