Distributive Prediction Of Energy Consumption Among Operations In Process Industries

In recent years, the disordered energy consumption has been making the energy supply increasingly tense. China’s steel industries are high energy-consuming ones, so they are facing a very severe situation. Improving energy utilization and reducing energy consumption can not only reduce production costs, but also increase the efficiency of enterprises and improve the competitiveness of enterprises. In this paper, the energy media consumption problem of process industry is used as the research background. The problem has been modeled as a multi-stage optimization using approximate dynamic programming method. The main contents are as follows:(1) Current energy management of iron and steel enterprises focus on the energy consumption forecasts based on demand side management, and paid more attention to the total energy consumption. This kind of bundled energy statistical methods can not make operators well understand the specific distribution for each media. In this paper, predictions of energy media consumption among production operations are carried out, which can be a good balance of energy media consumption, regeneration, loss and diffuse.(2) For solving distributive prediction of energy consumption problem, an approximate dynamic programming algorithm based on neural network is designed. Approximation dynamic programming is proposed in recent years, which is the effective tool to solve the large-scale stochastic optimization problem. The core of ADP is the value function approximation. Among the four non-parametric methods for value function approximation:K neighborhood, kernel regression, local polynomial regression and neural network, neural network has the highest accuracy and stability. The data stability and prediction accuracy have been improved using the approximate dynamic programming algorithm based on the neural network(3) Based on the investigation, theoretical analysis and algorithm results, the balance systems of energy media consumption distributive prediction are designed and developed. The modular design, each module function, display interface, etc. are introduced in detail.(4) Steel is a typical process industry, which covers multiple kinds of energy medium and consumes large quantity of energy. In order to verify the proposed energy medium distribution prediction algorithms in a wide range of process industry, the energy medium consumption research is extended to beer company, and the energy medium forecasting system is developed for a beer company.