Research On Energy Consumption Prediction And Dynamic Allocation In Iron And Steel Industry

China is a country with large energy consumption, which energy consumption per unit product is averagely higher47%than the advanced world level. And the proportion of Iron&Steel industry energy consumption is about10%in national total energy consumption, and its energy cost accounts for about one-third of steel production cost. Therefore, the Iron and Steel industry of China can be characterized by intensive energy consumption and low energy efficiency. Our country steel enterprises have huge energy saving potential.Based on typical energy consumption process of steel enterprises, this thesis predicts the energy consumption so as to manage scientifically the energy consumption in a future horizon. Based on the energy consumption prediction, this thesis researches energy dynamic allocation problem. Aiming at realization of energy optimal allocation, in this problem, the outermost layer of the framework is the energy dynamic allocation, in which the energy demands are got through the inner embedded prediction. The main contributions of this thesis are as follows:1) Based on the characteristic description of steel enterprise energy consumption prediction, this thesis modeled MDP (Markov Decision Processes, MDP). There are many factors that impact energy consumption such as yield and temperature, but they all are random and require online and real-time prediction. Although, there lacks the knowledge of energy consumption mechanism, we have a lot of history data. This thesis projects energy demand prediction problems into a stochastic dynamic programming framework and models a stochastic MDP, which effectively deals with uncertainty and be suitable for online and real-time energy consumption prediction.2) Considering the characteristics of MDP prediction model for energy prediction in iron and steel industry, this thesis develop ADP (Approximate dynamic programming, ADP) algorithm. Because of the randomness of the problem and lack of state transition probability, we cannot get optimal solution using dynamic programming algorithm. And, discrete degree of the system states will make problems produce huge scale, so dynamic programming will face a dimension disaster. Based on advantages of ADP, model-free and data-driven, high adaptability to the environment, low-cost and solving dimension disaster, this thesis projects prediction problem into a stochastic dynamic programming framework and solves it using ADP, which effectively deals with uncertainty. ADP algorithm policy is given by using Robbins-Monro (RM) stochastic approximation approach3) The energy allocation plan with the dynamic demand is put forward. Traditional energy allocation plans all assume that energy demand is constant, however, according to the new problems extracted from practical production, energy demand is changing with time. Under that circumstance, this thesis proposes energy allocation plan model with the dynamic demand, and the model’s objective function is to minimize the total cost of production, and energy demands update dynamically.4) According to the characteristics of the model of energy allocation model with the dynamic demand, based on energy demand prediction, this thesis develops algorithm to solve the model. The outside loop of the model is dynamic programming, each node of dynamic programming is linear programming model, and each energy demand of the linear programming model is update dynamically. The algorithm developed in this thesis has three layers:the outside loop is solved using the dynamic programming algorithm, each node of dynamic programming is solved using CPLEX, and each energy demand of the linear programming model is solved using ADP algorithm.5) According to the typical process of iron and steel industry, this thesis designs and implements decision support system with embedded prediction model and above algorithms in the core business logic module so as to effectively realize the energy prediction and guide practical production and help the operators to overcome the difficulties of manual operation and greatly improve the enterprise information level and office efficiency.