CN-121998502-A - Multi-mode intelligent analysis method and system for electric personnel capability assessment

CN121998502ACN 121998502 ACN121998502 ACN 121998502ACN-121998502-A

Abstract

The invention provides a multi-mode intelligent analysis method and a system for electric power talent ability assessment, which relate to the technical field of electric power industry, wherein the method comprises the steps of carrying out dynamic region division and combination in a three-dimensional feature space formed by an operation behavior mode, a state judgment index and interaction response characteristics according to a virtual type template to generate a scoring interval suitable for a current virtual imaging object; the method comprises the steps of establishing a personalized score adjustment rule set through interval mapping and normalization operation, calibrating initial capability assessment scores based on the personalized score adjustment rule set to obtain optimized capability scores, carrying out multi-dimensional weighted aggregation on the optimized capability scores to generate comprehensive capability assessment results, wherein the assessment results cover four virtual imaging dimensions of operation behavior modes, interaction response characteristics, state judgment indexes and execution stability. According to the invention, the matching degree is improved through dynamic distribution of modal weights, optimization of data processing, group difference calibration and deviation correction.

Inventors

ZHU QIANLI
ZHANG JINTAO
YUAN QIAN
WEN CHENCHEN
Sun Qiaohuan
YANG YANG
JI CHUNYANG
WANG YONG

Assignees

陕西华智云享科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (10)

1. A multi-modal intelligent analysis method for electric talent ability assessment, the method comprising: Step 1, based on a preset weight configuration strategy, carrying out weighted fusion on multi-modal input data with time sequence alignment and regular format to generate a fusion multi-modal data matrix; Step 2, inputting the multi-modal feature vector into a pre-trained neural network regression model to perform high-dimensional mapping calculation to obtain an initial capacity assessment value; step 3, based on the initial capacity evaluation value, extracting typical mode characteristics from the voice and the action modes, calculating the difference measurement between the typical mode characteristics and the reference modes in the virtual talent template library, and generating a characteristic difference vector; Step 4, according to the virtual type template, dynamic region division and combination are carried out in a three-dimensional feature space formed by an operation behavior mode, a state judgment index and interaction response characteristics, so as to generate a scoring interval which is suitable for the current virtual imaging object; Step 5, calibrating the initial capacity assessment score based on the personalized score adjustment rule set to obtain an optimized capacity score; And 6, performing multidimensional weighted aggregation on the optimized capability scores to generate comprehensive capability assessment results, wherein the assessment results cover four virtual imaging dimensions of operation behavior modes, interaction response characteristics, state judgment indexes and execution stability.
2. The method for multi-modal intelligent analysis of electrical talent ability assessment according to claim 1, further comprising, prior to step 1: The method comprises the steps of collecting a multi-mode original data stream generated in the power operation and maintenance process, wherein the multi-mode original data stream comprises a text record, a voice signal, an action sensing sequence and a numerical log; and dynamically configuring fusion weight coefficients for each mode data according to the operation environment grade and the operation task type to generate a weight distribution scheme.
3. The method for intelligent analysis of electric power talent ability assessment according to claim 2, wherein step 1, based on a preset weight configuration strategy, performs weighted fusion on multi-modal input data with time sequence alignment and regular format to generate a fusion multi-modal data matrix, extracts multi-modal feature vectors based on the matrix, and is used for representing operation behavior modes, interaction response characteristics and state judgment indexes, and comprises the following steps: based on the weight distribution scheme, performing time domain and space domain alignment processing on the regular multi-modal data to generate a time-space aligned multi-modal data sequence; based on the space-time aligned multi-mode data sequence, carrying out weighted fusion calculation according to fusion weight coefficients corresponding to all modes, and generating a weighted fusion multi-mode data matrix; Performing principal component analysis on the multi-mode data matrix after weighted fusion to obtain a principal component feature set after dimension reduction; based on the main component feature set, obtaining a final discrimination projection vector through linear discrimination analysis; And performing space projection transformation on the main component feature set by utilizing the final discrimination projection vector, extracting the feature combination with the maximum category distinction degree, and forming a multi-mode feature vector for representing the operation behavior mode, the interaction response characteristic and the state discrimination index.
4. The method for intelligent analysis of electric power talent ability assessment according to claim 3, wherein step 2 of inputting the multi-modal feature vector into a pre-trained neural network regression model for high-dimensional mapping calculation to obtain an initial ability assessment score comprises: based on the multi-mode feature vector, performing feature scaling processing by adopting a Z-score standardization method to obtain a standardized feature vector; The normalized feature vector is input into a pre-trained neural network regression model, forward propagation calculation is carried out through an input layer, a hidden layer and an output layer of the neural network regression model, and a multidimensional initial capacity assessment vector is obtained through the output layer; Based on the initial capability evaluation vector, adopting a Sigmoid activation function to perform nonlinear transformation processing to obtain transformed vector data; Mapping the transformed vector data to a preset standard numerical value interval through linear scaling calculation to obtain a standardized output vector; Based on the standardized output vector and a preset weight coefficient corresponding to the operation behavior mode, the interaction response characteristic and the state judgment index, a weighted sum algorithm is adopted to calculate a weighted sum, and the weighted sum is the initial capacity assessment score.
5. The method of claim 4, wherein step 3, based on the initial capability assessment score, extracts typical modal features from speech and motion modalities and calculates a difference metric between the typical modal features and a reference pattern in a virtual talent template library to generate a feature difference vector, comprising: Screening a key feature subset from the original feature data of the voice and the action mode based on the initial capability assessment value; Calculating the degree of difference between the key feature subset and each reference pattern in the virtual talent template library by adopting a distance measurement algorithm based on the key feature subset to obtain an initial difference measurement result; Performing dimension reduction processing by a principal component analysis method based on the initial difference measurement result to obtain dimension reduced difference characteristics; based on the difference characteristics after dimension reduction, carrying out pattern classification by adopting a cluster analysis method, and identifying the nearest reference pattern category; Based on the reference pattern category, extracting a corresponding reference feature vector, and calculating the difference value of the key feature subset and the reference feature vector in each dimension; And generating a normalized characteristic difference vector through normalization processing based on the difference value in each dimension.
6. The method for intelligent analysis of electric talent ability assessment according to claim 5, wherein the fast matching and pattern mapping of the feature difference vector using a geometric hash algorithm to obtain an adapted virtual type template comprises: performing dimension reduction processing by a principal component analysis method based on the feature difference vector to obtain a low-dimensional feature; based on the low-dimensional features, performing pattern classification by adopting a cluster analysis method, and identifying the similarity between the pattern classification and each reference pattern in the virtual talent template library; based on the similarity, selecting the first K reference modes with the highest similarity as a candidate template set; based on the candidate template set, extracting a corresponding reference feature vector, and constructing a hash table structure; based on the hash table structure, performing quick matching by adopting a geometric hash algorithm, and calculating the matching degree of the characteristic difference vector and each candidate template; and selecting a candidate template with the highest matching degree as an adaptive virtual type template based on the matching degree.
7. The method for assessing the capacity of electric power personnel according to claim 6, wherein the step 4 is characterized by performing dynamic region division and merging in a three-dimensional feature space formed by an operation behavior mode, a state judgment index and an interaction response characteristic according to the virtual type template to generate a scoring interval adapted to a current virtual imaging object, and constructing a personalized scoring adjustment rule set through interval mapping and normalization operation, and comprises the following steps: Based on the virtual type template, constructing a three-dimensional feature space formed by three dimensions of an operation behavior mode, a state judgment index and an interaction response characteristic; Based on the three-dimensional feature space, performing initial region division by adopting a cluster analysis method to obtain a plurality of initial feature regions; Based on the initial feature region, carrying out merging operation on regions which are similar in feature distribution and adjacent to each other in space by taking a feature distribution mode defined in a virtual type template as a merging basis to obtain optimized feature region division; Extracting characteristic distribution parameters of each region based on the optimized characteristic region division to obtain a dynamic scoring interval adapted to the current virtual imaging object; based on the dynamic scoring interval, mapping the original scoring data into a corresponding interval range by adopting a linear mapping algorithm to obtain an interval mapping result; And carrying out data standardization processing by a normalization processing method based on the interval mapping result to form a personalized score adjustment rule set.
8. The method of claim 7, wherein step 5 of calibrating the initial capacity assessment score based on the personalized score adjustment rule set to obtain an optimized capacity score comprises: based on the personalized score adjustment rule set and the initial capability assessment score, matching calculation of score data and adjustment rules is carried out by adopting a rule matching algorithm, and rule matching results of all dimensions are obtained; Calculating the deviation degree between the initial capacity evaluation value and the corresponding threshold value of the personalized score adjustment rule set by adopting a difference measurement algorithm based on the rule matching result to obtain score deviation data; Based on the scoring deviation data, performing calibration calculation on the initial capacity assessment value through an interpolation compensation algorithm to obtain a calibrated scoring result; and normalizing the scoring data to a standard numerical value interval by adopting a normalization processing method based on the calibrated scoring result to obtain an optimized capability score.
9. The method for multi-modal intelligent analysis of electric power talent ability assessment according to claim 8, wherein step 6, performing multi-dimensional weighted aggregation on the optimized ability scores to generate a comprehensive ability assessment result, where the assessment result covers four virtual imaging dimensions of operation behavior mode, interaction response characteristic, state judgment index and execution stability, and the method comprises: Extracting scoring data of four dimensions of operation behavior mode, interaction response characteristic, state judgment index and execution stability based on the optimized capability score; calculating a comprehensive evaluation score by a weighted aggregation algorithm through preset weight coefficients of each dimension based on the scoring data of the four dimensions; mapping the score to a preset evaluation level interval through normalization processing based on the comprehensive evaluation score to obtain a standardized evaluation result; Based on the standardized evaluation result, a comprehensive capability evaluation report covering four dimensions of operation behavior mode, interaction response characteristic, state judgment index and execution stability is obtained.
10. A power human ability assessment multimodal intelligent analysis system implementing the method of any of claims 1-9, comprising: The extraction module is used for carrying out weighted fusion on the multi-modal input data subjected to time sequence alignment and format regulation based on a preset weight configuration strategy to generate a fusion multi-modal data matrix; The mapping module is used for inputting the multi-modal feature vector into a pre-trained neural network regression model to perform high-dimensional mapping calculation so as to obtain an initial capacity assessment value; the matching module is used for extracting typical mode characteristics from the voice and the action modes based on the initial capacity evaluation value, calculating the difference measurement between the typical mode characteristics and the reference modes in the virtual talent template library, and generating a characteristic difference vector; the rule set module is used for carrying out dynamic region division and combination in a three-dimensional feature space formed by an operation behavior mode, a state judgment index and interaction response characteristics according to the virtual type template to generate a scoring interval which is suitable for a current virtual imaging object; the calibration module is used for calibrating the initial capacity assessment score based on the personalized score adjustment rule set to obtain an optimized capacity score; The evaluation result module is used for carrying out multidimensional weighted aggregation on the optimized capability scores to generate comprehensive capability evaluation results, wherein the evaluation results cover four virtual imaging dimensions of operation behavior modes, interaction response characteristics, state judgment indexes and execution stability.

Description

Multi-mode intelligent analysis method and system for electric personnel capability assessment Technical Field The invention relates to the technical field of power industry, in particular to a multi-mode intelligent analysis method and system for electric power talent ability assessment. Background As the scale of the power system is continuously enlarged and the technical complexity is continuously improved, the requirements of the power industry on professional skills are increasingly urgent. Traditional talent ability assessment methods mainly rely on written test assessment, expert observation assessment or single-dimension practical operation records, and have significant shortcomings in objectivity, comprehensiveness and efficiency. Especially in the data processing aspect, the prior art is difficult to meet the requirements of multi-mode and high-dimensional data analysis, so that the accuracy of an evaluation result is limited. For example, the prior art generally adopts a simple data stitching or weighted average mode, lacks an effective alignment and depth fusion mechanism for heterogeneous data such as video, audio, motion sensing and the like, and cannot accurately capture the space-time correlation and semantic complementarity between multi-mode data by using a rough fusion mode, so that important characteristic information is lost. Secondly, the electric power operation has strong time sequence characteristics, the existing method often adopts a static characteristic extraction mode, ignores the dynamic change rule in the operation process, and lacks the alignment of time sequence data and the dynamic characteristic extraction capability, so that the evolution process and the time dependency relationship of the operation behavior cannot be accurately modeled. Thirdly, in the prior art, a manually designed feature extraction method is mostly adopted, the features are often limited to surface phenomena, and semantic information of a data deep layer is difficult to mine. The lack of an end-to-end deep feature learning mechanism results in limited feature representation capability and inability to adapt to complex and diverse operating environments. In recent years, although research has been attempted to apply machine learning methods to talent assessment fields, the methods have some limitations that, on one hand, most of the methods are based on single-modality data and cannot fully utilize complementary advantages of multi-modality data, and on the other hand, even though multi-modality data are adopted, systematic data processing flows are lacking, and particularly, the methods have some defects in key links such as space-time alignment, feature fusion, personalized calibration and the like. Disclosure of Invention The invention aims to solve the technical problem of providing a multi-mode intelligent analysis method and a system for electric power talent ability assessment, which are used for improving the assessment matching degree by dynamically distributing mode weights, optimizing data processing, calibrating group differences and correcting deviations. In order to solve the technical problems, the technical scheme of the invention is as follows: In a first aspect, a method for multimodal intelligent analysis of electrical talent ability assessment, the method comprising: Based on a preset weight configuration strategy, carrying out weighted fusion on the multi-modal input data with time sequence alignment and regular format to generate a fusion multi-modal data matrix; Inputting the multi-modal feature vector into a pre-trained neural network regression model to perform high-dimensional mapping calculation to obtain an initial capacity assessment value; Based on the initial capability assessment value, extracting typical mode characteristics from voice and action modes, calculating difference metrics between the typical mode characteristics and reference modes in a virtual talent template library, and generating characteristic difference vectors; According to the virtual type template, dynamic region division and combination are carried out in a three-dimensional feature space formed by an operation behavior mode, a state judgment index and an interaction response characteristic, so as to generate a scoring interval adapted to the current virtual imaging object; Calibrating the initial capacity assessment score based on the personalized score adjustment rule set to obtain an optimized capacity score; And carrying out multidimensional weighted aggregation on the optimized capability scores to generate comprehensive capability assessment results, wherein the assessment results cover four virtual imaging dimensions of operation behavior modes, interaction response characteristics, state judgment indexes and execution stability. In a second aspect, an electrical human ability assessment multimodal intelligent analysis system, comprising: The extraction module is used for carrying out weigh