Data supply evaluation method based on power marketing

Abstract

The invention relates to a business data processing method, in particular to a data supply evaluation method based on power marketing. The method comprises the following steps: collecting part of operation data; carrying out standardization processing on the data to obtain quantitative scores; and taking a weighted average score of the quantitative score to obtain a total score of data supply evaluation. According to the method, quantitative evaluation of the data supply capability, the service customer capability and the application support capability is realized through a quantitative model. According to the principal component analysis method, by extracting the main change direction of each capability related index, dimensionality is effectively reduced, redundant information is reduced, the data interpretation capability is improved, and potential modes and structures can be found conveniently.

Description

Technical Field

[0001] The present invention relates to a commercial data processing method, and in particular to a data supply evaluation method based on power marketing. Background Art

[0002] The power grid company has built an enterprise-level data middle platform to realize the cross-professional data aggregation of provincial companies, and has gradually carried out the construction of application scenarios based on the data middle platform. In order to further give play to the value of power data elements, enhance the service capabilities of the data middle platform, and promote the in-depth application of power marketing data, it is necessary to deepen the construction of digital measurement and marketing, carry out the construction of marketing service support capabilities at all levels, strengthen the quality management and value application of marketing data, organize and carry out the top-level design of marketing digital operations, and propose to build a data supply system.

[0003] By carrying out the work of providing professional marketing data, the power grid company has realized the ability to import massive data, build various data application scenarios, and export data services to the outside world, thus quickly releasing the value of data. In order to further improve the data supply capability and support grassroots business personnel to conveniently use data, a normalized data supply operation has been carried out. In order to ensure the efficient development of data supply operations, it is urgent to build a set of operation evaluation models to improve the data service capability of the data supply platform, improve the efficiency of data application support, avoid repeated data set construction work, and improve work efficiency.

[0004] Invention Method

[0005] The present invention provides a data supply evaluation method based on power marketing, and the specific technical solution includes the following processes:

[0006] S100: Collect some operational data, including: total number of supplied data tables, supplied data volume, total number of unsupplied data tables, unsupplied data volume, collection frequency, collection time, data accuracy, number of services provided, number of services not provided, demand feedback time, demand collection frequency, customer satisfaction, number of supported applications, and proportion of active applications;

[0007] S200: standardize these data to obtain quantitative scores;

[0008] S300: taking a weighted average of the quantitative scores of the seven indicators, namely, the total number of supplied data tables, the amount of supplied data, the total number of unsupplied data tables, the amount of unsupplied data, the collection frequency, the collection time, and the data accuracy, to obtain a quantitative score of the data supply capability;

[0009] The weighted average of the quantitative scores of the five indicators, namely, the number of services provided, the number of services not provided, the time for demand feedback, the frequency of demand collection, and customer satisfaction, is taken to obtain a quantitative score for the customer service capability.

[0010] Take the weighted average of the quantitative scores of the two indicators, the number of supported applications and the proportion of active applications, to obtain the quantitative score of application support capability;

[0011] S400: Take a weighted average of the quantitative scores of data supply capability, customer service capability, and application support capability to obtain a total score for data supply evaluation.

[0012] Preferably, the step S200 adopts principal component analysis method; the specific process includes:

[0013] S210: Make the mean of each indicator 0 and the variance 1;

[0014] For each feature X in the dataset i , the standardized feature Z i The calculation formula is:

[0015]

[0016] where μ i It is feature X i The mean value is calculated as follows:

[0017]

[0018] σ i It is feature X i The standard deviation is calculated as follows:

[0019]

[0020] S220: Calculate the covariance matrix; the formula is as follows:

[0021]

[0022] Where C is the covariance matrix, Z is the standardized data matrix, n is the number of data samples, and Z T is the transpose of the data matrix;

[0023] S230: eigenvalue decomposition;

[0024] Perform eigenvalue decomposition on the covariance matrix to obtain a set of eigenvalues ​​and corresponding eigenvectors; the eigenvalues ​​and eigenvectors are obtained by solving the characteristic equation:

[0025] det(C-λI)=0;

[0026] Among them, det represents the determinant, C is the covariance matrix, λ is the eigenvalue, and I is the identity matrix;

[0027] The eigenvector v corresponding to the eigenvalue λ is obtained by solving the following equation:

[0028] (C-λI)v=0;

[0029] S240: Select the main principal components;

[0030] According to the size of the eigenvalues, arrange them in descending order from large to small and select the explained variance;

[0031] S250: Calculate the principal component score;

[0032] The standardized data is projected onto the selected principal component direction to obtain the principal component score of each sample; the principal component score is obtained by multiplying the data by the eigenvector, reflecting the performance of the data in the new coordinate system, and the calculation formula is as follows:

[0033] Z′=ZV;

[0034] Preferably, the step S250 is followed by the following steps:

[0035] S260: linearly transform the principal component scores and scale them to between 0 and 100;

[0036] The minimum value of the original principal component score is min_score, and the maximum value is max_score. The calculation formula for the scaled score score_scaled is:

[0037]

[0038] Preferably, the step S260 is followed by the following steps:

[0039] S270: Weight optimization: After the principal component analysis is scored, the weights in the weighted average are determined through a data-driven weight optimization process to ensure that the contribution of each principal component meets business requirements; the specific process includes:

[0040] S271: Principal component contribution analysis; Specific process: Calculate the proportion of variance explained by each principal component for each indicator;

[0041] S272: Data-driven weight optimization; specific process:

[0042] The optimization objective function is as follows:

[0043]

[0044] Among them, w j is the weight of the jth principal component, Z ij is the score of the i-th sample on the j-th principal component, PC i is the comprehensive principal component score of the i-th sample;

[0045] Constraints: Ensure that all weights wj satisfy the conditions of non-negativity and sum to 1, that is,

[0046]

[0047] S273: Weight optimization algorithm: To minimize the objective function, an optimization method based on gradient descent is used. The specific steps are as follows:

[0048] Define the objective function: Set the objective function J(w) to be the sum of squares of the difference between the weighted principal component score and the comprehensive principal component score:

[0049]

[0050] Among them, W j is the weight parameter to be optimized and satisfies the constraint W j ≥0;

[0051] Introduce the Lagrangian multiplier λ and construct the Lagrangian function:

[0052]

[0053] Find the gradient of the Lagrangian function L(w,λ) and update the weights according to the gradient descent method:

[0054]

[0055] Among them, α is the learning rate, and the gradient calculation formula is:

[0056]

[0057] Through multiple rounds of iterations, the weight W is continuously updated j Until the objective function J(w) converges, that is, the optimization result meets the given error threshold or reaches the maximum number of iterations; the optimal weights finally obtained are calculated by user weighted average, and the weights can be re-optimized regularly according to changes in business needs.

[0058] The present invention realizes the quantitative evaluation of data supply capability, customer service capability and application support capability through quantitative models. The principal component analysis method effectively reduces the dimension and redundant information by extracting the main change direction of each capability-related indicator, improves the data interpretation ability, and facilitates the discovery of potential patterns and structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of the data supply evaluation method of the present invention.

[0060] Figure 2 These are some actual values ​​of the 14 indicators in the embodiment of the present invention.

[0061] Figure 3 yes Figure 2 The results after data standardization.

[0062] Figure 4 yes Figure 3 The covariance matrix result after the data is calculated.

[0063] Figure 5 yes Figure 4 The calculated eigenvalues ​​of the data.

[0064] Figure 6 yes Figure 4 The computed feature vector of the data.

[0065] Figure 7 yes Figure 6 The principal component scores and the final score after scaling. DETAILED DESCRIPTION

[0066] A data supply evaluation method based on power marketing includes the following process:

[0067] S100: Collect some operational data, including: total number of supplied data tables, supplied data volume, total number of unsupplied data tables, unsupplied data volume, collection frequency, collection time, data accuracy, number of services provided, number of services not provided, demand feedback time, demand collection frequency, customer satisfaction, number of supported applications, and proportion of active applications;

[0068] Total number of supplied data tables, the total number of supplied data of this unit under the supplied directory, unit: sheets; supplied data volume, the amount of supplied data of this unit under the supplied directory, unit: items / GB; Total number of unsupplied data tables, the total number of unsupplied data of this unit under the supplied directory, unit: sheets; unsupplied data volume, the amount of unsupplied data of this unit under the supplied directory, unit: items / GB; Collection frequency, the frequency of data collection, unit: times / month; Collection time, the time required for each data collection, unit: hours; Data accuracy, the accuracy of collected data, unit: %;

[0069] Number of services provided, the number of services provided by the unit under the supply catalog, unit: times; Number of services not provided, the number of services not provided by the unit under the supply catalog, unit: times; Demand feedback time, the time from demand submission to feedback, unit: hours; Demand collection frequency, the frequency of collecting demands, unit: times / month; Customer satisfaction, customer satisfaction with demand feedback and processing, unit: %;

[0070] Number of supported applications, the total number of supported applications per day, unit: pcs; Active application ratio, the active application ratio refers to the proportion of applications called more than 3 times in the past month to the total number of applications, unit: %;

[0071] Specific data such as Figure 2 shown.

[0072] S200: Standardize these data to obtain quantitative scores; the specific process includes:

[0073] S210: Make the mean of each indicator 0 and the variance 1;

[0074] For each feature X in the dataset i , the standardized feature Z i The calculation formula is:

[0075]

[0076] where μ i It is feature X i The mean value is calculated as follows:

[0077]

[0078] σ i It is feature X i The standard deviation is calculated as follows:

[0079]

[0080] Figure 2 The data after normalization is as follows Figure 3 As shown;

[0081] S220: Calculate the covariance matrix; the formula is as follows:

[0082]

[0083] Where C is the covariance matrix, Z is the standardized data matrix, n is the number of data samples, and Z T is the transpose of the data matrix;

[0084] Figure 3 The covariance matrix after data calculation is as follows Figure 4 As shown;

[0085] S230: eigenvalue decomposition;

[0086] Perform eigenvalue decomposition on the covariance matrix to obtain a set of eigenvalues ​​and corresponding eigenvectors; the eigenvalue represents the variance of each principal component, and the eigenvector represents the direction of the principal component; the eigenvalue and eigenvector are obtained by solving the characteristic equation:

[0087] det(C-λI)=0;

[0088] Among them, det represents the determinant, C is the covariance matrix, λ is the eigenvalue, and I is the identity matrix;

[0089] The eigenvector v corresponding to the eigenvalue λ is obtained by solving the following equation:

[0090] (C-λI)v=0;

[0091] Figure 4 After calculating the data, the eigenvalues ​​are as follows Figure 5 As shown, the feature vector is Figure 6 shown.

[0092] S240: Select the main principal components;

[0093] According to the size of the eigenvalues, arrange them in descending order from large to small and select the explained variance; Figure 5 As shown, the first value 3.28206087 is the largest explained eigenvalue, and the first column vector of the feature matrix is ​​selected, which is the largest principal component;

[0094] S250: Calculate the principal component score;

[0095] The standardized data is projected onto the selected principal component direction to obtain the principal component score of each sample; the principal component score is obtained by multiplying the data by the eigenvector, reflecting the performance of the data in the new coordinate system, and the calculation formula is as follows:

[0096] Z′=ZV;

[0097] S260: linearly transform the principal component scores and scale them to between 0 and 100;

[0098] The minimum value of the original principal component score is min_score, and the maximum value is max_score. The calculation formula for the scaled score score_scaled is:

[0099]

[0100] The principal component scores and the final scores after scaling are shown in Figure 6 shown.

[0101] S270: Weight optimization: After the principal component analysis is scored, the weights in the weighted average are determined through a data-driven weight optimization process to ensure that the contribution of each principal component meets business requirements; the specific process includes:

[0102] S271: Principal component contribution analysis; Specific process: Calculate the variance ratio of each principal component to each indicator; Used to preliminarily determine its weight in the weighted average. The principal component with a higher explained variance ratio should have a greater impact on the comprehensive score;

[0103] S272: Data-driven weight optimization; Specific process: Further optimize weight distribution to achieve balanced contribution to the principal components and improve the business relevance of weighted results; The optimization objectives are as follows:

[0104] Objective function: Minimize the difference between the weighted principal component score and the original score, so that the optimized weighted score can more accurately reflect the contribution of each business indicator; the specific objective function is as follows:

[0105]

[0106] Among them, w j is the weight of the jth principal component, Z ij is the score of the i-th sample on the j-th principal component, PC i is the comprehensive principal component score of the i-th sample;

[0107] Constraints: Ensure that all weights wj satisfy the conditions of non-negativity and sum to 1, that is,

[0108]

[0109] S273: Weight optimization algorithm: To minimize the objective function, an optimization method based on gradient descent is used. The specific steps are as follows:

[0110] Define the objective function: Set the objective function J(w) to be the sum of squares of the difference between the weighted principal component score and the comprehensive principal component score:

[0111]

[0112] Among them, W j is the weight parameter to be optimized and satisfies the constraint W j ≥0;

[0113] Introducing Lagrange multipliers: In order to meet the weight constraints, the Lagrange multiplier λ is introduced to construct the Lagrange function:

[0114]

[0115] Gradient descent update: Find the gradient of the Lagrangian function L(w,λ) and update the weights according to the gradient descent method:

[0116]

[0117] Among them, α is the learning rate, and the gradient calculation formula is:

[0118]

[0119] Iterative optimization: Through multiple rounds of iterations, the weight W is continuously updated jUntil the objective function J(w) converges, that is, the optimization result meets the given error threshold or reaches the maximum number of iterations;

[0120] Dynamic update mechanism: The optimal weight is finally obtained by user weighted average calculation, and the weight can be regularly re-optimized according to changes in business needs;

[0121] S300: taking a weighted average of the quantitative scores of the seven indicators, namely, the total number of supplied data tables, the amount of supplied data, the total number of unsupplied data tables, the amount of unsupplied data, the collection frequency, the collection time, and the data accuracy, to obtain a quantitative score of the data supply capability;

[0122] The weighted average of the quantitative scores of the five indicators, namely, the number of services provided, the number of services not provided, the time for demand feedback, the frequency of demand collection, and customer satisfaction, is taken to obtain a quantitative score for the customer service capability.

[0123] Take the weighted average of the quantitative scores of the two indicators, the number of supported applications and the proportion of active applications, to obtain the quantitative score of application support capability;

[0124] S400: Take the weighted average of the quantitative scores of data supply capability, customer service capability, and application support capability to obtain the total score of data supply evaluation. The specific steps are as follows:

[0125] Define the objective function for the total evaluation score: the quantitative scores of “data supply capability”, “customer service capability” and “application support capability” are S 供给 , S 客户 and S 应用 ; According to business needs, define the weights of each capability in the total score as w 供给 、w 客户 and w 应用 ; satisfy w 供给 +w 客户 +w 应用 =1; the calculation formula of the total evaluation score Stotal is as follows:

[0126] S 总 =w 供给 ×S 供给 +w 客户 ×S 客户 +w 应用 ×S 应用 ,

[0127] Weight optimization: In order to determine the optimal weight combination so that the total evaluation score can accurately reflect the impact of each capability on the overall operational effect, a data-driven weight optimization method is used to select w 供给 、w 客户 and w 应用 The optimal value of ; the optimization process is as follows:

[0128] Set the total score minimization error objective function to improve the business relevance of the weighted total score. For example, by minimizing the deviation between each capability score and the actual business goal, the optimization objective function is:

[0129] min((S 供给 -αS 总 ) 2 +(S 客户 –βS 总 ) 2 +(S 应用 -γs 总 ) 2 ),

[0130] Among them, α, β, and γ are parameters related to business priority and are used to balance the relative importance of various capabilities;

[0131] Introduce the Lagrangian multiplier λ to ensure that the weights satisfy the constraint that the sum is 1; construct the Lagrangian function:

[0132]

[0133] By gradient descent, adjust w 供给 、w 客户 、w 应用 value, so that the Lagrangian function reaches the optimal solution, thereby obtaining the optimal weight combination of the weighted sum.

Claims

1. A data supply evaluation method based on power marketing, characterized in that: The process includes the following: S100: Collect some operational data, including: total number of supplied data tables, supplied data volume, total number of unsupplied data tables, unsupplied data volume, collection frequency, collection time, data accuracy, number of services provided, number of services not provided, demand feedback time, demand collection frequency, customer satisfaction, number of supported applications, and proportion of active applications; S200: standardize these data to obtain quantitative scores; S300: taking a weighted average of the quantitative scores of the seven indicators, namely, the total number of supplied data tables, the amount of supplied data, the total number of unsupplied data tables, the amount of unsupplied data, the collection frequency, the collection time, and the data accuracy, to obtain a quantitative score of the data supply capability; The weighted average of the quantitative scores of the five indicators, namely, the number of services provided, the number of services not provided, the time for demand feedback, the frequency of demand collection, and customer satisfaction, is taken to obtain a quantitative score for the customer service capability. Take the weighted average of the quantitative scores of the two indicators, the number of supported applications and the proportion of active applications, to obtain the quantitative score of application support capability; S400: Take a weighted average of the quantitative scores of data supply capability, customer service capability, and application support capability to obtain a total score for data supply evaluation.

2. According to claim 1, a data supply evaluation method based on power marketing is characterized in that: The step S200 adopts principal component analysis method; The specific process includes: S210: Make the mean of each indicator 0 and the variance 1; For each feature X in the dataset i , the standardized feature Z i The calculation formula is: where μ i It is feature X i The mean value is calculated as follows: σ i It is feature X i The standard deviation is calculated as follows: S220: Calculate the covariance matrix; the formula is as follows: Where C is the covariance matrix, Z is the standardized data matrix, n is the number of data samples, and Z T is the transpose of the data matrix; S230: eigenvalue decomposition; Perform eigenvalue decomposition on the covariance matrix to obtain a set of eigenvalues ​​and corresponding eigenvectors; the eigenvalues ​​and eigenvectors are obtained by solving the characteristic equation: det(C-λI)=0; Among them, det represents the determinant, C is the covariance matrix, λ is the eigenvalue, and I is the identity matrix; The eigenvector v corresponding to the eigenvalue λ is obtained by solving the following equation: (C-λI)v=0; S240: Select the main principal components; According to the size of the eigenvalues, arrange them in descending order from large to small, and select the principal component with the largest explained variance; S250: Calculate the principal component score; The standardized data is projected onto the selected principal component direction to obtain the principal component score of each sample; the principal component score is obtained by multiplying the data by the eigenvector, reflecting the performance of the data in the new coordinate system, and the calculation formula is as follows: Z′=ZV.

3. According to claim 2, a data supply evaluation method based on power marketing is characterized in that: The S250 is followed by the following steps: S260: linearly transform the principal component scores and scale the scores to between 0 and 100; The minimum value of the original principal component score is min_score, and the maximum value is max_score. The calculation formula for the scaled score score_scaled is:

4. The data supply evaluation method based on power marketing according to claim 3 is characterized in that: The S260 is followed by the following steps: S270: Weight optimization: After the principal component analysis is scored, the weights in the weighted average are determined through a data-driven weight optimization process to ensure that the contribution of each principal component meets business requirements; the specific process includes: S271: Principal component contribution analysis; Specific process: Calculate the proportion of variance explained by each principal component for each indicator; S272: Data-driven weight optimization; specific process: The optimization objective function is as follows: Among them, w j is the weight of the jth principal component, Z ij is the score of the i-th sample on the j-th principal component, PC i is the comprehensive principal component score of the i-th sample; Constraints: Ensure that all weights wj satisfy the conditions of non-negativity and sum to 1, that is, S273: Weight optimization algorithm: To minimize the objective function, an optimization method based on gradient descent is used. The specific steps are as follows: Define the objective function: Set the objective function J(w) to be the sum of squares of the difference between the weighted principal component score and the comprehensive principal component score: Among them, W j is the weight parameter to be optimized and satisfies the constraint W j ≥0; Introduce the Lagrangian multiplier λ and construct the Lagrangian function: Find the gradient of the Lagrangian function L(w,λ) and update the weights according to the gradient descent method: Among them, α is the learning rate, and the gradient calculation formula is: Through multiple rounds of iterations, the weight W is continuously updated j Until the objective function J(w) converges, that is, the optimization result meets the given error threshold or reaches the maximum number of iterations; The optimal weights are finally calculated by user weighted average, and the weights can be re-optimized regularly according to changes in business needs.

5. The data supply evaluation method based on power marketing according to claim 1 is characterized in that: The S400 includes the following specific processes: S410: Define the objective function for the total evaluation score; the quantitative scores of "data supply capability", "customer service capability" and "application support capability" are S 供给 , S 客户 and S 应用 ; Define the weights of each ability in the total score as w 供给 、w 客户 and w 应用 ; satisfy w 供给 +w 客户 +w 应用 =1; the calculation formula of the total evaluation score Stotal is as follows: S 总 =w 供给 ×S 供给 +w 客户 ×S 客户 +w 应用 ×S 应用 ; S420: Weight optimization: In order to determine the optimal weight combination so that the total evaluation score can accurately reflect the impact of each capability on the overall operational effect, a data-driven weight optimization method is used to select w 供给 、w 客户 and w 应用 The optimal value of ; the optimization process is as follows: The error minimization objective function of the total score is set to improve the business relevance of the weighted total score.

6. The data supply evaluation method based on power marketing according to claim 5 is characterized in that: In S420, by minimizing the deviation between each capability score and the actual business goal, the optimization objective function is: min((S 供给 -αS 总 ) 2 +(S 客户 -βS 总 ) 2 +(S 应用 -γS 总 ) 2 ), Among them, α, β, and γ are parameters related to business priority and are used to balance the relative importance of various capabilities.

7. The data supply evaluation method based on power marketing according to claim 5 is characterized in that: In S420, by minimizing the deviation between each capability score and the actual business goal, a Lagrangian function is constructed: By gradient descent, adjust w 供给 、w 客户 、w 应用 value, so that the Lagrangian function reaches the optimal solution, thereby obtaining the optimal weight combination of the weighted sum.

Images