关键词: 玻璃深加工, 光储一体化, 集成调度, 遗传算法, 深度学习

Abstract: With the growing global emphasis on environmental protection and sustainable development, the increasing pressure on companies to reduce energy cost and minimize carbon emissions is increasing. In the manufacturing sector, an energy-intensive industry, traditional fossil fuel energy consumption not only leads to high energy cost, but also has a serious impact on the environment. As a result, more and more companies are looking to renewable energy sources, particularly photovoltaic energy, to supply energy for their manufacturing processes. This transition not only helps to reduce long-term energy costs, but also helps companies to gain an edge in environmental regulations and market competition. In this paper, a photovoltaic (PV) glass deep-processing line shop using a photovoltaic storage energy supply system is used as a research object. It aims to explore how production scheduling can be optimized under photovoltaic energy conditions to reduce total cost and improve productivity.
   In this study, a three-stage mixed integer planning mathematical model is constructed by considering the characteristics of the photovoltaic glass deep-processing line. The model includes three stages: edging, coating and tempering. To effectively reduce the penalty cost of delayed delivery and the energy cost under time-sharing tariffs, we design a meta-heuristic algorithmic framework that combines the advantages of deep learning to optimize the production sequencing of glass deep-processing as well as the storage and deployment of photovoltaic power generation. Among them, the Monte Carlo simulation is used to obtain training sample data for the uncertainty of PV power generation. And the fast approximate evaluation of feasible solutions is achieved by fitting regression through feature engineering and convolutional neural network. The accuracies of the runtime and solutions of different algorithms are compared and analyzed by simulating the data of PV power generation and PV glass deep processing. Sensitivity analysis is then used to explore the impact of different PV power generation scenarios on production cost.
   It is shown that the integrated scheduling scheme incorporating photovoltaic (PV) energy storage systems performs superiorly in terms of the accuracy of runtime solutions. Through numerical experiments, it is verified that the designed convolutional neural network can quickly and accurately evaluate the feasible solutions. Comparison with commonly used heuristic algorithms verifies the effectiveness of this paper’s algorithm in terms of runtime and solution accuracy. Meanwhile, the integrated scheduling scheme in this study also shows significant advantages in terms of solution quality compared with the traditional independent running strategy. The study also reveals the trade-offs between the two core objectives under the integrated scheduling framework, and further compares the production costs in different photovoltaic power generation scenarios, which provides guidance at the practical operational level for the promotion and application of renewable energy in glass manufacturing.
   Further research can consider the real-time adjustment of power control strategies, formulate the real-time control problem as a Markov decision process, and use approximate dynamic methods such as reinforcement learning to optimize the solution. In addition, the optimal size allocation of PV panels and storage plants can be explored from a strategic level perspective. The NPV approach is utilized to measure the overall return of a company constructing a photovoltaic storage and energy supply system from the perspective of the full life cycle of a PV project.

Key words: glass deep-processing, photovoltaic generation and storage, integrated scheduling, genetic algorithm, deep learning