CN-122022560-A - Coal sampling representative evaluation method based on multi-modal data fusion

CN122022560ACN 122022560 ACN122022560 ACN 122022560ACN-122022560-A

Abstract

The invention relates to the technical field of coal quality detection and discloses a coal sampling representative evaluation method based on multi-modal data fusion, which comprises the steps of acquiring multi-modal data in the coal sampling process; the method comprises the steps of performing feature extraction on multi-modal data, constructing a sampling action compliance recognition model, judging the compliance of coal sampling action, outputting a first confidence coefficient, constructing a sampling point position distribution compliance model, calculating the compliance degree of an actual sampling point set and a theoretical sampling point set, outputting a second confidence coefficient, fusing the first confidence coefficient and the second confidence coefficient, and generating a sampling representative evaluation index.

Inventors

GAO WEI
ZHANG LEI
ZHANG WENXIN
GUO XIAOHU
BAO JUN
TIAN HAO

Assignees

杭州华电双冠能源科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260112

Claims (10)

1. A coal sampling representative evaluation method based on multi-modal data fusion, the method comprising: Acquiring multi-mode data in the coal sampling process; Extracting characteristics of the multi-mode data; based on the extracted features, a sampling action compliance recognition model is constructed, the compliance of the coal sampling action is judged by using the sampling action compliance recognition model, and a first confidence coefficient is output; Based on the extracted features, a sampling point position distribution coincidence degree model is constructed, the coincidence degree of an actual sampling point set and a theoretical sampling point set is calculated by using the sampling point position distribution coincidence degree model, and a second confidence coefficient is output; and fusing the first confidence coefficient and the second confidence coefficient to generate a sampling representative evaluation index.
2. The method of claim 1, wherein the multi-modal data includes control signals of a sampling device PLC, geographical location information of a sampling head, and video stream data, the method further comprising, prior to feature extraction of the multi-modal data: And carrying out space-time alignment on the control signal of the sampling equipment PLC, the geographical position information of the sampling head and the video stream data by adopting the UTC timestamp, and establishing the corresponding relation among the sending time of the sampling control instruction, the geographical position coordinates of the sampling head and the video image picture.
3. The method of claim 2, wherein determining compliance of the coal sampling action using the sampling action compliance recognition model, outputting a first confidence level, comprises: analyzing the video stream data frame by frame based on a target detection model of deep learning, and identifying key actions of coal sampling, wherein the key actions comprise whether a sampling head falls on the ground, whether a sample is successfully transferred to a sample barrel or not, and whether offer omission phenomenon exists or not; and verifying the consistency of the control signal of the sampling device PLC and the key action identified from the video stream data, and outputting a first confidence degree based on a consistency verification result.
4. The method of claim 1, wherein calculating the degree of coincidence of the actual set of sampling points and the theoretical set of sampling points using the sampling point distribution coincidence model, and outputting a second confidence level, comprises: Selecting a quantization mode matched with an application scene, and quantizing the current spatial distribution difference degree of an actual sampling point set and a theoretical sampling point set; and outputting the confidence coefficient corresponding to the current spatial distribution difference coefficient as a second confidence coefficient according to the mapping relation between the spatial distribution difference coefficient and the confidence coefficient.
5. The method of claim 4, wherein selecting a quantization mode that matches an application scene comprises: If the application scene is point set boundary matching evaluation, selecting and calculating Hausdorff distance as a quantization mode; If the application scene is the distribution overall similarity evaluation, selecting a distribution similarity based on kernel density estimation as a quantization mode; If the application scene is a rule area sampling evaluation, selecting the overlapping degree of the grid area covered by the sampling points as a quantization mode.
6. The method of claim 1, wherein the fusing the first confidence level and the second confidence level to generate a sampled representative evaluation index comprises: Respectively determining the weight corresponding to the first confidence coefficient and the weight corresponding to the second confidence coefficient; And fusing the first confidence coefficient with the weight corresponding to the first confidence coefficient and fusing the second confidence coefficient with the weight corresponding to the second confidence coefficient by adopting a weighted average method to obtain a sampling representative evaluation index.
7. The method of claim 1, wherein after fusing the first confidence and the second confidence to generate a sampled representative evaluation index, the method further comprises: Generating an evaluation report according to the sampling representative evaluation index; And when the index value in the evaluation report is lower than a preset index threshold value, generating alarm information.
8. A coal sample representative evaluation device based on multi-modal data fusion, the device comprising: The data acquisition module is used for acquiring multi-mode data in the coal sampling process; The feature extraction module is used for extracting features of the multi-mode data; the sampling action compliance recognition module is used for judging the compliance of the coal sampling action and outputting a first confidence coefficient; The coincidence degree calculating module is used for constructing a sampling point position distribution coincidence degree model based on the extracted features, calculating the coincidence degree of an actual sampling point set and a theoretical sampling point set by using the sampling point position distribution coincidence degree model, and outputting a second confidence degree; and the fusion module is used for fusing the first confidence coefficient and the second confidence coefficient and generating a sampling representative evaluation index.
9. An electronic device, comprising: A memory and a processor, the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, thereby executing the coal sampling representative evaluation method based on multi-mode data fusion according to any one of claims 1 to 7.
10. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the coal sample representative evaluation method based on multimodal data fusion of any one of claims 1 to 7.

Description

Coal sampling representative evaluation method based on multi-modal data fusion Technical Field The invention relates to the technical field of coal quality detection, in particular to a coal sampling representative evaluation method based on multi-mode data fusion. Background The coal sampling is used as a source link of coal quality detection, and the representativeness of the result directly determines the authenticity and effectiveness of the subsequent sample preparation and test results, so that the method is a key basis for coal trade settlement and enterprise cost control. In the prior art, most of the methods rely on the field inspection of a supervisor with abundant experience, and the methods have the advantages of high cost, low efficiency and easiness in interference of human factors, and are difficult to realize accurate evaluation of coal sampling. Disclosure of Invention The invention provides a coal sampling representative evaluation method based on multi-mode data fusion, which aims to solve the problem that accurate evaluation of coal sampling is difficult to realize because the prior art relies on field inspection of a supervisor. In a first aspect, the invention provides a coal sampling representative evaluation method based on multi-modal data fusion, the method comprising: Acquiring multi-mode data in the coal sampling process; Extracting characteristics of the multi-mode data; Based on the extracted features, a sampling action compliance recognition model is constructed, the compliance of the coal sampling action is judged by using the sampling action compliance recognition model, and a first confidence coefficient is output; Based on the extracted features, a sampling point position distribution coincidence degree model is constructed, the coincidence degree of an actual sampling point set and a theoretical sampling point set is calculated by using the sampling point position distribution coincidence degree model, and a second confidence coefficient is output; And fusing the first confidence coefficient and the second confidence coefficient to generate a sampling representative evaluation index. According to the invention, the limitation of single data acquisition is broken through by acquiring multi-modal data, the problem of incomplete data acquisition is solved from the source, the characteristics are extracted from the multi-modal data, so that the characteristics related to sampling representativeness are screened out, the compliance of coal sampling actions is judged by utilizing a sampling action compliance recognition model, sampling illegal operation is accurately recognized, the coincidence degree of an actual sampling point set and a theoretical sampling point set is calculated by utilizing a sampling point distribution coincidence degree model, the difference between the actual sampling point set and the theoretical sampling point set is scientifically quantized, the prior art is replaced by relying on supervisor evaluation, and sampling representativeness evaluation indexes are generated by combining sampling compliance and space coverage. In an alternative embodiment, the multi-modal data includes a control signal of the sampling device PLC, geographical location information of the sampling head, and video stream data, and the method further includes, prior to feature extraction of the multi-modal data: And carrying out space-time alignment on the control signal of the sampling equipment PLC, the geographical position information of the sampling head and the video stream data by adopting the UTC timestamp, and establishing the corresponding relation among the sending time of the sampling control instruction, the geographical position coordinates of the sampling head and the video image picture. According to the invention, the UTC time stamp is adopted to perform space-time alignment on the multi-mode data, so that a corresponding relation is established, the space-time consistency of the multi-mode data is ensured, and the data association deviation is eliminated. In an alternative embodiment, determining compliance of the coal sampling action using a sampling action compliance recognition model, outputting a first confidence level, comprising: analyzing video stream data frame by frame based on a target detection model of deep learning, and identifying key actions of coal sampling, wherein the key actions comprise whether a sampling head falls to the ground, whether a sample is successfully transferred to a sample barrel or not, and whether offer omission phenomenon exists or not; and verifying the consistency of the control signal of the sampling device PLC and the key action identified from the video stream data, and outputting a first confidence degree based on a consistency verification result. According to the invention, the video stream data is analyzed by adopting the target detection model based on deep learning, the key action recognition is recognized, the r