CN-121980461-A - Industrial energy-saving monitoring method based on Copula graph model

CN121980461ACN 121980461 ACN121980461 ACN 121980461ACN-121980461-A

Abstract

The invention discloses an industrial energy-saving monitoring method based on a Copula graph model, which relates to the technical field of industrial energy efficiency monitoring and energy-saving monitoring and comprises the steps of collecting multiple types of energy consumption original data and preprocessing to obtain multiple types of energy consumption time sequence; the method comprises the steps of fitting an edge probability distribution function to various energy time sequence by a non-parameter density estimation method, obtaining a corresponding uniform distribution sequence through probability integral transformation, constructing a Copula graph model by taking the uniform distribution sequence as a node variable and adopting a graph structure learning algorithm, converting real-time energy consumption data in a period to be monitored into real-time uniform distribution values, inputting the real-time uniform distribution values into the Copula graph model, calculating a joint probability density value, comparing the joint probability density value with a preset energy-saving abnormal threshold value, positioning a dominant energy type causing energy consumption abnormality based on the contribution degree of each node variable, and outputting energy-saving monitoring early warning information. The invention can improve the accuracy and the interpretability of industrial energy-saving supervision.

Inventors

CHEN JINHUAN
SHEN WENJIE
XING JING
Ren Qinyang
Ruan Xiuling
ZHOU RUN
ZHOU YONGPENG

Assignees

广东资环新能源有限公司

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. An industrial energy-saving monitoring method based on a Copula graph model is characterized by comprising the following steps of, Acquiring multiple types of energy consumption original data through an intelligent monitoring terminal of the Internet of things deployed on an industrial production site, and preprocessing the multiple types of energy consumption original data to obtain multiple types of energy consumption time sequence sequences; fitting the corresponding edge probability distribution function to the multiple classes of energy time sequence by adopting a non-parameter density estimation method, and converting each class of energy time sequence into a corresponding uniform distribution sequence through probability integral conversion to obtain multiple classes of uniform distribution sequences; Constructing a topological structure of the Copula graph model by taking the multi-class uniformly distributed sequences as node variables and adopting a graph structure learning algorithm to obtain a Copula graph model after training; converting the real-time energy consumption data acquired in the period to be monitored into real-time uniform distribution values, inputting the real-time uniform distribution values into a Copula graph model after training, and calculating a joint probability density value; And comparing the joint probability density value with a preset energy-saving abnormal threshold value, positioning a dominant energy source type causing abnormal energy consumption according to the contribution degree of each node variable in the Copula graph model to the joint probability density value, and outputting energy-saving monitoring and early warning information.
2. The industrial energy-saving monitoring method based on the Copula graph model of claim 1, wherein the energy-saving monitoring and early warning information output method is that, Normalizing the local abnormal contribution degree of each node variable in the Copula graph model to obtain the abnormal contribution degree duty ratio of each node variable; Arranging the abnormal contribution degree duty ratio in descending order according to the numerical value, selecting a node variable corresponding to the abnormal contribution degree duty ratio positioned at the first position after the sorting as a dominant abnormal node variable, and judging the dominant energy type; and generating energy-saving monitoring early warning information, pushing the energy-saving monitoring early warning information to a monitoring terminal of an energy management system for display, and storing the energy-saving monitoring early warning information in an energy-saving monitoring database.
3. The industrial energy-saving monitoring method based on the Copula graph model of claim 1, wherein comparing the joint probability density value with a preset energy-saving anomaly threshold comprises: based on a Copula graph model after training, constructing experience distribution by utilizing joint probability density values corresponding to all samples in a historical training data set, and selecting an alpha percentile of the experience distribution as a preset energy-saving abnormal threshold; When the joint probability density value is larger than or equal to a preset energy-saving abnormal threshold value, judging that the energy consumption state at the current moment is a normal state, and executing conventional monitoring and recording operation; When the joint probability density value is smaller than a preset energy-saving abnormal threshold value, judging that the energy consumption state at the current moment is an abnormal state, and immediately starting a multi-stage early warning and diagnosis response mechanism.
4. The industrial energy-saving monitoring method based on the Copula graph model of claim 3, wherein the joint probability density value is obtained by the following method, In a period to be monitored, acquiring real-time energy consumption data through an intelligent monitoring terminal of the Internet of things, and carrying out validity check on the real-time energy consumption data; Performing probability integral transformation on the verified real-time energy consumption data through an edge probability distribution function to obtain real-time uniform distribution values corresponding to various real-time energy consumption data, and forming a real-time uniform distribution vector; inputting the real-time uniform distribution vectors into a Copula graph model after training, and extracting all edge sets based on a final topological structure of the Copula graph model; and calculating a joint probability density value corresponding to the real-time uniform distribution vector based on tree topology decomposition characteristics of the Copula graph model.
5. The industrial energy-saving monitoring method based on the Copula graph model of claim 4, wherein the trained Copula graph model comprises the following steps: Each type of uniform distribution sequence in the multiple types of uniform distribution sequences is respectively defined as a node variable And all the node variables are combined into a node variable set ; For the node variable set Calculating mutual information values by adopting a mutual information estimation method based on rank correlation for any two node variables; traversing the node variable set All node variables are paired and combined to obtain a mutual information matrix M; based on the mutual information matrix M, constructing a topological structure of a Copula graph model by adopting a maximum weight spanning tree algorithm; For each edge in the topological structure, two node variables connected according to the edge are changed Tail dependent features of corresponding uniformly distributed sequence data are selected, a binary Copula function is selected, and a maximum likelihood estimation method is adopted to determine dependent parameters of the binary Copula function; Performing fitting goodness evaluation on the binary Copula function selected by each side in the topological structure by adopting a red pool information criterion, and determining an initial topological structure passing the fitting goodness evaluation as a final topological structure of a Copula graph model; and packaging the final topological structure, the binary Copula function type corresponding to each edge in the final topological structure and the dependent parameter combination of the binary Copula function to obtain a Copula graph model after training.
6. The industrial energy-saving monitoring method based on the Copula graph model of claim 5, wherein the maximum weight spanning tree algorithm is a Chow-Liu algorithm, the mutual information value is used as the weight value of the edge connecting the two corresponding node variables, the edge with the maximum weight value is selected from the mutual information matrix M to be sequentially added into the topological structure, and when the newly added edge and the existing edge form a loop, the edge is skipped until the topological structure is connected with all the node variables.
7. The industrial energy-saving monitoring method based on the Copula graph model of claim 5, wherein the acquisition method of the multi-class uniform distribution sequence is that, Sequentially extracting each type of energy time sequence from multiple types of energy time sequence as a target energy time sequence, wherein the target energy time sequence comprises N time-sequence energy consumption observations ; Fitting an edge probability density function to the target energy time sequence by adopting a non-parameter density estimation method, and carrying out numerical integration on the edge probability density function to obtain an edge cumulative distribution function; Substituting the energy consumption observation values into the edge cumulative distribution function based on each energy consumption observation value in the target energy consumption time sequence, and calculating a corresponding probability integral transformation value; All the energy consumption observation values in the target energy consumption time sequence are subjected to probability integral transformation to obtain all probability integral transformation values, and an even distribution sequence corresponding to the target energy consumption time sequence is formed according to the original time sequence; And respectively executing the operations on each type of energy-consuming time sequence in the multiple types of energy-consuming time sequence to obtain multiple types of uniformly distributed sequences.
8. The industrial energy-saving monitoring method based on the Copula graph model of claim 7, wherein the method for acquiring the edge cumulative distribution function is, Selecting a Gaussian kernel function as a kernel function; calculating an optimal bandwidth parameter of a nuclear density estimation method by adopting a Silverman rule of thumb; constructing a kernel density estimation function of a target energy time sequence based on the Gaussian kernel function and the optimal bandwidth parameter; And carrying out numerical integration on the kernel density estimation function from minus infinity to X to obtain an edge cumulative distribution function corresponding to the target energy time sequence.
9. The computer equipment comprises a memory and a processor, wherein the memory stores a computer program, and the computer equipment is characterized in that the processor realizes the steps of the industrial energy-saving monitoring method based on the Copula graph model according to any one of claims 1-8 when executing the computer program.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the Copula graph model-based industrial energy saving monitoring method as set forth in any one of claims 1 to 8.

Description

Industrial energy-saving monitoring method based on Copula graph model Technical Field The invention relates to the technical field of industrial energy efficiency monitoring and intelligent energy saving management, in particular to an industrial energy saving monitoring method based on a Copula graph model. Background Along with the continuous advancement of industrial digitization and intelligent processes, an industrial energy system gradually presents complex characteristics of multi-energy parallel supply, multi-equipment coupling operation and multi-working condition dynamic change. In order to realize the refined energy management under the 'double-carbon' target background, the industrial energy-saving monitoring technology gradually evolves from traditional manual inspection and single-index statistical analysis to the direction based on the perception of the Internet of things, data-driven modeling and intelligent analysis. At present, various energy consumption metering devices such as electric power, steam, fuel gas, water and the like are commonly deployed in industrial sites, and energy efficiency evaluation and anomaly identification are realized through centralized acquisition and analysis of energy consumption data. However, in the existing energy-saving monitoring method, multiple dependence on threshold judgment of single energy dimension or multivariate analysis model based on linear correlation assumption is difficult to accurately describe nonlinear dependence relationship between different energy types in time sequence dimension, and false alarm or missing alarm is easy to occur particularly under complex production load change and multi-energy cooperative consumption scene. In addition, the traditional method only pays attention to whether the abnormality occurs or not, but lacks the capability of finely tracing the cause of the abnormality, and is difficult to identify the dominant energy type causing the energy consumption abnormality, so that the practical application value of the energy-saving monitoring result in production regulation and control and energy-saving decision is restricted. CN114185960B discloses an optimization decision management method of town water, energy and environmental systems based on Copula functions, which realizes the optimal configuration and the joint risk management of a multi-resource system under uncertain conditions by constructing Copula joint distribution functions among water quantity, energy quantity and environmental indexes. The proposal fully utilizes the depicting capability of the Copula function to the variable related structure, and realizes the cooperative optimization of the water-energy-environment system on the macroscopic level. However, the method is mainly oriented to town-level resource planning and long-term decision-making problems, focuses on resource allocation and risk balance at a system level, is not designed for real-time energy consumption monitoring requirements of an industrial production site, does not relate to an anomaly detection mechanism based on time sequence energy consumption data, further lacks quantitative analysis of contribution degree of multiple types of energy sources in specific anomaly events, and is difficult to meet the requirements of industrial energy conservation monitoring on instantaneity, positioning performance and operability. The paper Copula model-based multidimensional data space scanning monitoring method provides a method for constructing multidimensional variable joint distribution by utilizing the Copula model and realizing anomaly monitoring by combining space scanning statistics. The research breaks through the limitation of the traditional multivariate statistical process control on distribution assumption, can improve the monitoring sensitivity of the multi-dimensional data in a runaway state to a certain extent, and can discover the trend change of the data earlier. Meanwhile, the method aims at anomaly detection and does not provide a structured analysis path of anomaly sources, and cannot locate a dominant anomaly variable in a complex energy consumption system, so that energy consumption responsibility division and targeted regulation and control under an energy-saving monitoring scene are not facilitated. In summary, although the related technology based on the Copula function or the Copula model has certain theoretical advantages in the aspects of multivariate correlation modeling and joint distribution depiction, the related technology focuses on macroscopic resource optimization, quality process monitoring or statistical out-of-control judgment, and generally has the problems that the technology is difficult to directly adapt to multi-source energy consumption time sequence data of an industrial site, lacks fine modeling on an energy variable condition dependent structure, cannot realize dominant energy type positioning after energy consumption abnormality occurs, and t