Household Electricity Demand Response And Multi-objective Optimization Based On Nonintrusive Load Monitoring

As an important part of power demand-side management,residents have strong load flexibility and high demand response potential,so it is important to explore the behavior of residential power consumption in the context of the current "double carbon" of peaking carbon dioxide emissions and carbon neutrality.However,the large variability of residential electricity power consumption behavior and the high diversity of decision-makers make it very difficult to explore its demand response potential.With the continuous development of smart grid technology,a large amount of data has been accumulated,especially the new generation of smart meters that can collect power load data at a frequency of a few seconds to tens of seconds.These data reflect the real power consumption behavior of users and hold great research value.However,residential electricity power data has typical big data features such as large scale,high dimensionality,high timeliness and low-value density,and it is difficult to extract features from them by traditional statistical methods.Therefore,how to use the new generation of data mining techniques to discover the patterns of residential power consumption behavior from smart meter data,to make efficient and energy-saving electricity consumption plans,and to improve the level of household demand response is the main problem of this paper.The paper focuses on this problem from the following aspects:Firstly,the nonintrusive load monitoring(NILM)technology is adopted to decompose the load of each electrical appliance from the total load of residential smart meters,and fine-grained characteristics of power consumption behavior are obtained.Traditional research on residents' power consumption behavior is mostly based on social surveys and the total household load.The data scale is small and the granularity is too coarse,so it is difficult to find the deep behavior patterns.In this paper,we propose a deep learning model based on attention mechanism for NILM,which can more accurately decompose the power of specific appliances from the total load by introducing the attention mechanism to enhance the learning ability of neural networks,and the experiments on public datasets demonstrate that the proposed model is ahead of current mainstream models.In order to solve the problem of deep learning models relying on large-scale training data,this paper conducts a study of transfer learning based on the proposed model.By adding multiple self-attention layers,the model is able to learn the abstract features of load signals,which can be transferred across domains to other data without training from the beginning,which can reduce the model training cost and solve the model generalization problem.Secondly,the characteristics of residential electricity consumption based on NILM are studied,and the short-term forecasting models are constructed.Load decomposition can obtain fine-grained features at the appliance level: one is time-level features,such as appliance turn-on time,running time,start-up frequency and active power,etc.The other is interrelated features between applinaces,such as the sequence of appliance turn-on,and the habits of appliances used with each other.Translating these features into suitable mathematical expressions is a prerequisite for further analysis and optimization,as well as the basis for demand prediction.Short-term load forecasting is an important decision-making basis for electric power companies to formulate residents' demand response plans.Then,a demand response-based residential electricity consumption optimization model is developed.Demand response is the use of price and other stimulus signals to encourage customers to actively adjust their electricity consumption behavior to cooperate with the grid to complete load regulation,but customers often lack understanding of incentive policies and their own electricity consumption habits,resulting in a lack of demand response.In this paper,electricity consumption patterns are extracted from historical electricity consumption data and consumption behavior is represented using the appliance state matrix.A multi-objective optimization model is established considering both comfort and expenses,and a particle swarm algorithm is used to solve for the optimal time to turn on household appliances.The usability of the proposed optimization model is demonstrated by experiments on real data sets.The optimization model established in this paper can be applied to the home energy management system to improve the automation level of demand response.Finally,an intelligent recommendation model based on collaborative filtering for residential electricity consumption plans is proposed to recommend the energy saving experience of high demand response users to low demand response users.The traditional optimization model starts from individual household power consumption data without considering the energy saving experience of other households,and previous studies have demonstrated that sharing energy saving knowledge has a significant promotion effect on residents' energy saving behavior.In this paper,we first distinguish the level of users' demand response by clustering algorithm,extract the usage plans of shiftable appliances from high demand response users as the energy saving solutions to be recommended,and then use the similarity of the usage patterns of unshiftable devices to identify high demand response users with similar electricity consumption habits as the target users,and recommend the energy saving plans to the target users based on the user similarity and the ratings of high demand response users.To address the shortcomings of the classical collaborative filtering algorithm,the paper proposes to use a neural network collaborative filtering algorithm to enhance the adaptability and accuracy of the recommendation model.