Evaluation method of digital transformation electric power project

Abstract

The invention discloses an evaluation method for a digital transformation power project, and relates to the field of big data mining, and the method comprises the steps: obtaining a multi-source heterogeneous data set D of the digital transformation power project; preprocessing the multi-source heterogeneous data set D to obtain a standardized data set D '; according to the standardized data set D ', constructing a multi-dimensional evaluation index model M through feature extraction and index mapping; respectively calculating a technical index weight matrix WT and a service index weight matrix WB by using the multi-dimensional evaluation index model M to form a weight matrix W; and according to the standardized data set D'and the weight matrix W, evaluating the digital transformation power project through a fuzzy comprehensive evaluation algorithm. Aiming at the problem that a traditional linear model ignores a non-linear relationship among power project indexes in the prior art, the method improves the power project evaluation precision through time sequence dynamic analysis, secondary fuzzy evaluation and the like.

Description

Technical Field

[0001] This application relates to the field of big data mining, and particularly to an evaluation method for digital transformation power projects. Background Art

[0002] As a basic industry of the national economy and the core field of the energy revolution, the power industry is accelerating the promotion of digital and intelligent transformation. With the rapid development of new-generation information technologies such as artificial intelligence, big data, Internet of Things, and cloud computing, digital transformation has become the only way for power enterprises to improve operational efficiency, optimize resource allocation, and enhance market competitiveness.

[0003] In the process of digital transformation of the power industry, power models have become the key technical support. Here, power models mainly refer to various prediction, optimization, control, and decision-making models constructed based on data-driven and mechanism analysis, including but not limited to load forecasting models, power grid stability analysis models, equipment fault diagnosis models, energy consumption behavior models, power market trading models, etc. These models directly affect the safe operation of the power system, the efficient operation of the power market, and the optimal allocation of energy resources, and their performance directly relates to the success or failure of digital transformation.

[0004] Regarding the evaluation of power models, traditional methods mainly conduct single evaluations from technical dimensions such as model accuracy and calculation efficiency. For example, the accuracy of prediction models is evaluated through indicators such as root mean square error (RMSE) and mean absolute percentage error (MAPE), or the performance of optimization models is evaluated through indicators such as convergence speed and calculation time. However, with the in-depth digital transformation of the power industry, this single technical dimension evaluation can no longer meet the comprehensive evaluation needs in complex scenarios.

[0005] Currently, the evaluation methods for power models and digital transformation projects have the following main problems: Traditional evaluation methods mostly adopt linear weighted models, assuming that each evaluation indicator is independent of each other, and obtaining the evaluation result through simple weighted summation. However, in the actual power system, there are complex non-linear relationships between technical indicators and business indicators, economic benefits and social benefits. For example, a 1% increase in load forecasting accuracy may not result in linear growth in economic benefits, but shows non-linear changes affected by various factors. Summary of the Invention

[0006] Aiming at the fact that traditional linear models in the prior art ignore the non-linear relationships between power project indicators, this application provides an evaluation method for digital transformation power projects, which improves the evaluation accuracy of power projects through time series dynamic analysis and secondary fuzzy evaluation.

[0007] The present application provides an evaluation method for a digital transformation power project, comprising: S1, obtaining a multi-source heterogeneous data set D of a digital transformation power project, wherein the multi-source heterogeneous data set D includes power project technical indicators D T and business indicators D B ; S2, preprocess the multi-source heterogeneous data set D to obtain a standardized data set D'; S3, based on the standardized data set D', construct a multidimensional evaluation index model M through feature extraction and index mapping; S4, use the multidimensional evaluation index model M to calculate the technical index weight matrix W T And business indicator weight matrix W B , forming a weight matrix W; S5, evaluating the digital transformation power project through a fuzzy comprehensive evaluation algorithm based on the standardized data set D' and the weight matrix W. The digital transformation power project in this application mainly refers to the power model.

[0008] Furthermore, technical indicator D T Including model performance indicators; the model performance indicators include model accuracy indicators and model stability indicators; business indicators D B Including cost-effectiveness indicators; the cost-effectiveness indicators include economic benefit indicators E(t) and social benefit indicators S(t).

[0009] Furthermore, S2 obtains the standardized data set D', including: using the quartile method IQR to analyze the technical indicators D T Perform outlier processing to obtain technical indicator D T '; Using Box-Cox transformation to analyze the technical indicator D T 'Perform distribution transformation, transform non-normal distribution data into approximate normal distribution, and obtain standardized technical indicators D T "; For business indicators D B Perform time series analysis and nonlinear processing to obtain standardized business indicators D B '.

[0010] Furthermore, we get the standardized business indicator D B ', including: performing correlation analysis on the economic benefit index E(t) and the social benefit index S(t) to obtain the coupling relationship matrix C of the mutual influence degree between the indicators; using the time sliding window analysis method to perform time series weighting processing on the coupling relationship matrix C to obtain the dynamic weight coefficient W(t) including the time influence, W(t) = e -αt, where \(t\) represents the time interval relative to the evaluation reference point, and \(a\) is the optimal decay coefficient obtained by fitting historical data; for the economic benefit index \(E(t)\), the weighted economic benefit index \(E'\) at each time point \(t\) is calculated using the dynamic weight coefficient \(W(t)\), \(E'=\sum[E(t)\times W(t)]\); for the social benefit index \(S(t)\), the optimal time-delay parameter \(\tau\) is fitted according to historical data, and the time-delay response function \(R(\tau)=1 - e\) -βτ ; according to the time-delay response function \(R(\tau)\), the social benefit index \(S(t)\) is subjected to a time-lag transformation \(S'(t)\), \(S'(t)=S(t - \tau)\times R(\tau)\); using the modified dynamic weight coefficient \(W'(t)\), the social benefit index data \(S''\) after quantifying the lag effect is obtained, \(S''=\sum[S'(t)\times W'(t)]\); a non-linear transformation is performed on the weighted economic benefit index \(E'\) and the social benefit index data \(S''\) after quantifying the lag effect to obtain the standardized business index data \(E\) norm and \(S\) norm , \(E\) norm =F(E') and \(S\) norm =F(S'), where \(F\) represents an exponential function mapping transformation, which converts the index value to the unified \([0, 1]\) interval and retains the original non-linear distribution characteristics.

[0011] Among them, the time-sliding window analysis method and the dynamic weight coefficient \(W(t)=e\) are introduced -αt , effectively capturing the dynamic change law of the power project benefits over time and overcoming the limitations of traditional static evaluation. Through the time-delay response function \(R(\tau)=1 - e\) -βτ and the modified dynamic weight coefficient \(W'(t)=e\) -α't , the lag manifestation characteristics of social benefits are scientifically quantified, making the evaluation results more in line with the actual project benefit development law.

[0012] Furthermore, the modified dynamic weight coefficient \(W'(t)\) has the following expression: \(W'(t)=e\) -α't ; \(a' = a\times(1 - \gamma)\); where \(\gamma\) is the social benefit persistence coefficient, which is used to characterize the long-term impact characteristics of the social benefit index compared with the economic benefit index; where \(\gamma\) is the social benefit persistence coefficient, which is used to characterize the long-term impact characteristics of the social benefit index compared with the economic benefit index; where, through the exponential function mapping transformation \(F\) and non-linear transformation, the original non-linear distribution characteristics of the index are retained, overcoming the limitations of traditional linear evaluation models.

[0013] Furthermore, in S3, a multi-dimensional evaluation index model M is constructed, including: extracting the model accuracy characteristics and model stability characteristics from the standardized technical index \(D\) T ” to construct the technical dimension index subspace \(MT\); for the standardized business index \(D\) B 's economic benefit index \(E\)norm and the social benefit index S norm Extract the time series features and non-linear distribution features to construct the subspace M of business dimension indicators B ; Analyze the interaction between technical indicators and business indicators based on the coupling relationship matrix C to construct the cross-dimensional joint feature space M J ; Combine the subspace M of technical dimension indicators T , the subspace M of business dimension indicators B and the cross-dimensional joint feature space M J for fusion mapping to obtain the comprehensive evaluation index model M containing multi-dimensional evaluation indicators

[0014] Furthermore, in S4, form the weight matrix W, including: Based on the model accuracy feature and model stability feature in the subspace M of technical dimension indicators T , use the analytic hierarchy process to construct the technical indicator weight matrix W T ; Based on the time series features and non-linear distribution features extracted from the subspace M of business dimension indicators B , calculate the time series weighted importance of business indicators according to the modified dynamic weight coefficient W'(t) to obtain the business indicator weight matrix W B ; Use the coupling relationship matrix C as the quantization basis for the correlation strength between indicators, calculate the interaction influence coefficient between technical indicators and business indicators in the cross-dimensional joint feature space M J , establish the cross-evaluation matrix W C , the cross-evaluation matrix W C characterizes the contribution degree of technical indicator improvement to business indicator improvement; Combine the technical indicator weight matrix W T , the business indicator weight matrix W B and the cross-evaluation matrix W C to construct the multi-dimensional comprehensive weight matrix W through the block matrix combination method where λ T , λ B , λ C are the overall weight coefficients of the technical dimension, business dimension and cross dimension respectively, and satisfy λ T +λ B +2λ C =1

[0015] Among them, analyzing the interaction between technical indicators and business indicators based on the coupling relationship matrix C reveals the deep-level correlation mechanism between indicators, making up for the deficiency of the traditional method of evaluating indicators separately. Through the cross-evaluation matrix W C , quantitatively characterize the contribution degree of technical indicator improvement to business indicator improvement, and realize the collaborative evaluation from the technical and business perspectives

[0016] Furthermore, S5 evaluates the digital transformation power project based on the standardized data set D' and the weight matrix W through the fuzzy comprehensive evaluation algorithm, including: establishing an evaluation factor set U, including the standardized indicators D of the technical dimension T "The model accuracy index and model stability index, as well as the business dimension standardization index D B The economic benefit index E in ' norm and social benefit index S norm ; Establish the evaluation level set V = {V1, V2, ..., V n}, define n-level evaluation criteria, including excellent, good, average, poor and unqualified; construct the time series dynamic fuzzy relationship matrix R(t) = {r ij (t)} m×n , where r ij (t) represents the membership of the i-th evaluation factor to the j-th evaluation level at time t; the nonlinear membership function μ is used for technical indicators T =F(D T ”), F represents the exponential function mapping transformation; the membership function of the business indicator is modified by the time-delay response in, represents the original membership function without considering the time lag effect, D B '(t-τ) represents the business indicator value at the time point (t-τ), R(τ) is the time-delay response function); for the block matrix weight matrix Decompose the technical indicator weight component W T,adj =λ T ×W T , business indicator weight component W B,adj =λ B ×W B , cross-influence weight component W C,adj =λ C ×W C ; Use the two-level fuzzy evaluation mechanism to calculate the fuzzy evaluation results of technical dimensions B T =W T,adj ◇R T and business dimension fuzzy evaluation results B B =W B,adj ◇R B , where ◇ represents the mixed fuzzy synthesis operation. Based on the coupling relationship matrix C and the cross-evaluation matrix W C , calculate the impact matrix of technical indicators on business indicators in, Represents a coupling-aware composition operation.

[0017] Evaluation results by integrating technical dimensions B T , Business dimension evaluation results BB and the influence degree matrix I to obtain the comprehensive evaluation result wherein is the vector outer product operation, η1, η2, η3 are integrated weight coefficients and satisfy η1 + η2 + η3 = 1; the main evaluation level is determined according to the maximum membership degree principle, that is, G = V k , satisfying b k = max(b j ), j = 1, 2,......, n; meanwhile, calculate the comprehensive evaluation score S = ∑(b j ×s j ), wherein, s j is the score of the j-th level; combining the refined evaluation results G T and G B of the technical dimension and the business dimension, generate an evaluation report

[0018] Furthermore, ◇ represents the hybrid fuzzy composition operation, defined as (a◇b) j = ∑(a i ×b i ×δ i ) + (1 - δ i )×max(min(a i , b i ))), δ i is the linearity parameter of the i-th index, determined according to the non-linear distribution characteristics of the index

[0019] Furthermore represents the coupling perception composition operation, defined as quantify the non-linear contribution degree of the improvement of technical indicators to the improvement of business indicators

[0020] Compared with the prior art, the advantages of the present application are as follows

[0021] In the digital transformation power project, due to the time difference between policy changes, market fluctuations, technology updates and social responses, there are a large number of non-linear relationships between economic benefit indicators and social benefit indicators. The traditional linear model simply superimposes the weights of each indicator and uses static fixed weights for calculation, which has defects such as low evaluation accuracy, inability to reflect the time lag effect of social benefits and inability to capture the complex coupling relationship between indicators in the digital transformation power project. The present application proposes a non-linear evaluation method based on time series dynamic analysis and secondary fuzzy evaluation, which combines the coupling relationship matrix and the time lag response function to achieve the precise quantification of the time lag effect of social benefits, the scientific characterization of the interaction between technology and business indicators, and the significant improvement of evaluation accuracy BRIEF DESCRIPTION OF THE DRAWINGS

[0022] This application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same numbers represent the same structures, where:

[0023] Figure 1 is an exemplary flowchart of an evaluation method for a digital transformation power project according to some embodiments of this application;

[0024] Figure 2 is a comprehensive evaluation system for power projects according to some embodiments of this application;

[0025] Figure 3 is a radar chart of the scores of first-level indicators according to some embodiments of this application;

[0026] Figure 4 is a schematic diagram of the scores of second-level indicators according to some embodiments of this application;

[0027] Figure 5 is an exemplary flowchart of fuzzy evaluation according to some embodiments of this application. Detailed implementation manners

[0028] The methods and systems provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0029] As Figure 1 shown, obtain a multi-source heterogeneous data set D of a digital transformation power project, where the multi-source heterogeneous data set D includes power project technical indicators D T and business indicators D B ; preprocess the multi-source heterogeneous data set D to obtain a standardized data set D'; according to the standardized data set D', construct a multi-dimensional evaluation index model M through feature extraction and index mapping; use the multi-dimensional evaluation index model M to calculate the technical index weight matrix W T and the business index weight matrix W B respectively to form a weight matrix W; evaluate the digital transformation power project through a fuzzy comprehensive evaluation algorithm according to the standardized data set D' and the weight matrix W.

[0030] S1. Obtain a multi-source heterogeneous data set D of a digital transformation power project. Extract model accuracy indicators from the operation log system of the power model, such as the MAPE value (mean absolute percentage error) of the load prediction model, the convergence time of the power grid stability analysis model, the recognition accuracy of the fault diagnosis model, etc. Obtain model stability indicators from the system monitoring platform, such as the performance fluctuations of the model under different load conditions, the accuracy drift rate during long-term operation, the robustness test results for abnormal inputs, etc. Obtain the performance differences of the model in different scenarios through A / B test records, and establish a multi-dimensional technical performance index set DT .

[0031] Extract direct economic benefit indicators E(t) from the enterprise financial system, including cost savings from energy scheduling optimization, the economic value of reduced grid losses, and revenue growth from power trading optimization, to form time series data. Collect social benefit indicators S(t) through questionnaire surveys, third-party evaluation reports, and government public data, including power supply reliability improvement index, changes in user satisfaction, carbon emission reduction, and degree of optimization of energy consumption structure. Establish an indicator collection cycle and data quality control mechanism to ensure the temporal consistency and comparability of data, and construct business indicators D B .

[0032] S2, preprocess the multi-source heterogeneous data set D to obtain the standardized data set D'. Use the quartile method IQR to analyze the technical indicators D T The specific steps for outlier processing are as follows: For each technical indicator, calculate its first quartile Q1 and third quartile Q3, and obtain the interquartile range IQR = Q3-Q1. Set the outlier judgment standard, and usually regard data points outside the range of [Q1-1.5×IQR, Q3+1.5×IQR] as outliers. For the identified outliers, they are not deleted directly, but a segmented replacement strategy is adopted: for extreme outliers (outside the range of [Q1-3×IQR, Q3+3×IQR]), determine whether they are caused by equipment failure or data collection errors, and if so, eliminate them. For moderate outliers, use the nearest average or quantile replacement method for correction. For outliers that may reflect real performance fluctuations (such as model performance under special working conditions), retain and mark them, and perform sensitivity analysis later. Through this processing, the technical indicator D is obtained. T ', providing a clean data foundation for subsequent transformations.

[0033] Technical indicator D T The specific implementation of the Box-Cox transformation is as follows: First, perform a normality test on each technical indicator, such as the Shapiro-Wilk test or QQ plot analysis, to identify whether there is a significant skewed distribution. Apply the Box-Cox transformation to non-normally distributed indicators: Or Y(λ)=ln(X),λ=0. The optimal λ value is determined by the maximum likelihood estimation method to make the transformed data closest to the normal distribution: for model accuracy indicators (such as MAPE values), it is usually observed that the distribution is right-skewed in the high-precision range, and λ is usually between 0.1 and 0.5. For stability indicators (such as the coefficient of volatility), a left-skewed distribution may be observed. In this case, λ may be negative or require data inversion before transformation. Apply a specific λ parameter to each indicator to obtain a standardized technical indicator D T”. This method is particularly effective for highly non-linear distributed data that is difficult to effectively standardize using Z-score, avoiding the distortion of the distribution pattern by traditional methods.

[0034] Construct the association analysis and coupling relationship, and use the Dynamic Time Warping (DTW) algorithm to analyze the temporal similarity between the economic benefit index E(t) and the social benefit index S(t), identifying potential time-delay patterns. Combine the Granger causality test and cross-correlation analysis to determine the degree of association between the indicators at different time lags. Construct the coupling relationship matrix C, where C ij represents the coupling degree between the i-th economic benefit index and the j-th social benefit index, with a value range of [0, 1]: indicators with a completely positive correlation have corresponding values close to 1 (such as a reduction in power grid losses and a reduction in carbon emissions); indicators with no obvious correlation have corresponding values close to 0 (such as some short-term cost savings and long-term social benefits); indicator pairs with time-delay correlation use the lag cross-correlation coefficient.

[0035] Adopt the time-sliding window method (such as a 24-month sliding window) to analyze the stability of the coupling relationship under different time windows. According to the timeliness characteristics of the indicators, construct the dynamic weight coefficient W(t) = e -αt : For the load forecasting model, fitting from historical data gives α approximately 0.15 (monthly data), indicating that for each additional month of time interval, the influence of the indicator decays by approximately 14%. For the equipment status forecasting model, the α value may be smaller (about 0.05 to 0.08), reflecting its long-term benefit characteristics. When fitting the α value, use historical project case data, and determine the value that best matches the actual project benefits through optimization methods such as grid search or gradient descent.

[0036] Weighted processing of economic benefit indicators, collect the economic benefit indicators E(t) at different time points t, including direct cost savings, reduced maintenance costs, economic gains brought by improved power quality, etc. Apply the dynamic weight coefficient W(t) to calculate the weighted economic benefit indicator E': for short-term projects (such as distribution network optimization), the recent weights are higher, and the benefits within 6 to 12 months are mainly evaluated. For long-term projects (such as power grid planning models), use a smaller α value to balance short-term and long-term benefits. Adopt the sliding accumulation method to incrementally calculate the impact of the latest data on the overall evaluation and achieve dynamic updates of the evaluation.

[0037] Time-delay processing of social benefits, analyze historical data to identify the typical time-delay parameters τ of different types of social benefit indicators: the power supply reliability indicator may have a lag effect of 3 to 6 months; the change in user satisfaction may have a lag of 6 - 12 months; the optimization of the regional energy structure may have a lag of 12 to 24 months; construct the time-delay response function R(τ) = 1 - e -βτ, where β is usually between 0.1 and 0.3 (monthly data), reflecting the characteristic that the social benefits gradually emerge over time. Apply its specific time-lag parameter to each social benefit indicator S(t) to calculate the time-lagged transformation value S'(t) = S(t - τ) × R(τ): This processing method takes into account both the lag time (t - τ) of indicator realization and quantifies the intensity of this lag effect through R(τ).

[0038] To quantify the lag effect, introduce the social benefit persistence coefficient γ, whose typical value range is from 0.2 to 0.5, reflecting that the social benefits have a longer lasting impact period than the economic benefits: for carbon emission reduction benefits, the γ value can reach 0.4 to 0.5; for the improvement of electricity penetration rate, γ can be set to 0.3 to 0.4; for the improvement of short-term user experience, γ can be set to 0.1 to 0.2; calculate the modified dynamic weight coefficient W'(t) = e -α't , where a' = a × (1 - γ): This makes the weight decay rate of social benefit indicators slower than that of economic benefit indicators, conforming to its long-term impact characteristics. Apply the modified weights to calculate the social benefit indicator S” = ∑[S'(t) × W'(t)] after quantifying the lag effect.

[0039] Nonlinear transformation: For the weighted economic benefit indicator E' and the social benefit indicator S”, adopt the exponential function mapping transformation F instead of simple linear normalization: for the economic benefit indicator, the function of the form can be used, where k is the shape parameter; for the cost indicator, the function of the form can be used; this nonlinear transformation retains the original distribution characteristics and maps all indicators to the unified [0, 1] interval: for growth-type indicators, the conversion curve is steeper in the middle and low value intervals, reflecting the law of diminishing marginal benefits; for control-type indicators, the curve is steeper in the interval close to the target value, reflecting the increasing difficulty of fine control; the shape parameter k is determined by fitting historical data and is usually between 1.5 and 3. Different types of indicators use different parameters to ensure that the converted distribution can best reflect the actual impact characteristics of the indicators.

[0040] S3. According to the standardized data set D', construct a multi-dimensional evaluation index model M through feature extraction and index mapping. For the standardized technical indicators D TExtract the model accuracy features and model stability features, and construct the technical dimension index subspace MT. For the load forecasting model, extract the steady-state accuracy features (average accuracy during the stable operation period) and dynamic accuracy features (accuracy performance during the load mutation period) from the standardized MAPE, RMSE and other data. For the fault diagnosis model, extract the sensitivity feature (true positive rate) and specificity feature (true negative rate) respectively, and construct the area under the ROC curve as the comprehensive accuracy feature. For the optimization decision model, extract the optimization objective achievement degree and constraint satisfaction rate to form the decision quality feature vector. Use principal component analysis (PCA) to compress the high-dimensional accuracy indicators into 2-3 main feature dimensions, retaining more than 85% of the information content.

[0041] Extract the model stability features, calculate the coefficient of variation of each accuracy indicator under different operating conditions, and form the stability evaluation indicators. Through sliding window analysis, calculate the drift rate of the model accuracy to evaluate the long-term stability. Construct the sensitivity features to quantify the response degree of the model to the input data perturbation. Establish the interpretability features to evaluate the transparency and traceability of the model decision-making process.

[0042] Construct the feature pair <P, S>, where P represents the accuracy feature vector and S represents the stability feature vector. On this basis, define the technical dimension index subspace M T ={<P i , S i >|i = 1, 2,....., n}, where n is the number of evaluation models. Use the multidimensional scaling (MDS) method to visually reduce the dimension of the high-dimensional feature space to assist experts in understanding the distribution of technical indicators.

[0043] For the standardized business indicator D B 's economic benefit indicator E norm and social benefit indicator S norm Extract the time series features and non-linear distribution features, and construct the business dimension index subspace M B .

[0044] Extract the time series features, apply time series decomposition to the economic benefit indicator E norm to extract the trend feature, seasonal feature and periodic feature. Calculate the slope of the cumulative benefit curve of the social benefit indicator S norm to quantify the benefit growth rate. Use the autoregressive integrated moving average (ARIMA) model to extract the time series pattern features and predict the future benefit trend. Construct the time series feature vector TS = {trend, seasonality, velocity, acceleration} to comprehensively characterize the time evolution law of the business indicators.

[0045] Extract the non-linear distribution features, for E norm and Snorm Kernel density estimation (KDE) was applied to obtain the distribution density function f(x). Statistical characteristics such as skewness, kurtosis, and entropy were calculated to quantify the nonlinear characteristics of the distribution. Quantile regression was used to extract eigenvalues at different quantiles (0.25, 0.5, and 0.75). The nonlinear distribution eigenvector NL = {skewness, kurtosis, entropy, quantiles} was constructed.

[0046] Fusion of time series features TS and nonlinear distribution features NL to construct business dimension feature tuples<E,S> , where E represents economic benefit characteristics and S represents social benefit characteristics. Define the business dimension indicator subspace

[0047] M B ={<E i ,S i >|i=1,2,.....,m}, where m is the number of business indicators. Apply factor analysis to identify potential common factors and simplify the dimensionality of the business indicator space.

[0048] Based on the coupling relationship matrix C, the interaction between technical indicators and business indicators is analyzed to construct a cross-dimensional joint feature space MJ. First, the technology-business mapping relationship is analyzed. Using the element values of the coupling relationship matrix C, the impact graph (ImpactGraph) of technical indicators to business indicators is constructed. For each pair of technical indicator-business indicator combination (i, j), its direct impact intensity DI is calculated. ij and indirect impact path set IP ij . Apply path analysis methods to identify the impact paths and bottlenecks of key technical indicators on business indicators.

[0049] Extract interactive impact features and calculate the comprehensive impact of each technical indicator on all business indicators For each business indicator, calculate the degree to which it is affected by all technical indicators Construct a technology-business interaction response curve, fitting a nonlinear response function to changes in technology indicators and business indicators. Extract characteristic parameters of the response function, such as sensitivity, saturation point, and threshold point.

[0050] Construct the interaction feature matrix J, where element J ij The feature vector representing the impact of the i-th technical indicator on the j-th business indicator. Define the cross-dimensional joint feature space M J = {J|J is the interaction feature matrix}. Apply association rule mining to discover nonlinear association patterns between technical indicators and business indicators.

[0051] The technical dimension index subspace M T , business dimension indicator subspace M Band the cross - dimensional joint feature space M J Perform fusion mapping to obtain a comprehensive evaluation index model M that contains multi - dimensional evaluation indexes. Construct a three - dimensional relationship graph G=(V, E), where the vertex set V contains technical index nodes and business index nodes, and the edge set E represents the correlation relationship between indexes. Define the adjacency matrix A of the graph, where non - zero elements indicate the existence of a correlation between indexes, and zero elements indicate no correlation. Calculate the centrality index of the graph to identify key influencing indexes and bridging indexes.

[0052] Adopt the tensor fusion method to represent M T , M B and M J as a multi - dimensional tensor T. Apply tensor decomposition techniques (such as Tucker decomposition) to extract core features and reduce dimensions. Construct a collaborative filtering matrix between multi - dimensional indexes to identify potential index combination patterns. Define the comprehensive evaluation index model M=(M T , M B , M J , R), where R represents the set of mapping relationship sets between three sub - spaces. Design a set of mapping functions Φ={φ T , φ B , φ J}, which respectively represent the mapping methods from each sub - space to the comprehensive evaluation result. Construct a hierarchical evaluation structure to support drill - down analysis from the comprehensive score to the sub - dimensions. Implement an adaptive threshold mechanism to dynamically adjust the evaluation criteria according to the characteristics of different types of power projects.

[0053] S4. Based on the model accuracy characteristics and model stability characteristics in the technical dimension index subspace MT, use the analytic hierarchy process to construct a technical index weight matrix WT, and construct a three - layer structure: the target layer (comprehensive evaluation of technical indexes), the criterion layer (model accuracy, model stability), and the index layer (specific technical indexes). In the criterion layer, determine the relative importance of model accuracy and stability according to the application scenarios of different power models. In the index layer, subdivide various technical indexes. For example, under prediction accuracy, it can be divided into point prediction accuracy and interval prediction accuracy.

[0054] Organize power system experts and data science experts to evaluate the index importance using the Delphi method. Construct a pairwise comparison judgment matrix B, where the element b ij represents the importance degree of index i relative to index j, using a 1 - 9 scale. For different types of power models (such as load forecasting, fault diagnosis, optimal scheduling, etc.), construct a specific set of judgment matrices.

[0055] Calculate the maximum eigenvalue λ max of the judgment matrix and the corresponding eigenvector. Conduct a consistency test, calculate the consistency index and the consistency ratio C R =CI / R I When C R <0.1, it is considered that the judgment matrix has satisfactory consistency; otherwise, experts need to readjust the judgment matrix. Normalize the eigenvector to obtain the weights of each index and form the technical index weight matrix WT. Collect the judgment results of multiple experts to construct a combined judgment matrix. Based on the background and experience of experts, set the expert weight coefficient. Use the weighted geometric mean method to fuse the judgment results of multiple experts to obtain a more objective technical index weight matrix WT.

[0056] Based on the time series features and non-linear distribution features extracted from the business dimension index subspace M B , calculate the time series weighted importance of business indicators according to the corrected dynamic weight coefficient W'(t) to obtain the business indicator weight matrix W B For each business indicator, based on its time series feature vector T S , calculate the long-term trend importance T I , seasonal fluctuation importance S I and emergency response importance E I . Construct the time series importance vector T W ={T I , S I , E I}, which reflects the influence of indicators at different time scales. For the economic benefit indicator E norm and the social benefit indicator S norm , design different time series importance calculation parameters respectively.

[0057] Based on the non-linear distribution feature vector N L , calculate the distribution skewness adjustment coefficient S A and kurtosis adjustment coefficient K A . Construct the distribution correction coefficient D C =f(S A , K A ), which is used to adjust the basic weight to better reflect the actual distribution characteristics of the indicators. For highly non-linear distribution indicators, increase the sensitivity weight in the extreme value region.

[0058] Apply the corrected dynamic weight coefficient W'(t)=e -α't , calculate the weighted importance of business indicators at different time points. For the economic benefit indicator, design a short-term response type weight function W E (t)=W'(t)×T W . For the social benefit indicator, design a long-term cumulative type weight function W S (t)=W'(t)×(1 + γ×t)×T W . Integrate the weighted results at different time points to form the comprehensive weight vector of business indicators.

[0059] Combine the weight vectors of each index of economic benefits and social benefits to form the business index weight matrix WB. Use the entropy weight method to optimize the initial weights and increase the weights of indexes with large amounts of information. Conduct sensitivity analysis to test the impact of different weight configurations on the evaluation results and ensure the robustness of weight allocation.

[0060] Use the coupling relationship matrix C as the quantification basis for the correlation strength between indexes, and calculate the cross-dimensional joint feature space M J The interaction influence coefficient between technical indexes and business indexes in it, and establish the cross-evaluation matrix W C . Based on the coupling relationship matrix C, extract the direct influence coefficient DI of technical index i on business index j ij . For time-varying coupling relationships, calculate their time-weighted average influence intensity. Construct the direct influence coefficient matrix DI as the basis for cross-evaluation.

[0061] Construct an index influence propagation network, with each index as a node and the edge weight as the direct influence coefficient. Apply an influence propagation algorithm (such as a PageRank variant) to calculate the comprehensive influence coefficient considering multi-step propagation. Identify key propagation paths and influence hubs, and analyze the optimal path for technical improvement.

[0062] Collect historical project case data, analyze the corresponding relationship between technical index improvement and business index enhancement. Establish a technology-business response curve and fit the non-linear response function R(x) = a×(1 - e -b×x ). Calculate the marginal gain of business indexes brought by each unit of technical index improvement and quantify the contribution degree of technical improvement.

[0063] Integrate the quantification results of direct influence coefficients, indirect influence coefficients and contribution degrees to construct the initial cross-evaluation matrix W C,init . Apply the fuzzy comprehensive evaluation method to adjust the initial matrix to form the final cross-evaluation matrix W C . Design a cross-validation mechanism to verify the accuracy of the cross-evaluation matrix through historical cases.

[0064] The technical index weight matrix W T , the business index weight matrix W B , and the cross-evaluation matrix W C , and construct a multi-dimensional comprehensive weight matrix W through the block matrix combination method. Based on the project type and application scenario, determine the technical dimension weight coefficient λ T , the business dimension weight coefficient λ B , and the cross-dimension weight coefficient λ C : For R & D projects, the technical dimension weight λ T can be set to 0.4 - 0.5; for commercialization projects, the business dimension weight λ BCan be set to 0.4 - 0.5; cross - dimension weight λ C Usually set to 0.1 - 0.2; ensure that the weight coefficient satisfies the constraint condition λ A +λ B +λ C = 1. Design an adaptive adjustment mechanism for the weight coefficient to dynamically adjust the dimension weight according to the project phase.

[0065] Scale the technical index weight matrix W T to obtain λ T ×W T . Scale the business index weight matrix WB to obtain λ B ×W B . Scale the cross - evaluation matrix WC to obtain λ C ×W C .

[0066] Adopt the 2×2 block matrix form to combine the three scaled weight matrices: Among them, the upper - left block represents the self - evaluation weight of technical indicators, the lower - right block represents the self - evaluation weight of business indicators, and the upper - right and lower - left blocks represent the cross - evaluation weights between technical and business indicators. Check whether the combined weight matrix W meets the consistency requirement. Use genetic algorithm or particle swarm optimization algorithm to fine - tune the elements in the weight matrix to minimize the evaluation error. Verify the performance of different weight configurations on the test dataset and select the optimal weight matrix. Design a weight sensitivity analysis tool to help decision - makers understand the impact of weight changes on the evaluation results.

[0067] As Figure 5 shown in S5, according to the standardized dataset D' and the weight matrix W, evaluate the digital transformation power project through the fuzzy comprehensive evaluation algorithm. First, establish an evaluation factor set U, and screen key model accuracy indicators from the standardized technical indicators D T ": such as the MAPE value of the load forecasting model (better than 3% is the first - level standard), the F1 score of the fault diagnosis model (higher than 0.9 is excellent), and the convergence characteristic index of the optimization model. Select core model stability indicators: including the accuracy fluctuation coefficient (measuring the stability of the model in different scenarios), the long - term drift rate (measuring the stability of the model performance over time), and the robustness index (measuring the resistance of the model to abnormal inputs). Adopt a progressive screening mechanism, and set different index sets for different types of power models. For example, the distribution network optimization model focuses on calculation efficiency indicators, while the load forecasting model pays more attention to the peak - valley forecasting accuracy.

[0068] [[ID=4)]Business dimension index screening: Extract economic benefit indicators E from the standardized business indicators D B '. norm: including direct cost savings rate, operation efficiency improvement ratio, resource utilization improvement degree, fault loss reduction amount, etc. Select social benefit indicator S norm : including power supply reliability improvement index, user satisfaction change, carbon emission reduction amount, energy consumption structure optimization degree, etc. Apply principal component analysis (PCA) for dimensionality reduction, and retain key indicators that explain more than 85% of the total variance to avoid redundancy of evaluation factors.

[0069] Index hierarchical organization: Construct a three-level evaluation factor hierarchical structure: overall goal layer (comprehensive evaluation of digital transformation) → (first-level indicator) dimension layer (technology dimension, business dimension) → (second-level indicator) index layer (specific evaluation indicators). For the technology dimension indicators, further group them into model accuracy class, model stability class, and model applicability class. For the business dimension indicators, subdivide them into short-term economic benefit class, long-term economic benefit class, direct social benefit class, and indirect social benefit class. As Figure 3 shown, it is the radar chart of the first-level indicator scores in this embodiment, Figure 4 is the radar chart of the second-level indicator scores in this embodiment; this embodiment mainly evaluates the carbon emission analysis model and the charging pile analysis model, where the first-level indicators include model performance, model reliability, model availability, application results, and cost-benefit; the second-level indicators include model accuracy, model stability, data sufficiency, technological advancement, scope of application, and industry recognition.

[0070] Establish an evaluation grade set V = {V1, V2,......, V n}, V1 (excellent): The indicator performance far exceeds expectations and reaches the industry-leading level (90 - 100 points); V2 (good): The indicator performance exceeds expectations and is significantly higher than the industry average level (80 - 89 points); V3 (average): The indicator performance meets expectations and conforms to the industry average level (70 - 79 points); V4 (poor): The indicator performance fails to meet expectations and is lower than the industry average level (60 - 69 points); V5 (unqualified): The indicator performance is significantly lower than expectations and has obvious defects (below 60 points).

[0071] Indicator specialization grade standard: For prediction models, the excellent grade is defined as: MAPE < 2% (day-ahead prediction) or < 5% (medium- and long-term prediction); for optimization models, the excellent grade is defined as: objective function improvement > 12% and constraint satisfaction rate > 99.5%; for economic benefit indicators, the excellent grade is defined as: return on investment > 25% or cost savings > 15%; for social benefit indicators, the excellent grade is defined as: power supply reliability improvement > 5% or user satisfaction improvement > 15%.

[0072] Level boundary blurring: Set a level transition interval. For example, the transition range from good to excellent is 88 - 92 points. An indicator may have different degrees of membership to two adjacent levels simultaneously. Apply the fuzzy boundary theory and define the smooth transition between levels through the S-shaped membership function to avoid the "cliff effect" in traditional evaluations.

[0073] Construct a time-series dynamic fuzzy relation matrix \(R(t)=\{r ij (t)\} m×n , Time dimension sampling: Establish an evaluation time series \(\{t_1,t_2,\cdots,t k \}\). Usually, select a quarter or half-year as the evaluation interval. For short-term projects (within 1 year), conduct monthly evaluations; for medium- and long-term projects (1 - 3 years), conduct quarterly evaluations. Set key time points to strengthen sampling, such as densifying sampling before and after project milestone time points to capture the changing trends.

[0074] Dynamic membership degree calculation: For each time point \(t\) and each evaluation factor \(i\), calculate its membership degree \(r ij (t)\) to each evaluation level \(j\): For technical indicators, calculate the membership degree based on the proximity of the actual value of the indicator at this time point to the level standard. For business indicators, calculate the membership degree considering the time lag effect. Form a time-series membership degree matrix set \(\{R(t_1),R(t_2),\cdots,R(t k )\}\) to reflect the dynamic changes of evaluations over time.

[0075] Time-series consistency guarantee: Apply time-series smoothing techniques to ensure that the changes in membership degrees between adjacent time points conform to the principle of continuity, avoiding drastic fluctuations without actual reasons. Design an anomaly detection mechanism to identify and analyze the reasons behind the mutation points of membership degrees. Introduce a time window moving average to reduce the impact of short-term random factors on evaluations.

[0076] For indicators with seasonal variations (such as electricity load forecasting), introduce a seasonal correction factor to avoid evaluation biases caused by seasonal fluctuations. Identify periodic patterns and interpret the evaluation results in the context of the corresponding cycle to improve the rationality of time-series evaluations.

[0077] Nonlinear membership degree functions for technical indicators. For cost-type indicators (the smaller the value, the better, such as MAPE), design a decreasing exponential function: where \(k\) is the shape parameter; for benefit-type indicators (the larger the value, the better, such as accuracy), design an increasing exponential function: For interval-type indicators (with an optimal interval, such as model complexity), design a bilateral exponential function: Based on the actual distribution characteristics of the metrics, determine the optimal shape parameter k: For metrics sensitive to high-precision regions (such as recognition rates close to 100%), set a larger k value (usually 3 - 5); for metrics sensitive to medium-precision regions, set a medium k value (usually 1.5 - 3); for metrics sensitive to low-precision regions, set a smaller k value (usually 0.8 - 1.5); Use historical data samples to fit the optimal k value to make the membership degree distribution closest to the expert evaluation results. Organize power system experts and data science experts to set key inflection point values (such as the boundary value between excellent and good). Based on the expert judgment results, fine-tune the shape of the membership function to conform to industry consensus. Introduce an expert correction coefficient δ exp , allowing fine-tuning of the calculation results in specific scenarios.

[0078] For multiple technical metrics of the same type (such as prediction accuracies at different time scales), design a combined membership function: μ T,comb = w1×μ T1 + w2×μ T2 +...... + w n ×μ Tn , where the weight coefficient w i is determined according to the relative importance of each metric and satisfies ∑w i = 1.

[0079] For business metrics, use a membership function F B (D B ', τ) with time-delay response correction. For each type of business metric, determine the optimal time-delay parameter τ through historical data analysis: For short-term economic benefit metrics (such as operating cost savings), the time delay is short, and τ is usually 1 - 3 months; for medium-term economic benefit metrics (such as maintenance cost reduction), τ is usually 3 - 6 months; for long-term social benefit metrics (such as power supply reliability improvement), τ is usually 6 - 18 months; Use lag cross-correlation analysis to identify the optimal lag period between technical metric improvement and business metric enhancement.

[0080] Calculate the original membership degree without considering the time-delay effect For economic benefit metrics, usually use an S-shaped membership function; for social benefit metrics, usually use a convex membership function; Optimize the membership function parameters for different types of metrics to reflect their specific distribution characteristics.

[0081] For the evaluation of business metrics at time point t, actually use the metric value at time point (t - τ): D B '(t - τ)), and apply the time-delay response function R(τ) = 1 - e -βτ for correction. The β parameter reflects the intensity of the time-delay effect: For fast-response metrics, the β value is larger (usually 0.3 - 0.5); for slow-response metrics, the β value is smaller (usually 0.05 - 0.2); Calculate the time-delay corrected membership degree:

[0082] For long-term social benefit indicators (such as the improvement of electricity penetration rate), consider the benefit accumulation characteristics: i ranges from t - T to t - 1; where D(t - i) is a decay function that describes the decay effect of historical benefits on the current evaluation; γ is an accumulation coefficient, usually between 0.1 and 0.3, reflecting the intensity of the continuous influence of historical benefits.

[0083] Block decomposition of the weight matrix: Decompose the comprehensive weight matrix W into three components according to the block matrix structure: the weight component of technical indicators W T,adj = λ T ×W T ; the weight component of business indicators W B,adj = λ B ×W B ; the weight component of cross-influence W C,adj = λ C ×W C . Ensure that the dimensions of each component match. When the number of technical indicators is n and the number of business indicators is m, W T,adj is an n×n matrix, W B,adj is an m×m matrix, and W C,adj is an n×m matrix.

[0084] Determination of the overall weight coefficient: Dynamically determine the weight coefficients of each dimension according to the type and development stage of the power project: For projects in the initial R & D stage: λ T = 0.5, λ B = 0.3, λ C = 0.1; For projects in the promotion and application stage: λ T = 0.4, λ B = 0.4, λ C = 0.1; For projects in the mature operation stage: λ T = 0.3, λ B = 0.5, λ C = 0.1; Verify that the weight coefficients satisfy the constraint condition: λ T + λ B + 2λ C = 1.

[0085] Adaptive adjustment during the project life cycle: Design a time-varying function of the weight coefficient to automatically adjust the dimension weights as the project progresses:

[0086] λ C (t) is adjusted accordingly to maintain the sum constraint; where t is the project progress time, T is the total project cycle, and ρ and σ are adjustment rate parameters.

[0087] Special scenario weight offset: Design a weight offset mechanism for specific evaluation objectives: Technical feasibility evaluation: Increase λ T Weight (+0.1 - 0.2); Business value evaluation: Increase λ B Weight (+0.1 - 0.2); Technical transformation efficiency evaluation: Increase λ C Weight (+0.05 - 0.1); After offset, perform weight normalization to ensure that the constraint conditions are still met. Adopt a two - level fuzzy evaluation mechanism to calculate the evaluation results of the technical dimension and the business dimension respectively.

[0088] First - level evaluation of the technical dimension: Construct a fuzzy relation matrix R for the subset of technical indicators T ; Each technical indicator corresponds to a row, and the evaluation levels correspond to the columns; The matrix element r ij represents the membership degree of technical indicator i to evaluation level j; First - level evaluation of the business dimension: Construct a fuzzy relation matrix R for the subset of business indicators B ; Each business indicator corresponds to a row, and the evaluation levels correspond to the columns; The matrix element r ij represents the membership degree of business indicator i to evaluation level j.

[0089] Implementation of hybrid fuzzy composition operation: Determine the linearity parameter δ of each indicator i : For indicators with an approximately linear distribution, δ i is set higher (0.7 - 0.9); For indicators with an obviously non - linear distribution, δ i is set lower (0.3 - 0.6); Estimate δ through the linear fitting R 2 value of the historical data of the indicator i : δ i ≈min(0.9, R 2 +0.2).

[0090] Execute the hybrid fuzzy composition operation B T = W T,adj ◇R T : For each evaluation level j, calculate This operation combines the advantages of the weighted average method (suitable for linear relationships) and the maximum - minimum method (suitable for non - linear relationships); It adaptively processes indicators with different distribution characteristics.

[0091] Fuzzy evaluation of the technical dimension: Calculate the fuzzy evaluation vector B T of the technical dimension, and each element (B T ) j represents the membership degree of the technical dimension to the j - th evaluation level. Perform normalization processing: Ensure that the sum of membership degrees is 1. Verify the rationality of the evaluation results and adjust the linearity parameter δ if necessary iBusiness dimension fuzzy evaluation: Calculate the business dimension fuzzy evaluation vector B using the same hybrid fuzzy synthesis operation. B = W B,adj ◇R B Consider the membership degree of business indicators after correcting for time-delay effects to make the evaluation results more objective and accurate. For the economic benefit and social benefit sub-dimensions, they can be calculated separately and then weighted and combined.

[0092] Based on the coupling relationship matrix C and the cross-evaluation matrix W C , calculate the influence degree matrix I of technical indicators on business indicators, and the coupling perception synthesis operation mechanism: Realize the coupling perception synthesis operation For each pair of technical indicator i and business indicator j, calculate This operation is similar to the "OR" operation in probability theory and can reasonably quantify the synergistic influence degree between two indicators; when both factors are strong (close to 1), the result approaches 1; when either factor is weak, the result decays rapidly.

[0093] Nonlinear contribution degree quantification: For each element of the influence degree matrix I, it is interpreted as the nonlinear contribution degree of technical indicator i to business indicator j. Set the contribution degree threshold (usually 0.3 - 0.5) to identify the key influence paths. Construct an influence heat map to intuitively display the influence intensity distribution of technical indicators on business indicators.

[0094] Influence link analysis: Based on the influence degree matrix I, mine the most effective path from technical improvement to business improvement. Calculate the overall influence of each technical indicator: Calculate the degree of being influenced of each business indicator: Identify high-influence technical indicators and high-sensitivity business indicators to provide a basis for optimizing resource allocation.

[0095] Influence time difference analysis: Conduct a comparative analysis of the influence degree matrices {I(t1), I(t2),......, I(t k )} at different time scales. Identify the patterns of influence intensity changing with time, such as enhanced type, decay type, and fluctuation type. Establish an influence response time model to predict the time window when business benefits appear after technical improvement.

[0096] By integrating the technical dimension evaluation result BT, the business dimension evaluation result BB, and the influence degree matrix I, obtain the comprehensive evaluation result B. Determine the integration weight coefficients: According to the project characteristics and evaluation objectives, determine three integration weight coefficients: The technical dimension weight η1: Usually between 0.3 and 0.5; The business dimension weight η2: Usually between 0.3 and 0.5; The interaction influence weight η3: Usually between 0.1 and 0.3; Ensure that the three weight coefficients satisfy the constraint condition: η1 + η2 + η3 = 1; Use the analytic hierarchy process or entropy weight method to scientifically determine the optimal weight configuration.

[0097] Implementation of vector cross - product operation: Calculate the vector cross - product of the evaluation results in the technology dimension and the business dimension

[0098] Form an n×m - dimensional matrix; this matrix represents the joint membership degree of the technology evaluation level i and the business evaluation level j co - occurring; Combine the cross - product result with the influence degree matrix: I·(BT⊕BB) to strengthen the evaluation weight of the path with high influence degree.

[0099] Calculation of comprehensive evaluation result: Execute the comprehensive evaluation formula: The first item reflects the evaluation in the pure technology dimension; the second item reflects the evaluation in the pure business dimension; the third item reflects the interactive influence evaluation between the technology and business dimensions; Normalize the comprehensive evaluation result: Ensure that the sum of membership degrees is 1.

[0100] Sensitivity analysis and verification: By adjusting the integrated weight coefficient, analyze the impact of parameter changes on the evaluation result. Use historical case data to verify the accuracy and stability of the comprehensive evaluation result. Establish evaluation accuracy indicators, such as the consistency with expert evaluation, the accuracy rate of predicting business indicators, etc.

[0101] Determine the main evaluation level and generate an evaluation report according to the maximum membership degree principle. Determination of the main evaluation level: Apply the maximum membership degree principle to determine the main evaluation level: G = V k , satisfying b k = max(b j ), j = 1, 2,......, n; Considering the proximity of membership degrees, when the difference between the second - largest membership degree and the largest membership degree is less than the threshold (usually 0.1), it is identified as a fuzzy evaluation and a dual - level result is given. Calculate the evaluation certainty index: As an indicator of the confidence level of the evaluation result.

[0102] Calculation of comprehensive score: Calculate the comprehensive evaluation score: S = ∑(b j ×s j ), where s j is the score of the j - th level (e.g., excellent = 95 points, good = 85 points, etc.). Generate a score interval: [S - ΔS, S + ΔS], where ΔS reflects the uncertainty of the evaluation, ΔS = α×(1 - Certainty)×Range(s), and α is an adjustment parameter (usually 0.3 - 0.5). The sub - scores of different dimensions are also calculated in the same way for easy comparison and analysis.

[0103] Refinement of evaluation result generation: Refined evaluation result of the technology dimension G T : Includes the evaluation results of each technical sub - dimension such as model accuracy and model stability. Refined evaluation result of the business dimension GB : The evaluation results include various business sub - dimensions such as economic benefits and social benefits. Establish a comparison chart between dimensions to visually display the relative strengths of evaluations in each dimension. Identify key advantage points and improvement points, and calculate the contribution degree of indicators to the total score.

[0104] Intelligent generation of evaluation reports: Generate a structured evaluation report, including the following core parts: Comprehensive evaluation results: Overall grade, score and its confidence interval; Dimension evaluation results: Evaluation grades and scores of the technology dimension and business dimension; Key indicator analysis: The most influential technology indicators and the most sensitive business indicators; Time - series trend analysis: The change trend of key indicators over time and inflection point analysis; Advantages and disadvantages: Outstanding highlights and aspects that need improvement of the project; Improvement suggestions: Based on the impact degree matrix, put forward targeted improvement suggestions.

Claims

1. An evaluation method for digital transformation power projects, characterized in that Including: S1. Obtain a multi-source heterogeneous dataset D of digital transformation power projects, where the multi-source heterogeneous dataset D includes power project technical indicators D T and business indicators D B ; S2. Preprocess the multi-source heterogeneous dataset D to obtain the standardized dataset D'. S3. Construct a multi-dimensional evaluation index model M based on the standardized dataset D' through feature extraction and index mapping. S4. Use the multi-dimensional evaluation index model M to calculate the technical index weight matrix W T and the business index weight matrix W B , to form the weight matrix W; S5. Evaluate the digital transformation power project through the fuzzy comprehensive evaluation algorithm according to the standardized dataset D' and the weight matrix W.

2. The evaluation method of the digital transformation power project according to claim 1, wherein: Technical indicator D T including model performance indicators; the model performance indicators include model accuracy indicators and model stability indicators; Business metric D B including cost-benefit metrics; the cost-benefit metrics include economic benefit metric E(t) and social benefit metric S(t).

3. The evaluation method of the digital transformation power project according to claim 2, wherein: S2. Obtain the standardized dataset D', including: Use the interquartile range (IQR) method to perform outlier processing on the technical indicator D T to obtain the technical indicator D T '; Use the Box-Cox transformation to perform a distribution transformation on the technical indicator D T ' to convert non-normal distribution data into an approximately normal distribution, obtaining the standardized technical indicator D T ”; For business metric D B Perform time series analysis and nonlinear processing to obtain the standardized business metric D B '.

4. The evaluation method of the digital transformation power project according to claim 3, wherein: Obtain the standardized business metric D B ', including: Conduct a correlation analysis on the economic benefit index E(t) and the social benefit index S(t) to obtain the coupling relationship matrix C of the mutual influence degree between the indexes. Using the time-sliding window analysis method to perform time-series weighting processing on the coupling relationship matrix C, a dynamic weight coefficient W(t) including the influence of time is obtained, W(t) = e -αt , where t represents the time interval relative to the evaluation reference point, and a is the optimal decay coefficient fitted according to historical data; For the economic benefit index E(t), calculate the weighted economic benefit index E' at each time point t using the dynamic weight coefficient W(t), E' = ∑[E(t) × W(t)]. For the social benefit index S(t), the optimal time-delay parameter τ is fitted based on historical data, and the time-delay response function R(τ) = 1 - e -βτ ; According to the time-delay response function R(τ), perform a time-lag transformation S'(t) on the social benefit index S(t), S'(t) = S(t - τ) × R(τ). Using the modified dynamic weight coefficient W'(t), obtain the social benefit index data S” after quantifying the lag effect, S” = ∑[S'(t) × W'(t)]. Perform a non - linear transformation on the weighted economic benefit index E' and the social benefit index data S" after quantifying the lag effect to obtain the standardized business index data E norm and S norm , E norm = F(E') and S norm = F(S'), where F represents an exponential function mapping transformation that converts the index values to the unified [0, 1] interval and retains the original non - linear distribution characteristics.

5. The evaluation method of the digital transformation power project according to claim 4, wherein: The modified dynamic weight coefficient W'(t) has the following expression: W'(t) = e -α't a' = a × (1 - γ) where γ is the social benefit persistence coefficient, used to characterize the long-term influence characteristics of the social benefit index compared with the economic benefit index.

6. The evaluation method of the digital transformation power project according to any one of claims 2 to 5, wherein: S3. Construct a multi-dimensional evaluation index model M, including: For the standardized technical index D T "Extract the model accuracy feature and the model stability feature to construct the technical dimension index subspace MT; For the standardized business metric D B 's economic benefit metric E norm and social benefit metric S norm Extract time series features and non-linear distribution features to construct the business dimension metric subspace M B ; Analyze the interaction effects between technical indicators and business indicators based on the coupling relationship matrix C, and construct a cross-dimensional joint feature space M J ; Perform a fusion mapping on the technical dimension index subspace M T , the business dimension index subspace M B and the cross-dimension joint feature space M J to obtain a comprehensive evaluation index model M containing multi-dimensional evaluation indexes.

7. The evaluation method of the digital transformation power project according to claim 6, wherein: S4. Form the weight matrix W, including: Based on the model accuracy feature and model stability feature in the technical dimension index subspace M T use the analytic hierarchy process to construct the technical index weight matrix W T ; Based on the time series features and non-linear distribution features extracted from the business dimension index subspace M B According to the corrected dynamic weight coefficient W'(t), calculate the time series weighted importance of the business indicators to obtain the business indicator weight matrix W B ; Using the coupling relationship matrix C as the quantification basis for the correlation strength between indicators, calculate the cross-dimensional joint feature space M J The interaction influence coefficient between technical indicators and business indicators in J , and establish the cross-evaluation matrix W C , the cross-evaluation matrix W C represents the contribution degree of the improvement of technical indicators to the improvement of business indicators; The technical index weight matrix W T , the business index weight matrix W B , and the cross-evaluation matrix W C are combined by block matrix to construct a multi-dimensional comprehensive weight matrix W , where λ T , λ B , λ C are the overall weight coefficients of the technical dimension, business dimension and cross dimension respectively, and satisfy λ T + λ B + 2λ C = 1.

8. The evaluation method of the digital transformation power project according to claim 7, wherein: S5. Evaluate the digital transformation power project through the fuzzy comprehensive evaluation algorithm, including: Establish an evaluation factor set U, which includes the technical dimension standardization index D T The model accuracy index and model stability index in "", as well as the business dimension standardization index D B The economic benefit index E in '' norm And the social benefit index S norm ; Establish an evaluation level set V = {V1, V2,......, V n}, define the evaluation criteria for n levels, and the evaluation criteria include excellent, good, average, poor, and unqualified; Construct a time-series dynamic fuzzy relation matrix \(R(t)=\{r ij (t)\} m×n , where \(r ij (t)\) represents the membership degree of the \(i\)-th evaluation factor to the \(j\)-th evaluation level at time \(t\); Adopt a non-linear membership function μ for technical indicators T = F(D T ”), where F represents an exponential function mapping transformation; Membership function with time-delay response correction for business metrics Among them, represents the original membership function without considering the time-delay effect, and D B '(t - τ) represents the value of the business metric at the time point (t - τ), and R(τ) is the time-delay response function); Decompose the weight matrix in the form of a block matrix The weight component W of the technical index T,adj = λ T ×W T , the weight component W of the business index B,adj = λ B ×W B , the weight component W of the cross - influence C,adj = λ C ×W C ; Adopt a two-level fuzzy evaluation mechanism to calculate the fuzzy evaluation results B of the technical dimension and the fuzzy evaluation results B of the business dimension respectively T =W T,adj ◇R T and the fuzzy evaluation results B of the business dimension B =W B,adj ◇R B , where ◇ represents the hybrid fuzzy composition operation Based on the coupling relationship matrix C and the cross-evaluation matrix W C , calculate the influence degree matrix of technical indicators on business indicators Among them, represents the coupling perception synthesis operation; By integrating the evaluation result B of the technical dimension T , the evaluation result B of the business dimension B and the influence degree matrix I, the comprehensive evaluation result is obtained wherein is the vector outer product operation, and η1, η2, η3 are integration weight coefficients and satisfy η1 + η2 + η3 = 1; Determine the main evaluation level according to the principle of maximum membership degree, i.e., G = V k , satisfying b k = max(b j ), j = 1, 2,......, n; At the same time, calculate the comprehensive evaluation score S = ∑(b j × s j ), where s j is the score of the j-th level; Combine the refined evaluation results G T , G B of the technical dimension and the business dimension to generate an evaluation report.

9. The evaluation method of the digital transformation power project according to claim 8, wherein: ◇ represents the hybrid fuzzy composition operation, defined as (a◇b) j = ∑(a i × b i × δ i ) + (1 - δ i ) × max(min(a i , b i ))), where δ i is the linearity parameter of the i-th index and is determined according to the non-linear distribution characteristics of the index.

10. The evaluation method of the digital transformation power project according to claim 8, wherein: Denotes a coupled perception synthesis operation, defined as Quantify the non-linear contribution degree of the improvement of quantization technical indicators to the improvement of business indicators.

Images