Calculation power thermal evaluation method based on multi-dimensional data analysis and decision making system thereof

Abstract

The invention provides a computing power thermal evaluation method based on multi-dimensional data analysis and a decision making system thereof, and relates to the technical field of computing power analysis, and the method comprises the steps: obtaining an original data set, and carrying out the preprocessing of the original data set to generate a standardized data set; performing vectorization processing and heterogeneous fusion on the standardized data set to generate a unified fusion data set; a hierarchical evaluation system is constructed, correlation analysis is performed based on evaluation index features, and a computing power evaluation feature set is formed; different initial index weights are configured for different types of regions; calculating an initial calculation force thermal value of each region; performing iterative optimization on the initial index weight by using an optimization algorithm; and recalculating the regional computing power thermal value based on the optimized index weight, generating a thermal time-space distribution map, and generating a computing power resource allocation scheme according to the thermal gradient difference. According to the method, the systematicness and objectivity of computing power resource assessment are improved, and a reliable decision basis is provided for regional computing power resource allocation.

Description

Technical Field

[0001] The present invention relates to the technical field of computing power analysis, and particularly to a computing power heat evaluation method based on multi-dimensional data analysis and its decision-making system. Background Art

[0002] Existing computing power analysis and evaluation decisions mainly adopt statistical index analysis methods and empirical evaluation methods. The statistical index analysis method mainly evaluates based on hardware indicators such as the number of racks, server scale, CPU utilization rate, etc., and conducts comprehensive analysis by combining economic indicators such as energy consumption and investment scale. The empirical evaluation method qualitatively judges and classifies the regional computing power development level based on historical data and expert experience. In practical applications, some regions adopt a scoring model based on a single dimension, or use a simple weighted average method for computing power resource evaluation and decision-making analysis.

[0003] However, the above technical solutions have obvious deficiencies: First, the single-dimensional evaluation method is difficult to comprehensively reflect the complexity of regional computing power development and ignores the mutual influence between multi-dimensional indicators; Second, the simple linear scoring model cannot accurately depict the non-linear influence characteristics of indicators on computing power development; Third, the existing evaluation methods lack in-depth analysis of policy documents and are difficult to quantify the actual impact of the policy environment on computing power development; Finally, the static evaluation results are difficult to adapt to the dynamic changes of computing power requirements, and the visualization degree of the evaluation results is low, which is not conducive to intuitive understanding and quick decision-making. Summary of the Invention

[0004] In view of this, the present invention proposes a computing power heat evaluation method based on multi-dimensional data analysis and its decision-making system, which provides decision-making support for the reasonable allocation and efficient utilization of regional computing power resources by integrating multi-source heterogeneous data, constructing a hierarchical evaluation system, introducing a non-linear weighted model, and dynamic visualization display.

[0005] The technical solution of the present invention is implemented as follows: On the one hand, the present invention provides a computing power heat evaluation method based on multi-dimensional data analysis, including: S1. Obtain an original data set containing active subject data, economic data, resource environment data, and policy documents, and preprocess the original data set to generate a standardized data set; S2. Perform vectorization processing and heterogeneous fusion on the standardized data set to generate a unified fusion data set; S3. Construct a hierarchical evaluation system, extract corresponding evaluation index features from the fusion data set, and conduct correlation analysis based on the evaluation index features to form a computing power evaluation feature set; S4. Use the analytic hierarchy process and regional clustering algorithm to configure different initial index weights for different types of regions; S5. Combine the computing power evaluation feature set and the initial index weights, and calculate the initial computing power heat values of each region through a non-linear weighted model; S6. Obtain the actual computing power requirements of each region, and use an optimization algorithm to iteratively optimize the initial index weights to generate optimized index weights; S7. Recalculate the regional computing power heat values based on the optimized index weights, generate a heat map of spatio-temporal distribution, and generate a computing power resource allocation plan according to the heat gradient difference.

[0006] Preferably, the preprocessing of the original data set in step S1 includes: Define data cleaning rules through a rule engine, and clean the original data set according to the data cleaning rules. Among them, the data cleaning rules include outlier correction thresholds, missing value filling strategies, and data integrity scoring criteria; Use the dynamic time warping algorithm to align the cleaned data with different collection periods to a unified reference time axis, and eliminate periodic noise interference based on Fourier transform to generate a standardized data set.

[0007] Preferably, the active entity data includes population data, household data, and enterprise data; the economic data includes GDP data, industrial data, transportation and logistics data, education data, financial data, commercial data, agricultural data, and medical data; the resource and environment data includes power generation data, power structure data, temperature data, and IDC facility data; the policy documents include computing power hub policies, regional planning policies, industrial support policies, and energy management policies.

[0008] Preferably, step S2 includes: S21. For the active entity data, economic data, and resource and environment data in the standardized data set, use the piecewise adaptive normalization method for vectorization processing to convert data with different dimensions into a unified numerical vector; S22. Build a data quality evaluation model, calculate the quality score based on the three dimensions of data integrity, timeliness, and accuracy, determine the weights of each dimension, generate a comprehensive quality score, and use the comprehensive quality score as an adjustment coefficient to weighted-adjust the initial numerical vector to generate a quality-weighted numerical vector; S23. Conduct in-depth semantic analysis on the policy documents, extract policy clauses related to computing power, quantify the policy impact intensity based on the term frequency-inverse document frequency method, and establish an association rule between the policy clauses and each quality-weighted feature vector; S24. Dynamically adjust the vectors affected by the policy in the quality-weighted numerical vector according to the association rule; S25. Align and integrate the adjusted numerical vectors according to the time dimension and the space dimension to generate a unified integrated data set.

[0009] Preferably, step S21 includes: S211. Perform normalization processing on economic data after logarithmic transformation: In the formula, is the economic data after normalization processing; is the original economic data; is the minimum value of the economic data; is the maximum value of the economic data; is the smoothing factor, and its value range is ; S212. Construct a piecewise mapping function to perform normalization processing on the active subject data, and determine the threshold set based on the inflection point analysis of the historical computing power demand curve , where n is the number of thresholds. When the active subject data , the mapping function is as follows: In the formula, is the active subject data after normalization processing; is the linear transformation coefficient; is the bias constant; and are determined by calculating the piecewise endpoint values; S213. Perform non-linear mapping on the resource and environment data: In the formula, is the resource and environment data after normalization processing; is the normalization coefficient; is the non-linear adjustment factor; is the original resource and environment data; is the environmental suitability threshold; S214. Organize all the normalized values of the same data source at the same time point into a numerical vector : In the formula, s is the unique identifier of the data source; t is the time point.

[0010] Preferably, the hierarchical evaluation system is a two-layer evaluation index system: The first-level indicators include computing power demand level, computing power support conditions, and computing power development environment; The second-level indicators include: Under the computing power demand level, there are the density of information technology enterprises, the proportion of Internet enterprises, the number of large and medium-sized enterprises, and the digital coverage of key industries; Under the computing power support conditions, there are the capacity of IDC facilities, the power guarantee ability, the network bandwidth level, and the adaptability of the computer room environment; Under the computing power development environment, there are the proportion of information industry investment, the degree of industrial policy support, the degree of resource element guarantee, and the completeness of infrastructure.

[0011] Preferably, in step S3, the correlation analysis adopts a feature redundancy elimination algorithm based on locality-sensitive hashing. By calculating the Pearson correlation coefficient of the index pairs in the hash bucket and combining with a set threshold, significant features are screened and retained as the computing power evaluation feature set.

[0012] Preferably, in step S5, the calculation formula of the non-linear weighted model is: In the formula, is the initial computing power heat value of the j-th region; is the initial index weight of the k-th item; is the standardized value of the k-th feature in the computing power evaluation feature set of the j-th region; is the non-linear adjustment coefficient of the k-th feature; is the policy influence intensity of the j-th region; is the policy influence coefficient; is the number of features in the computing power evaluation feature set.

[0013] Preferably, the method for generating the heat time-space distribution map includes: using the regional location set as the construction unit and the regional computing power heat value as the threshold to construct an initial Voronoi space structure; dynamically adjusting the computing power heat field structure of adjacent regions based on the time-space propagation function, and the time-space propagation function considers the geographical distance and heat diffusion parameters between regions; generating a time-space distribution map supporting real-time interaction by dynamically updating the Voronoi heat region boundary.

[0014] On the other hand, the present invention also provides a computing power decision-making system based on multi-dimensional data analysis. The system is used to execute the method described in any one of the above, and the system includes: A data acquisition module, which is used to collect active subject data, economic data, resource and environment data, and policy documents, and perform standardized preprocessing to generate a standardized data set; A data fusion module, which is used to perform vectorization processing and heterogeneous fusion on the standardized data set to generate a unified fusion data set; A feature analysis module, which is used to construct a hierarchical evaluation system, extract evaluation index features from the fusion data set, and perform correlation analysis and feature selection to form a computing power evaluation feature set; A regional clustering module, which is used to configure different initial index weights based on the analytic hierarchy process and the regional clustering algorithm; A heat evaluation module, which is used to combine the computing power evaluation feature set and the initial index weight, calculate the initial computing power heat value of the region through a non-linear weighted model, and optimize the index weight according to the output of the weight optimization module to update the computing power heat value; A weight optimization module, which is used to iteratively optimize the initial index weight according to the actual computing power demand; The result display module is used to generate a heat distribution atlas of time and space according to the computing power heat value and output a computing power resource allocation plan.

[0015] The present invention has the following beneficial effects compared with the prior art: By constructing a complete computing power heat evaluation method, the full-process automatic processing from data collection, feature analysis to evaluation decision-making is realized, which improves the systematicness and objectivity of computing power resource evaluation and provides a reliable decision-making basis for regional computing power resource allocation; The vectorization processing and heterogeneous fusion method are adopted to standardize multi-source data, solve the problem that different types of data are difficult to analyze uniformly, and improve the data utilization efficiency and analysis accuracy; Based on the analytic hierarchy process and regional clustering algorithm, the differential initial weights are configured, which overcomes the problem of unreasonable weight configuration in traditional evaluation methods and improves the accuracy and credibility of evaluation results; Through the feature redundancy elimination algorithm, the evaluation indexes are optimized and screened, the data redundancy is reduced, the effectiveness of feature expression is improved, and the evaluation system is made more concise and efficient; The nonlinear weighted model is introduced to calculate the heat value, which effectively depicts the nonlinear influence characteristics of evaluation indexes on computing power heat and improves the fitting degree of the model to the actual situation; Based on the spatio-temporal computing power heat field drawing method of dynamic Voronoi diagram, the real-time dynamic display of heat distribution is realized, and the visualization effect and interaction performance of evaluation results are enhanced. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is the method flow chart of the present invention; Figure 2 It is the technical implementation diagram of the present invention; Figure 3 It is the system framework diagram of the present invention. Detailed Embodiments

[0018] Next, in combination with the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] As Figure 1 shown, the present invention provides a computing power and heat evaluation method based on multi-dimensional data analysis, including: S1. Obtain an original data set containing active subject data, economic data, resource and environmental data, and policy documents, and preprocess the original data set to generate a standardized data set; S2. Perform vectorization processing and heterogeneous fusion on the standardized data set to generate a unified fusion data set; S3. Construct a hierarchical evaluation system, extract corresponding evaluation index features from the fusion data set, and perform correlation analysis based on the evaluation index features to form a computing power evaluation feature set; S4. Use the analytic hierarchy process and regional clustering algorithm to configure different initial index weights for different types of regions; S5. Combine the computing power evaluation feature set and the initial index weights, and calculate the initial computing power and heat values of each region through a non-linear weighted model; S6. Obtain the actual computing power requirements of each region, and use an optimization algorithm to iteratively optimize the initial index weights to generate optimized index weights; S7. Recalculate the regional computing power and heat values based on the optimized index weights to generate a heat time-space distribution map, and generate a computing power resource allocation plan according to the heat gradient difference.

[0020] Please refer to Figure 2, the present invention first collects multi-source heterogeneous data such as active subject data, economic data, resource and environmental data, and policy documents, and realizes data standardization and fusion through vectorization processing; then constructs a hierarchical evaluation system and extracts evaluation index features, and optimizes the feature set by using a feature redundancy elimination algorithm; then configures differential initial weights based on the analytic hierarchy process and regional clustering algorithm, and optimizes the weights according to the actual computing power requirements, and calculates the regional computing power heat value in combination with a non-linear weighted model; finally, realizes the visual display of the heat time-space distribution through a dynamic Voronoi diagram, and forms a complete computing power resource evaluation and decision-making scheme. In the specific implementation process, text vectorization technology is used to perform semantic analysis on policy documents and extract policy orientation features; key evaluation indicators are screened through correlation analysis and principal component analysis methods, and a multi-level index evaluation system is established; the regional clustering algorithm is used to classify evaluation regions to achieve differential weight configuration; a spatio-temporal propagation function is introduced to dynamically adjust the computing power heat field and generate a real-time interactive distribution map. This technical solution realizes the precision of computing power resource evaluation and the scientific nature of decision-making through multi-dimensional data analysis and dynamic optimization mechanism.

[0021] Specifically, in an embodiment of the present invention, the active subject data includes population data, household data, and enterprise data; the economic data includes GDP data, industrial data, transportation and logistics data, education data, financial data, commercial data, agricultural data, and medical data; the resource and environmental data includes power generation data, power structure data, temperature data, and IDC facility data; the policy documents include computing power hub policies, regional planning policies, industrial support policies, and energy management policies.

[0022] In this embodiment, the active subject data is obtained through the population census data, enterprise registration database, and relevant statistical yearbooks released by the national statistical department; the economic data comes from the statistical bulletins, industry development reports, and professional data monitoring platforms released by statistical departments at all levels; the resource and environmental data is mainly obtained from the power industry operation monitoring system, meteorological department observation database, and data center infrastructure census data; the policy documents are obtained through public channels such as policy documents, planning outlines, and industry management measures released on government portals at all levels.

[0023] Specifically, techniques such as web crawler technology, application programming interface (API) calls, database exchanges, and file parsing can be adopted to obtain multi-source heterogeneous data by constructing an automated data acquisition system. For example, statistical data and policy documents on government portals are crawled in real time through distributed web crawler technology, and a smart parsing engine is used to perform structured extraction on statistical reports in various formats such as PDF and Excel; docking with various professional data platforms based on RESTful API interfaces to achieve timed synchronization of economic indicators and environmental data; for data with high real-time requirements (such as power data and meteorological data), a Kafka-based streaming data acquisition mechanism is adopted, and real-time access to data is achieved through a message queue.

[0024] Specifically, in an embodiment of the present invention, the preprocessing of the original data set in step S1 includes: Defining data cleaning rules through a rule engine, and cleaning the original data set according to the data cleaning rules. Among them, the data cleaning rules include outlier correction thresholds, missing value filling strategies, and data integrity scoring criteria.

[0025] Specifically, a rule engine based on the Rete algorithm can be used to define and execute data cleaning rules. This rule engine converts data cleaning rules into executable rule chains by constructing a rule network, realizing dynamic configuration and automated execution of the rules.

[0026] In this embodiment, the outlier correction threshold is calculated using the interquartile range (IQR) method, and data exceeding Q3 + 1.5IQR or below Q1 - 1.5IQR is identified as an outlier; the missing value filling strategy adopts multiple imputation methods, and mean filling, regression filling, or nearest neighbor filling is selected according to data characteristics; the data integrity scoring criteria are comprehensively calculated based on the time coverage rate, space coverage rate, and attribute integrity of the data.

[0027] The dynamic time warping algorithm is used to align the cleaned data with different collection cycles to a unified reference time axis, and periodic noise interference is eliminated based on Fourier transform to generate a standardized data set.

[0028] Specifically, the dynamic time warping algorithm (DTW) realizes elastic matching of data in the time dimension by calculating the optimal warping path between different time series, thereby mapping data with different collection cycles to a unified reference time axis. Then, the fast Fourier transform (FFT) is applied to the aligned data, and periodic noise components are identified and filtered through frequency domain analysis to achieve smooth processing of the data, and finally a standardized data set is generated.

[0029] Specifically, in an embodiment of the present invention, step S2 includes: S21. For the active subject data, economic data, and resource and environment data in the standardized dataset, a piecewise adaptive normalization method is used for vectorization processing to convert data with different dimensions into a unified numerical vector. In this embodiment, due to the significant differences in the distribution characteristics and numerical ranges of different types of data, the piecewise adaptive normalization method can better maintain the original distribution characteristics and relative relationships of the data. Step S21 includes: S211. Perform logarithmic transformation and then normalization on the economic data: In the formula, is the economic data after normalization processing; is the original economic data; is the minimum value of the economic data; is the maximum value of the economic data; is the smoothing factor, and its value range is ; S212. Construct a piecewise mapping function to normalize the active subject data, and determine the threshold set based on the inflection point analysis of the historical computing power demand curve , where n is the number of thresholds. When the active subject data , the mapping function is as follows: In the formula, is the active subject data after normalization processing; is the linear transformation coefficient; is the bias constant; and are determined by calculating the piecewise endpoint values; S213. Perform non-linear mapping on the resource and environment data: In the formula, is the resource and environment data after normalization processing; is the normalization coefficient; is the non-linear adjustment factor; is the original resource and environment data; is the environmental suitability threshold; S214. Organize all the normalized numerical values of the same data source at the same time point into a numerical vector : In the formula, s is the unique identifier of the data source; t is the time point.

[0030] In this embodiment, the active subject data has obvious hierarchical characteristics. Using a piecewise mapping function can better distinguish data characteristics at different levels; economic data generally shows exponential growth characteristics, and through logarithmic transformation, it can be converted into a linear relationship for subsequent analysis; resource and environmental data has non-linear fluctuation characteristics, and using non-linear mapping can better capture the dynamic change law of the data. All the normalized values of the same data source at the same time point are organized into a numerical vector with a fixed dimension according to the preset feature dimension order, ensuring that the vectors at different time points have the same dimension and feature correspondence relationship.

[0031] S22. Construct a data quality evaluation model, calculate the quality score based on three dimensions of data integrity, timeliness, and accuracy, determine the weight of each dimension, generate a comprehensive quality score, and use the comprehensive quality score as an adjustment coefficient to perform weighted adjustment on the initial numerical vector to generate a quality-weighted numerical vector.

[0032] In this embodiment, due to the characteristics of the acquired data, such as multi-source heterogeneity, time-series continuity, and large differences in numerical distribution, integrity, timeliness, and accuracy are selected as the quality evaluation dimensions.

[0033] Among them, integrity is calculated based on the data item coverage rate, and the completeness of the data is measured by the ratio of the actually acquired data volume to the theoretical data volume: In the formula, is the integrity score, and the value range is [0, 1]; is the number of actually acquired data items; is the number of data items that should be acquired as expected.

[0034] Timeliness is based on the principle of exponential decay, and the time delay effect of the data is quantified by setting a time decay coefficient: In the formula, is the timeliness score, and the value range is [0, 1]; is the time decay coefficient, which is used to control the timeliness decay rate; is the data update time; is the benchmark evaluation time.

[0035] Accuracy is based on statistical principles, and the credibility of the data is evaluated by combining the standard deviation and the proportion of outliers: In the formula, is the accuracy score, and the value range is [0, 1]; is the proportion of outliers; is the data standard deviation; is the maximum allowable standard deviation.

[0036] The weight of each dimension is determined by the entropy weight method. By calculating the information entropy, the dispersion degree of each dimension is reflected to achieve objective weight assignment: Wherein, is the weight of the i-th dimension, where i = 1, 2, 3; is the information entropy of the i-th dimension.

[0037] Finally, the weighted scores of the three dimensions are summed to obtain the comprehensive quality score: Wherein, is the comprehensive quality score, and its value range is [0, 1].

[0038] Taking the comprehensive quality score Q as the adjustment coefficient, the initial numerical vector V is adjusted by linear weighting: V' = Q × V, generating a quality-weighted numerical vector.

[0039] This objective weighting method based on data characteristics can dynamically reflect the data quality status, avoid the deviation caused by subjective judgment, and at the same time, through the quality weighting mechanism, realize the automatic down-weighting processing of low-quality data.

[0040] S23. Perform in-depth semantic analysis on the policy document, extract the policy clauses related to computing power, quantify the policy impact intensity based on the term frequency-inverse document frequency method, and establish the association rules between the policy clauses and each quality-weighted feature vector.

[0041] In this embodiment, when performing in-depth semantic analysis on the policy document, the pre-trained BERT model can be used to process it. The input of the model is the original policy document, and the output is the semantic vector representation of the text and the key entity information. Based on the output results of the model, the text fragments related to computing power are extracted by using the topic recognition algorithm, and combined with the term frequency-inverse document frequency (TF-IDF) method, the policy clauses related to computing power are screened out from them. Specifically, by setting the core dictionary and similarity threshold in the computing power field, the text fragments with semantic similarity higher than the threshold are identified as relevant policy clauses.

[0042] For each policy clause j, calculate its policy impact intensity P j : Wherein, represents the term frequency of the keyword in this policy clause, represents the inverse document frequency of the keyword, is the policy level weight coefficient.

[0043] Construct the association intensity model between the policy clause and the quality-weighted feature vector. Let the quality-weighted feature vector be V', and for each feature component k in the vector, calculate its correlation coefficient R with the policy clause j jk : Wherein, is the semantic similarity, obtained by calculating the cosine similarity between the policy clause and the description of the feature component; is the quality score of this feature component; They are the correlation benchmark coefficient and the quality adjustment coefficient respectively.

[0044] Based on the correlation coefficient matrix R, a fusion rule is established: when R jk is greater than the set threshold, the correlation relationship between policy clause j and feature component k is established, and the correlation strength is used as the weight basis for subsequent analysis. This method based on semantic analysis and correlation modeling not only maintains the consistency of the data processing logic with the previous steps but also realizes the quantitative mapping from policy documents to numerical features.

[0045] S24. Dynamically adjust the vectors affected by policies in the quality-weighted numerical vector according to the correlation rule.

[0046] In this embodiment, let the adjustment coefficient matrix at time t be: , where is the identity matrix, is the global adjustment factor, which is used to control the overall intensity of policy impact; is the policy time decay matrix, and its element , is the difference between the policy release time and the current time, is the time decay coefficient; is the correlation coefficient matrix at time t.

[0047] Apply the adjustment coefficient matrix to the quality-weighted numerical vector V' to obtain the adjusted numerical vector V'': V'' t =A t ×V' t , where for the correlation coefficient R jk less than the threshold, the original value remains unchanged.

[0048] S25. Align and integrate the adjusted numerical vector according to the time dimension and the space dimension to generate a unified fusion data set.

[0049] In this embodiment, for the adjusted numerical vector V'' obtained in step S24, time dimension alignment is first performed. Set a unified time granularity Δt (such as day, week, month), and map the data with different timestamps to the standard time point. For time point t, the numerical calculation formula after alignment is: , where is the time offset, , , are the time weighting coefficients and satisfy , which is used to smooth the time series data.

[0050] Then, perform spatial dimension alignment. Based on geographical information coding, establish a spatial mapping function M(x,y) to unify data of different spatial scales into a standard spatial grid: Among them, M(x,y) is a spatial mapping matrix used to handle spatial distribution differences.

[0051] Finally, construct a fusion dataset D, whose structure is: , where T is the set of time indices, S is the set of spatial indices, and v is the aligned feature vector. This alignment and integration method based on spatio-temporal two dimensions realizes the unified expression of heterogeneous data.

[0052] Specifically, in an embodiment of the present invention, step S3 includes: constructing a hierarchical evaluation system, which is a two-layer evaluation index system: The first-level indicators include computing power demand level, computing power support conditions, and computing power development environment; The second-level indicators include: Under the computing power demand level, the density of information technology enterprises, the proportion of Internet enterprises, the number of large and medium-sized enterprises, and the digital coverage of key industries are set; Under the computing power support conditions, the IDC facility capacity, power guarantee ability, network bandwidth level, and adaptability of the computer room environment are set; Under the computing power development environment, the proportion of information industry investment, the degree of industrial policy support, the degree of resource element guarantee, and the completeness of infrastructure are set.

[0053] In this embodiment, the constructed hierarchical evaluation system is based on the internal logic of computing power development and is evaluated from three dimensions: the demand side, the supply side, and the environment side. The computing power demand level reflects the actual demand intensity of the region for computing power resources; the computing power support conditions reflect the basic ability of the region to provide computing power services; the computing power development environment represents the guarantee degree of the sustainable development of the regional computing power industry. This multi-dimensional evaluation system design not only conforms to the law of computing power development but also can comprehensively reflect the regional computing power level.

[0054] For the calculation of each second-level indicator, extract the corresponding features from the fusion dataset: Density of information technology enterprises: Calculate the number of IT enterprises per unit area based on enterprise registration data and industrial classification data; Proportion of Internet enterprises: Statistically calculate the proportion of Internet-related enterprises using enterprise business scope data; IDC facility capacity: Calculate comprehensively by combining data such as the number of racks in the data center and the design load; Power guarantee ability: Obtain by weighting indicators such as power supply stability and load rate; Degree of industrial policy support: Quantitatively calculate based on the analysis results of the policy documents in steps S23-24. And so on for other indicators to form a complete index calculation system.

[0055] In the correlation analysis stage, the Locality-Sensitive Hashing (LSH) algorithm is used to eliminate feature redundancy. First, the feature vectors of n secondary indicators are constructed into an n×m feature matrix F, where m is the number of samples. The execution process of the LSH algorithm is as follows: Design a family of hash functions H: , where a is a random vector, b is a random offset, and r is a bucket width parameter.

[0056] Calculate the Pearson correlation coefficient between feature vectors: , where x and y are feature pairs falling into the same hash bucket.

[0057] Feature screening rule: When (θ is the correlation threshold), retain the feature with larger information content and remove redundant features.

[0058] Through this method, the finally formed computing power evaluation feature set not only maintains the integrity of the evaluation dimensions but also ensures the independence between features.

[0059] Specifically, in an embodiment of the present invention, step S4 includes: Based on the computing power evaluation feature set, use the K-means clustering algorithm to classify regions. During the clustering process, key factors such as the computing power demand characteristics, infrastructure conditions, and industrial development level of the regions are mainly considered, and the regions are divided into different types such as computing power leading type, demand-driven type, and basic support type.

[0060] Then, for different types of regions, use the analytic hierarchy process to determine the relative importance of each level of indicators. When constructing the judgment matrix, fully consider the characteristics of the region type. For example, for regions of the computing power leading type, strengthen the weight of computing power support conditions; for demand-driven type regions, appropriately increase the weight of indicators related to the computing power demand level; for basic support type regions, pay more attention to the weight allocation of the computing power development environment.

[0061] In this embodiment, the K-means clustering algorithm and the analytic hierarchy process used are both existing algorithms, and their specific implementation steps will not be elaborated here.

[0062] Specifically, in an embodiment of the present invention, in step S5, the calculation formula of the non-linear weighted model is: In the formula, is the initial computing power heat value of the j-th region; is the initial index weight of the k-th item; is the standardized value of the k-th feature in the computing power evaluation feature set of the j-th region; is the non-linear adjustment coefficient of the k-th feature; is the policy influence intensity of the j-th region; is the policy influence coefficient; is the number of features in the computing power evaluation feature set.

[0063] Specifically, the model reflects the interaction effects between indicators in a multiplicative form and introduces a policy influence term to achieve a comprehensive quantitative evaluation of the regional computing power development level. In the calculation process, first, the feature values in the computing power evaluation feature set are standardized to ensure the comparability of indicators with different dimensions. The feature standardization adopts the range standardization method to map all feature values to the interval [0, 1]. The non-linear adjustment coefficient is set based on the importance and change sensitivity of features and is determined by a method combining expert experience and data analysis, and is used to adjust the non-linear change characteristics of the contribution degrees of different features. The policy influence intensity comes from the policy document analysis results in step S23-24, and the policy influence coefficient is determined through historical data regression analysis and is used to balance the influence degree of policy factors on the computing power heat value. Using this model, the initial computing power heat values of all regions can be calculated.

[0064] Specifically, in an embodiment of the present invention, in step S6, the optimization algorithm adopts a hybrid optimization algorithm of batch normalization adjustment and momentum gradient descent and introduces a computing power feature adaptive weight mechanism. The iterative process of the optimization algorithm is as follows: (1). Prepare the input data, including the computing power evaluation feature sets of n regions , the initial index weights , k = 1, 2,..., m (m is the number of features), the actual computing power demand value , j = 1, 2,..., n. The feature importance coefficient is determined based on the feature correlation analysis results in S3.

[0065] (2). Initialize the parameters, including the regularization coefficient , the feature importance adjustment factor , the initial learning rate , the attenuation coefficient , the adaptive exponent , the momentum factor and the momentum term vector .

[0066] (3). Execute the iterative optimization process In each round of iteration t, execute: 1. Calculate the computing power heat value under the current weight as the predicted value , and the specific calculation method can still be calculated based on the non-linear weighted model.

[0067] 2. Calculate the error evaluation loss function: Among them, is the importance coefficient of the k-th feature, determined based on the feature correlation analysis result in step S3; is the regularization coefficient; m is the number of features; is the feature importance adjustment factor. In this loss function, the first term is the error term, the second term is the L2 regularization term, and the third term is the feature importance weighting term.

[0068] 3. For each weight Calculate the gradient: where That is, the gradient of the k-th weight.

[0069] 4. Batch normalize the gradient: where is the gradient after batch normalization, is the mean of all feature gradients, is the variance of all feature gradients, is a small constant in batch normalization to prevent division by zero.

[0070] 5. Update the momentum term: where is the momentum term of the k-th feature at the t-th iteration, is the momentum factor.

[0071] 6. Calculate the adaptive learning rate: where is the adaptive learning rate at the t-th iteration.

[0072] 7. Update all weights simultaneously: Stop the iteration when any of the following conditions is met: the change in Loss is less than the threshold for two consecutive rounds ; the maximum number of iterations is reached. The output result is the optimized set of index weights .

[0073] Specifically, in one embodiment of the present invention, step S7 includes: Based on the optimized index weights, recalculate the regional computing power heat value according to the non-linear weighting model, and construct the initial Voronoi spatial structure with the regional position set as the construction unit and the regional computing power heat value as the threshold; dynamically adjust the computing power heat field structure of adjacent regions based on the spatio-temporal propagation function, and the spatio-temporal propagation function considers the geographical distance between regions and the heat diffusion parameter; generate a spatio-temporal distribution map supporting real-time interaction by dynamically updating the Voronoi heat region boundary, and generate a computing power resource allocation plan according to the heat gradient difference.

[0074] Specifically, in this embodiment, the generation process of the heat spatio-temporal distribution map is as follows: A1. Construction of the initial Voronoi space structure. Its input is a set of regional positions , where is the geographical coordinate of the i-th region; the set of computing power and heat values ; for each regional position point , construct its Voronoi polygon : where p is an arbitrary point on the plane, represents the Euclidean distance.

[0075] A2. Dynamic adjustment of the spatio-temporal heat field. Design the spatio-temporal propagation function as: where, is the geographical distance between regions i and j; is the heat diffusion parameter, controlling the heat influence range; is the time fluctuation coefficient, reflecting the periodic change of heat over time; is the time period parameter, controlling the heat change frequency; t is the time variable.

[0076] A3. Superposition calculation of the heat field. The comprehensive heat value of region i at time t: where, is the spatial weight coefficient, , representing the distance attenuation effect; is the basic heat value of region i.

[0077] A4. Dynamic update of the Voronoi region boundary.

[0078] Calculate the heat gradient of adjacent regions:

[0079] Boundary adjustment function:

[0080] where B(i,j) is the dynamic boundary point between regions i and j.

[0081] In this embodiment, the process of generating the computing power resource allocation scheme is as follows: (1) Calculate the heat gradient matrix G:

[0082] (2) Identify high-gradient region pairs: If (gradient threshold), then mark (i,j) as the region pair that needs to be allocated.

[0083] (3) Generate allocation suggestions: For the marked region pair (i,j), calculate the suggested allocation amount: where: is the allocation coefficient, is the maximum single allocation capacity.

[0084] In summary, the computing power and thermal evaluation method proposed by the present invention establishes a complete technical system from data collection, preprocessing, feature extraction to thermal evaluation through multi-dimensional data intelligent analysis and deep integration. This method standardizes and integrates multi-source heterogeneous data such as active subjects, economic development, resource environment, etc., and combines piecewise adaptive normalization and policy semantic analysis to achieve accurate quantification of the regional computing power development level. Through key technologies such as a hierarchical evaluation system, feature selection based on locality-sensitive hashing, dynamic weight optimization, and Voronoi thermal field visualization, this method not only provides objective and reliable computing power evaluation results, but also realizes dynamic monitoring and intelligent allocation of regional computing power distribution, providing a powerful decision-making support tool for promoting regional computing power collaborative development and optimizing resource allocation.

[0085] In addition, as Figure 3 shown, the present invention also provides a computing power decision-making system based on multi-dimensional data analysis for executing any one of the above methods, and the system includes: A data acquisition module, configured to collect active subject data, economic data, resource environment data, and policy documents, and perform standardized preprocessing to generate a standardized data set; specifically, define data cleaning rules using a rule engine, including outlier correction thresholds, missing value filling strategies, and data integrity scoring criteria; align data with different collection cycles to a unified reference time axis through a dynamic time warping algorithm, and use Fourier transform to eliminate periodic noise interference.

[0086] A data fusion module, configured to perform vectorization processing and heterogeneous fusion on the standardized data set to generate a unified fusion data set; specifically, perform vectorization on active subject data, economic data, and resource environment data using a piecewise adaptive normalization method; construct a quality evaluation model based on data integrity, timeliness, and accuracy; use the term frequency-inverse document frequency method to quantify the policy impact intensity, and establish an association rule between policy clauses and feature vectors.

[0087] A feature analysis module, configured to construct a hierarchical evaluation system, extract evaluation index features from the fusion data set, and perform correlation analysis and feature selection to form a computing power evaluation feature set; specifically, construct a two-layer evaluation index system including computing power demand level, computing power support conditions, and computing power development environment; use a feature redundancy elimination algorithm based on locality-sensitive hashing to screen significant features by calculating the Pearson correlation coefficient of index pairs in the hash bucket.

[0088] A regional clustering module, configured to configure differentiated initial index weights based on the analytic hierarchy process and regional clustering algorithms; specifically, determine the index layer weights in combination with the analytic hierarchy process, group regions through clustering algorithms; and configure differentiated initial weight coefficients for different types of regions according to regional development characteristics and computing power demand levels.

[0089] A thermal evaluation module, which is used to combine the computing power evaluation feature set and the initial index weights, calculate the initial computing power heat value of the region through a non-linear weighting model, and optimize the index weights according to the output of the weight optimization module to update the computing power heat value; specifically, based on the non-linear weighting model calculate the heat value; through the feature non-linear adjustment coefficient and the policy influence coefficient to achieve multi-dimensional dynamic evaluation; support real-time update of the heat calculation results according to the optimized weights.

[0090] A weight optimization module, which is used to iteratively optimize the initial index weights according to the actual computing power requirements; specifically, adopt a hybrid optimization algorithm of batch normalization adjustment and momentum gradient descent; introduce a computing power feature adaptive weight mechanism; through the feature importance coefficient and the adaptive learning rate strategy to improve the optimization efficiency.

[0091] A result display module, which is used to generate a heat time-space distribution map according to the computing power heat value and output a computing power resource allocation plan. Specifically, construct a heat field based on the Voronoi space structure; use the time-space propagation function to dynamically adjust the heat field structure of adjacent regions; generate suggestions for computing power resource allocation between regions through heat gradient analysis; support real-time interactive update of the heat map.

[0092] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A computing power thermal evaluation method based on multidimensional data analysis, characterized in that: include: S1. Obtain the original data set containing active subject data, economic data, resource and environmental data, and policy documents, preprocess the original data set, and generate a standardized data set; S2, vectorize and fuse the standardized data set to generate a unified fused data set; S3. Build a hierarchical evaluation system, extract corresponding evaluation index features from the fusion data set, perform correlation analysis based on the evaluation index features, and form a computing power evaluation feature set; S4. Using the analytic hierarchy process and regional clustering algorithm, differentiated initial indicator weights are configured for different types of regions; S5. Combine the computing power evaluation feature set and the initial indicator weights to calculate the initial computing power thermal value of each region through a nonlinear weighted model; S6. Obtain the actual computing power requirements of each region, use the optimization algorithm to iteratively optimize the initial indicator weights, and generate optimized indicator weights; S7. Recalculate the regional computing power thermal value based on the optimized indicator weights, generate a thermal spatiotemporal distribution map, and generate a computing power resource allocation plan based on the thermal gradient differences.

2. According to claim 1, a computing power thermal evaluation method based on multidimensional data analysis is characterized in that: The preprocessing of the original data set in step S1 includes: Define data cleaning rules through the rule engine and clean the original data set according to the data cleaning rules. The data cleaning rules include outlier correction thresholds, missing value filling strategies, and data integrity scoring criteria. The dynamic time warping algorithm is used to align the cleaned data of different acquisition periods to a unified reference time axis, and the periodic noise interference is eliminated based on Fourier transform to generate a standardized data set.

3. According to claim 1, a computing power thermal evaluation method based on multidimensional data analysis is characterized in that: Active subject data include population data, household data, and enterprise data; economic data include GDP data, industrial data, transportation and logistics data, education data, financial data, commercial data, agricultural data, and medical data; resource and environmental data include power generation data, power structure data, temperature data, and IDC facility data; policy documents include computing hub policies, regional planning policies, industrial support policies, and energy management policies.

4. The method for thermal evaluation of computing power based on multidimensional data analysis according to claim 1 is characterized in that: Step S2 includes: S21. For the active subject data, economic data, and resource and environmental data in the standardized data set, the segmented adaptive normalization method is used for vectorization processing to convert data of different dimensions into a unified numerical vector; S22. Construct a data quality assessment model, calculate the quality score based on the three dimensions of data integrity, timeliness and accuracy, determine the weight of each dimension, generate a comprehensive quality score, use the comprehensive quality score as an adjustment coefficient to perform weighted adjustment on the initial numerical vector, and generate a quality weighted numerical vector; S23. Conduct deep semantic analysis on policy documents, extract policy clauses related to computing power, quantify the policy impact intensity based on the word frequency-inverse document frequency method, and establish association rules between policy clauses and each quality-weighted feature vector; S24, dynamically adjusting the vectors affected by the policy in the quality weighted numerical vector according to the association rule; S25. Align and integrate the adjusted numerical vectors according to the time dimension and the space dimension to generate a unified fusion data set.

5. The method for thermal evaluation of computing power based on multidimensional data analysis according to claim 4 is characterized in that: Step S21 includes: S211. Logarithmically transform economic data and then normalize them: In the formula, It is the normalized economic data; It is the original economic data; is the minimum value of economic data; is the maximum value of economic data; is the smoothing factor, and its value range is ; S212: Construct a segmented mapping function to normalize the active subject data, and determine the threshold set based on the inflection point analysis of the historical computing power demand curve. , n is the threshold number, when the active subject data When , the mapping function is as follows: In the formula, It is the normalized active subject data; is the linear transformation coefficient; is the bias constant; and Determined by calculating the segment endpoint values; S213. Perform nonlinear mapping on resource and environmental data: In the formula, It is the resource and environmental data after normalization; is the normalization coefficient; is the nonlinear adjustment factor; It is the original resource and environmental data; is the environmental suitability threshold; S214, organizing all normalized values ​​of the same data source at the same time point into a numerical vector : Where s is the unique identifier of the data source and t is the time point.

6. The method for thermal evaluation of computing power based on multidimensional data analysis according to claim 1 is characterized in that: The hierarchical evaluation system is a two-tier evaluation indicator system: The first-level indicators include computing power demand level, computing power support conditions, and computing power development environment; Secondary indicators include: The computing power demand level includes the density of information technology enterprises, the proportion of Internet enterprises, the number of large and medium-sized enterprises, and the digital coverage of key industries; The computing power support condition is based on the IDC facility capacity, power supply guarantee capability, network bandwidth level, and computer room environment adaptability; The computing power development environment includes the proportion of information industry investment, industrial policy support, resource element guarantee, and infrastructure completeness.

7. The method for thermal evaluation of computing power based on multidimensional data analysis according to claim 1 is characterized in that: In step S3, the correlation analysis adopts a feature redundancy elimination algorithm based on local sensitive hashing, by calculating the Pearson correlation coefficient of the indicator pairs in the hash bucket, combined with the set threshold to filter and retain significant features as the computing power evaluation feature set.

8. The method for computing power thermal evaluation based on multidimensional data analysis according to claim 4 is characterized in that: In step S5, the calculation formula of the nonlinear weighted model is: In the formula, is the initial computing power thermal value of the jth region; is the initial indicator weight of the kth item; The normalized value of the kth feature in the computing power evaluation feature set for the jth region; is the nonlinear adjustment coefficient of the kth feature; is the policy impact intensity of the jth region; is the policy impact coefficient; The number of features in the feature set for computational power evaluation.

9. The method for thermal evaluation of computing power based on multidimensional data analysis according to claim 1, characterized in that: The method for generating the thermal spatiotemporal distribution map includes: constructing an initial Voronoi spatial structure with a set of regional locations as a construction unit and a regional computing power thermal value as a threshold; dynamically adjusting the computing power thermal field structure of adjacent areas based on a spatiotemporal propagation function, wherein the spatiotemporal propagation function takes into account the geographical distance between regions and thermal diffusion parameters; and generating a spatiotemporal distribution map that supports real-time interaction by dynamically updating the Voronoi thermal region boundaries.

10. A computing power decision-making system based on multidimensional data analysis, characterized in that: The system is used to execute the method according to any one of claims 1 to 9, and the system comprises: The data acquisition module is used to collect active subject data, economic data, resource and environmental data, and policy documents, and perform standardized preprocessing to generate standardized data sets; The data fusion module is used to perform vectorization processing and heterogeneous fusion on the standardized data set to generate a unified fused data set; The feature analysis module is used to build a hierarchical evaluation system, extract evaluation indicator features from the fused data set, and perform correlation analysis and feature selection to form a computing power evaluation feature set; The regional clustering module is used to configure differentiated initial indicator weights based on the analytic hierarchy process and regional clustering algorithm; The thermal evaluation module is used to combine the computing power evaluation feature set and the initial indicator weights, calculate the initial computing power thermal value of the region through a nonlinear weighted model, and optimize the indicator weights according to the output of the weight optimization module to update the computing power thermal value; The weight optimization module is used to iteratively optimize the initial indicator weights according to the actual computing power requirements; The result display module is used to generate a thermal spatiotemporal distribution map based on the computing power thermal value and output a computing power resource allocation plan.

Images