Research On Air Conditioning Test Task Scheduling Problem Based On Energy Consumption Prediction

Numerous enthalpy testing should be carried out to ensure that a new airconditioning(AC)prototype can perform well under various conditions before the new product is put into mass production.Creating such a testing condition consumes high electricity usage.Take our studied company as an example,as a leading company in China’s AC testing industry,every year it consumes 15 million kwh of power,equivalently a 12 million RMB electricity bill.It will be of great economic and social value if the test plan can be reasonably arranged by means of operational optimization to reduce power consumption.Through the investigation,we found several common problems in the industry were as follows.First,the length of time to create a required environmental condition depends largely on the sequence of test tasks,which makes it difficult to determine the task testing time.And it causes the second common problem,it is hard for the companies to take full advantage of time of use policies to save money on electricity bills.Third,according to the developed test plan,it is difficult to give consideration to multiple criteria such as economy and fairness at the same time in the staff scheduling process.Adding to the complexity is the varying number of people required for different stages of testing.These common problems exist not only in enthalpy difference experimental enterprises,but also in other manufacturing enterprises.The research on this problem has great research and application value.The intelligent systems of enterprises collect a large number of production process data,which provides an opportunity to effectively solve the stated three problems.First of all,through big data mining,the testing time can be better predicted,and the test time and power consumption under different task sequences can be obtained.Therefore,task scheduling can be better arranged based on(time of used)pricing.After that,according to the reasonable task scheduling results,better personnel scheduling can be achieved.Following this idea,this paper carries out the following work.Firstly,a variety of data collected by the intelligent sensors in the enterprise laboratory were cleaned,explored,and analyzed with significance analysis.Then variables related to the machine adjustment time were screened out.Aiming at the situation that the machine adjusting time is correlated and cannot be described by the conventional distribution function,a prediction model based on random forest regression is built to predict it,which lays a foundation for solving the task scheduling problem in the future.Second,in view of the shortcomings of the separated data mining and mathematical programming stages,a hybrid algorithm based on random forest and genetic algorithm(RF-GA)is proposed,which can quickly predict the corresponding time and power consumption when the task sequence changes.In order to meet the high requirements of quality,convergence speed and robustness of solving algorithm for practical test arrangement,the task scheduling problem was transformed into a kind of generalized traveling salesman problem by extracting the main characteristic information of air conditioning test task,and a kind of Inver-over operator genetic algorithm suitable for the characteristics of this problem was selected and applied to the hybrid algorithm.Finally,a large number of simulation experiments are designed to verify effectiveness and efficiency of the hybrid algorithm.Thirdly,the idea of using surrogate model to fit the feasible solution is proposed to further optimize the hybrid algorithm.Broad Learning System is used as the surrogate model in this paper because of its high computational efficiency.A series of optimal solutions obtained by the hybrid algorithm were used as the training data of the surrogate model,and then the gradient descent method was used to find the optimal solution.The experimental results show that this method is practiced.Fourthly,based on the results of task scheduling,the problem of staff scheduling is studied.According to the economic and equity criteria,such as overtime payment and limited continuous working time,a staff scheduling model is obtained,and then the genetic algorithm is used to solve the staff scheduling problem.In the meantime,this paper discusses the difficulties and opportunities of jointly and phased solving the task and staff scheduling problems.Finally,considering the cost of task and staff scheduling,the method proposed in this paper is applied to the actual test of enterprises.The results show that the method can significantly reduces the energy consumption and total expenditure cost of enterprises,shortens the working time by about 25%,improves the energy saving efficiency by 18%,and saves the expenditure of enterprises by about 180,000RMB/month,bringing good economic and social benefits.