Anomaly Detection Method Study Of Operation State Of Office Building Based On Machine Learning

Building energy consumption accounts for more than 30% of the total energy consumption in China,which is an important part of achieving the goal of carbon peak and carbon neutralization on schedule.The energy use intensity of office buildings is 10 ～20 times that of residential buildings.Improper control strategies or energy use behaviors can lead to a large amount of energy waste,and even serious faults or accidents.Therefore,how to carry out the abnormal detection of building operation state and establish real-time fault alarm and early warning mechanism is of great significance to ensure the energy efficiency,safety and stability of building operation process.Machine learning technology can realize the extraction of mapping relationship of abstract data,effectively overcome the difficulties such as massive data,and is especially suitable for the research of time-series and multi-source building operation state anomaly detection.This paper proposes an abnormal detection method for building energy consumption and complex equipment system operation status of an office building in hot summer and warm winter areas,including the following study contents:(1)To analyze the characteristics of the energy consumption mode of the research object,and describing the building energy consumption supervision platform and air conditioning energy saving control system.The energy consumption modes of lighting sockets and other sub-items are extracted and discussed,and the abnormal types of building operation data are analyzed from two aspects of distortion anomaly and non-distortion anomaly,which lays the foundation for abnormal detection of building operation state.(2)For building energy consumption anomaly detection,a single variable anomaly detection method based on multi-step prediction is proposed,and a hierarchical early warning mechanism is established.Combined with the multi-input-multi-output prediction strategy,a multi-step prediction model based on LSTM is established,and the influence of different super parameters on the prediction accuracy of the model is discussed.The results show that compared with BP neural network and LSSVM,when the prediction step length is 24 hours,the average prediction accuracy is increased by 13.25 % and 4.23 %,respectively.Finally,according to the cumulative probability curve of the number of elements in the energy consumption set predicted outside the threshold interval,the energy consumption classification warning mechanism is established,and the accuracy of anomaly detection reaches 95.41 %.(3)In view of the abnormal detection of the operation state of the complex building equipment system,a multivariate abnormal detection method of the operation state of the complex building equipment system based on quadratic clustering is proposed.In view of the coupling and multi-source problems of operating parameters,K-means algorithm is used to divide I – V operating conditions.Then,the information entropy value of each subsystem under each working condition is calculated,and the SOM algorithm is used to realize the binary classification of abnormal data and normal data.The results show that the average accuracy of intra-class anomaly detection under each working condition is 97.07 %.Finally,the relationship between the intra-class anomaly rate and the proportion of minority classes in the total number and the number of intra-class operating states is analyzed,and the real-time alarm mechanism for the operating state of the multivariable complex equipment system is established.(4)From the practical application,analysis and design of building energy consumption and complex system operating state anomaly detection module.Following the principle of scientificity,compatibility,simplicity and systematicness,the functional requirements of the module are analyzed,including six parts of data selection and sub-item data processing.The development process of the module is expounded,and the design framework of the module is determined,which provides a solution from theoretical research to practical application for abnormal detection of building operation state.