Construction of demand response model of integrated energy system based on machine learning algorithm

: The "multi-energy era", which is complementary to new energy and fossil energy, has arrived. Demand-side management (DSM) has gradually gained worldwide attention because of its advantages of high flexibility and great response potential. However, the appearance of integrated energy system (IES) has broken the existing mode of independent operation of traditional energy systems, and made different forms of multi-energy flow more and more closely coupled. Demand response is the key measure to stimulate demand-side resources to participate in scheduling. IES can integrate various forms of energy, which brings new development to demand response. This paper studies the demand response project of IES, introduces the basic concept and popularization value of integrated demand response, and builds the demand response model of IES based on machine learning algorithm.


Introduction
Machine learning as a major component of modern intelligent technology, has been continuously innovated, improved and developed until now, and it still maintains a vigorous vitality today, setting off a wave of research frenzy of machine learning.It is widely used in almost all popular fields, including unmanned driving, robotics, medical treatment, finance, machinery, energy, network security, etc.It can be said that it is the most revolutionary science and technology in this era [1].Learning, as its name implies, is to let the machine realize people's learning behavior, thus improving the performance of the system through continuous self-learning, and finally giving a relatively correct judgment according to the new conditions [2].It integrates knowledge of many disciplines, including cross probability theory, statistics, cybernetics, algorithm complexity, computational theory in complex environment and other interdisciplinary fields.The core of machine learning is algorithm.In the learning process, the data is iteratively analyzed and calculated by some algorithm, and finally a reasonable judgment and prediction is made on the matter [3].Building a clean and low-carbon energy system and promoting the transformation of the energy industry has become the focus of the competitive development of the energy industry all over the world [4].The report of the 19th National Congress of the Communist Party of China has positioned the future development direction of China's energy, and it is necessary to establish an energy supply system that meets diversified consumption needs and is energy efficient.The IES couples various energy forms such as electric energy, thermal energy, natural gas and hydrogen energy.Different from the traditional energy system, it can not only meet the diversified energy supply demand, but also greatly improve the capacity of renewable energy consumption, with stronger reliability and flexibility [5].The IES can realize multi-energy coupling transformation and utilization, which provides the possibility to improve the utilization rate of renewable energy, enhance the comprehensive utilization rate of energy and solve the energy crisis and environmental pollution problems.The primary goal of the IES is to realize a large amount of renewable energy consumption, improve the comprehensive utilization of energy and ensure the safe, stable, flexible and efficient operation of the system [6].However, there are a lot of uncertainties in the IES, such as the fluctuation of renewable energy output and the fluctuation of different load demand [7].These uncertain factors will bring challenges to the safe and reliable operation of the system.The uncertainty of renewable energy output means that it is influenced by various factors such as meteorology and environment during power generation, which makes it difficult for us to predict the output sequence and amplitude of renewable energy.With the increasing penetration of renewable energy into the grid, this challenge is becoming more and more severe.Therefore, how to solve the uncertainty of supply and demand side in the IES and bring more stable optimization scheme to the system optimization scheduling is still a problem to be solved at present.

Research status of IES and demand response 2.1 Research status of IES
The IES effectively integrates various energy sources such as cold, heat, electricity and natural gas, and can fully tap the potential of each energy system, which has become an important direction of the current energy revolution and a new strategic competition focus of various countries [8].The U.S. Department of Energy put forward a comprehensive energy system development plan, aiming at improving the utilization rate of renewable energy, promoting the popularization and application of combined cooling, heating and power technology, and improving the reliability of social energy supply [9].According to the Energy Independence and Security Act of the United States, comprehensive energy planning must be placed in an important position in energy development, and a large amount of funds will be invested in the special research and implementation of comprehensive energy system planning in the later period.Countries all over the world have also launched related research and system construction projects.In the seventh R&D framework of the European Union, the requirements for the security and reliability of high-quality energy networks are put forward, so as to achieve further energy conservation in various industries [10].The pilot project is based on heating technologies such as cogeneration, and a multienergy energy system with high permeability and renewable energy is built.The ELECTRA demonstration project of the European Union is based on modern power communication technology and market mechanism, aiming to achieve the local consumption of large-scale renewable energy and the stable operation of large power grid.Germany has always listed energy transformation as one of the major national projects.Aiming at the typical energy problems in China, the German demonstration project in langenfeld, based on modern communication, automatic control and measurement technologies, is committed to building wireless city and minimizing the energy utilization of users [11].Manchester demonstration project in England has set up an interactive platform between IES and users, which can solve the problems of cascade utilization of energy, comprehensive coordination among energy sources and demand side management of various energy sources.

Present situation of demand response research
Traditional power demand response is simply the load transfer or energy consumption reduction in time scale, which greatly changes users' energy consumption habits, with low enthusiasm of users and limited response potential.In order to cope with large-scale distributed renewable energy access, people also need a long-term sustainable solution.With the continuous enrichment of research results and the gradual landing of key technologies, the researcher put forward the concept of demand response for the first time in 2015, and defined demand response as the derivative and expansion of traditional electric power on the demand response side in the integrated energy network.The multi-energy intelligent management system, which combines energy conversion with time transfer, enriches the response mode of demand side, strengthens the flexibility of demand side response, and is an important embodiment of the convergence of energy flow, information flow and value flow on the user side.At present, there are many research achievements on traditional demand response strategies at home and abroad.Due to the early start of foreign research, a relatively formed market system has been put into operation.At present, many countries have carried out research on demand response, Europe and North America are at the forefront of this field, and small-scale experiments have been carried out at the community level, and certain research results have been achieved in response types, basic devices, interactive ways and user participation forms." To sum up, at present, the research on demand response is still in the initial stage, lacking perfect interaction mechanism, and the transfer relationship between information flow and energy flow is not clear; In the process of interaction, the end user's adjustable ability is difficult to describe, and the superior network can't accurately issue the control instruction; In addition, the complementary characteristics of cold, heat, electricity and other energy sources in time and space are less considered, which leads to the failure to further tap the response potential of users.Therefore, it is an important exploration direction of demand response in the future to clarify the interaction mode and regulation strategy of demand response, to model demand response resources in a refined way and to quantify the regulation ability, which is of great significance for demand response to really participate in the optimal scheduling of IES.

Build the demand response model of IES based on machine learning algorithm 3.1 Model Construction Based on BP Neural
Network Before constructing the demand response model of IES based on machine learning algorithm, before choosing machine learning methods, the first thing to do is to choose the machine learning mode according to the characteristics of the subject.In this paper, the BP neural network is chosen to build and optimize the model, because of the following reasons: the neural network can use a simple single hidden layer network to update the weights and thresholds by iteration to fit arbitrary functions; Neural network can deal with high-dimensional feature vectors and extract features.Neural network has learning ability similar to human brain, and the trained network evaluation model can be generalized to untrained samples, which is suitable for evaluation research.The neural network is easy to expand.When the amount of information to be stored exceeds the capacity, the storage capacity of the network can be expanded by increasing the network scale.It is self-adaptive and can dynamically adapt to all kinds of samples as the training samples change.First, the determination of network layers.On this issue, a famous theorem gives a judgment, that is, a three-layer neural network with only input layer, output layer and hidden layer can approach any nonlinear system to the maximum extent, as long as the number of nodes in the hidden layer is not limited.Therefore, in necessity, the three-layer network model is enough.In addition, if the number of layers of the network is increased, the accuracy of the training results will be improved to some extent, but at the same time, the complexity of the system will be greatly increased, and the training efficiency of the neural network will be reduced.The result is probably not worth the candle.Therefore, the basic three-layer network structure is selected for the number of BP network layers in this paper.In the selection of the number of neurons in each layer, the number of nodes in the input layer is determined to be 25 by the above research on the demand response of the IES; According to the purpose and demand of this topic, that is, the performance of demand response model of IES, the number of nodes in the output layer is determined to be 4.The flow chart of BP neural network training is shown in Figure 1.

System structure design of demand
response model of IES Firstly, the purpose and core of the demand response model of IES are defined, that is, the evaluation of the demand response model of IES is realized.In order to achieve this goal, the required structural modules mainly include: data management module, improved BP neural network configuration module and evaluation result and analysis module.The following is a description of the specific contents and functions of each main module.The first block is the data management module, which mainly manages information data, including the input of collected data, the basic operation and processing of the input data, the viewing and deletion of the input data, etc.This module is the premise and foundation for the follow-up module or the whole evaluation model to run smoothly.The second block is the improved BP neural network configuration module, which is the key to realize the system goal.This module embeds the improved BP neural network model which has been constructed before, and users can use this module to obtain the evaluation results.This module is where the evaluation prototype system of this system is continuously maintained and updated.There will definitely be a better evaluation model in the future, and there will definitely be more optimized research results in the current evaluation index system, weights and evaluation methods.At that time, based on the latest research, the evaluation model and related indexes and coefficients can be optimized to update this system.The third block is the evaluation result and analysis module, which is the purpose module of the evaluation prototype system.It is used to display the evaluation results, thus providing analysis materials and foundation for future researchers to improve and optimize the energy system demand response model, and promoting the development of the research field of energy system demand response.The schematic diagram of the functional modules of the IES demand response model is shown in Figure 2.

Conclusions
In order to improve the consumption of renewable energy, maximize the utilization of comprehensive energy, solve the uncertainty of demand response, and realize the economic and stable operation of comprehensive energy system.Based on the machine learning method, this paper focuses on the demand response model of IES, and studies the evaluation system of demand response model of IES.On the basis of previous studies, from understanding the definition and connotation of IES demand response model, an evaluation index system of IES demand response model is obtained.By studying the IES demand response model and its evaluation methods in related fields, the IES demand response model is combined with the development status and bottleneck of IES, and the shortcomings and shortcomings of traditional evaluation methods are realized.For the first time, the demand response model of IES is introduced into several mature and widely used machine learning evaluation methods, and then the improved optimization strategy is added to construct the evaluation model of demand response model of IES, which is relative to the objective subject of demand response model of IES.In a sense, it adds a research attempt of machine learning method evaluation to the development of demand response model of IES.

Figure 1
Figure 1 BP neural network training flow chart

Figure 2
Figure 2 Schematic diagram of functional modules