Research On Consumption Prediction Model Of Fiber Refining Procession Based On Neural Network

Fiber refining process is one of the most crucial process of fiberboard production, and is also the large energy consumption step in the production of fiberboard. The energy consuming of fiber refining process accounts for about 40%of the producing total energy consumption. Quality of fiber separation has important influence on fiberboard properties, thereby affecting the quality of products and profits. The main purpose is to study and establish the prediction model of energy consumption in the fiber refining process with the introduction of BP neural network. Explore the production process parameters on fiber refining energy consumption and quality of fiber. According to the prediction model can be, in the fiber production process, predicting fiber refining energy consumption and quality of fiber to seek for the parameters balance point of low power consumption and high quality of fiber in the fiber refining process.Combined with the analysis results of production process in the fiber refining process, determined the factors of energy consumption and fiber quality. Using of gray relative analysis method to determine factors associate degree of fiber refining energy consumption and fiber quality. According to the grey relational degree determine the training data is cooking pressure, bark content, refining pressure, discharge screw speed, wood moisture content, refining energy consumption in the energy consumption predicting model. Bark content, cooking pressure, discharge valve opening degree, discharging screw speed, refining pressure, fiber screening values(20 to 120 mesh) is the training data of the fiber quality forecasting model, pursuant to proposed design hot mill consumption of fiber and fiber quality prediction model. Accordingly proposed design plan of the fiber refining energy consumption and fiber quality prediction model.Using BP neural network method to establish fiber refining energy consumption and fiber quality prediction model. In the modeling process, normalization training data, to determine the structural parameters of the model, the network layer is 3, hidden layer nodes is 8, hidden layer transfer function is’tansig’, the transfer function for the output layer is’logsig’, training mode is batch learning, training function is’train’, network algorithms is’traingdm’. Structure training samples for prediction model, and training them, obtain prediction models, the maximum relative error is 4.26%. To verify the reliability of prediction model, the prediction results show the maximum relative error is 3.33%.Base on the reliability of prediction model, combined with a single variable experiment, respectively predicting the influence trends of the main factors on the fiber refining energy consumption and fiber quality. And achieving the predicted values of refining energy consumption and fiber screening values(20 to 120 mesh) under the condition of multiple factors change. Using the prediction model to predict the product data, compared predicted results with actual values, the average relative error of refining energy consumption is 1.27%, and the average relative error of fiber screening values(20 to 120 mesh) is 1.99%. Bark content as the research object, study on its value range influence on fiber refining energy consumption and fiber quality. Through experiments achieve trends of fiber refining energy consumption and fiber quality with bark content changes. Comprehensive analysis of prediction results and the cost of raw materials, under the conditions, the best bark content is 13.2%. Under this case, it can meet production requirements and the cost of production is lowest.Accurately predicting energy consumption and fiber quality in the fiber refining process have important theoretical and practical significance in the aspect of reducing energy consumption in the fiberboard production, improving quality and optimizing the production parameter and achieving energy-saving operation of fiberboard production systems.