Electric power field digital project cost evaluation method based on impact factors

Abstract

The invention discloses a power field digital project cost review method based on impact factors, and belongs to the technical field of digital project cost review. The invention provides a solution for the problem that the score of the influence factor with floating score interval cannot be accurately evaluated in the practical application of intelligent review by using the model. Project data are automatically analyzed through big data analysis, a machine learning algorithm and other technical means, the cost is accurately predicted, potential risks are found in time, human errors are effectively avoided, resource allocation is optimized, a more scientific and objective cost basis is provided for decision making of power enterprises, deep development of digital transformation in the power field is promoted, and the economic benefits of the power enterprises are improved. And double improvement of economic benefits and social benefits is realized. According to the method, objectivity, scientificity and accuracy of a digital project cost review result in the electric power field are improved, and the working efficiency of digital project cost review can be improved.

Description

Technical Field

[0001] The invention belongs to the technical field of digital project cost review, and in particular relates to a digital project cost review method in the electric power field based on influencing factors. Background Art

[0002] With the rapid development of cutting-edge technologies such as smart grids, big data analysis and artificial intelligence, the power industry is also experiencing unprecedented changes in project management, operation and maintenance, and investment decisions. In particular, the review of digital project costs is not only directly related to the cost control, economic benefits and sustainable development of the project, but is also one of the key factors in measuring the success of the project.

[0003] Traditional cost review methods mostly rely on manual review, which has inherent defects such as low efficiency, strong subjectivity, and difficulty in standardization. As the scale of projects continues to expand and the complexity increases, traditional review methods have been unable to cope with the processing needs of massive data and complex review standards, resulting in long review cycles and difficulty in ensuring accuracy, which in turn affects the decision-making efficiency and implementation effect of the project.

[0004] Therefore, a new technical solution is urgently needed in the prior art to solve this problem. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a digital project cost review method in the power field based on influencing factors to solve the technical problem that manual cost review methods are difficult to cope with the processing requirements of massive data and complex review standards, resulting in long review cycles and poor accuracy.

[0006] The method for evaluating the cost of digital projects in the power field based on influencing factors includes the following steps, and the following steps are performed in sequence:

[0007] S1: Select a pre-trained NLP semantic understanding model to learn the vocabulary, grammar, semantics, and syntax of the power system for text data related to the power system, and form a dedicated model in the power knowledge field;

[0008] S2: Manually preliminarily process the impact factor evaluation standard given by the power system. The evaluation standard includes the impact factor name, impact factor description, impact factor evaluation standard, and specific evaluation level of the impact factor. At the same time, each level is also accompanied by the conditions for the project to meet the level and the score value of the level;

[0009] The impact factor is divided into fixed impact factor and floating impact factor according to the scoring method of the evaluation criteria: if the score given in the evaluation criteria has only one definite value, then the impact factor is a fixed impact factor; if at least one of the evaluation criteria gives a score in the form of an interval, then the impact factor is a floating impact factor;

[0010] Since the score of the floating impact factor contains an interval form, the evaluation score outputted by directly inputting it into the NLP semantic understanding model is inaccurate and of little reference value. Therefore, the floating impact factor is manually refined into multiple refined impact factors, and evaluation criteria and corresponding scores are formulated for each refined impact factor. At the same time, the text screening method and corresponding paragraphs and keywords of each fixed impact factor and each refined impact factor are formulated, and the proposed content is formed into a document and then input into the NLP semantic understanding model;

[0011] The specific steps for the impact factor score review are:

[0012] S21: If the impact factor is a fixed impact factor, directly execute S22; if the impact factor is a floating impact factor, select a refined impact factor, and manually determine the order of classification and refinement operations. At the same time, formulate evaluation criteria for each refined impact factor, give a corresponding score, and then execute S22;

[0013] S22: If the impact factor is a fixed impact factor, a corresponding search is performed on the evaluation criteria of the fixed impact factor to obtain the text screening method and corresponding paragraphs and keywords of the corresponding information of the impact factor, which are used to extract relevant text information from the feasibility study report and index the impact factor score;

[0014] If the impact factor is a floating impact factor, the evaluation criteria of each refined impact factor of the floating impact factor are searched accordingly to obtain the text screening method and corresponding paragraphs and keywords of the corresponding information of the impact factor, which are used to extract relevant text information from the feasibility study report and index the scores corresponding to each refined impact factor;

[0015] S3: Select the historical data that has been manually evaluated, and according to the corresponding screening method, for each impact factor evaluation standard, extract the corresponding information from the feasibility study report that has been reviewed. The impact factors include fixed impact factors and various detailed impact factors;

[0016] S4: The evaluation criteria processed in S3, their corresponding information, and the corresponding cost scores obtained through manual review and calculation are used to construct a data set. The data set has two columns: the first column is the evaluation criteria and review text of the impact factor, and the second column is the review result of the impact factor, i.e., the score and review basis. This data set is used to train the NLP semantic understanding model to build a vertical field model for intelligent review of digital projects in the power field.

[0017] S5: Based on the project feasibility report input by the user and the corresponding proposed evaluation criteria, an intelligent review is conducted using the vertical field big model of the intelligent review of digital projects in the power field to obtain the score of each influencing factor. The detailed influencing factors are integrated into the corresponding floating influencing factor scores through the mapping relationship calculation formula to obtain the score calculated for this reference standard in the digital project cost. The score is multiplied by the unit score cost reference value of the given evaluation standard to obtain the cost reference of the project on this influencing factor. All cost references are summed to obtain the final cost reference value given by the constructed intelligent review system for the project.

[0018] The specific method of manually determining the order of grading and refining the review steps is as follows:

[0019] If a floating impact factor is directly classified through the evaluation criteria given by the power system, the order of classification first and then refinement and grading is adopted, that is, classification is first carried out according to the given classification, and then the refined impact factors are graded, and corresponding evaluation criteria and scores are formulated for each grade of refined impact factors; for the floating impact factor with ambiguous and contradictory value standards in the operation order of classification first and then refinement and grading, the order of refinement first and then grading is adopted;

[0020] A floating impact factor is derived from different degrees or grades through the evaluation criteria given by the power system. The order of first refinement and then grading is adopted, that is, each refined impact factor is graded first, and corresponding evaluation criteria and scores are formulated for each grade of refined impact factors. The floating impact factor scores obtained by the mapping relationship calculation formula in step S5 are used to determine which degree or grade of the floating impact factors to which these refined impact factors belong in the evaluation criteria.

[0021] The mapping relationship calculation formulas include the following three:

[0022] The first mapping relationship is the basic mapping relationship, and the calculation formula is:

[0023]

[0024] Among them, s is the score calculated by the reference standard in the digital project cost, S max is the highest score for this evaluation criterion, S min is the minimum score of the evaluation criterion, n is the number of refined impact factors of the evaluation criterion, and w i is the score of the i-th refined impact factor, w max，j is the highest score of the jth refined impact factor;

[0025] The first mapping relationship is a basic mapping relationship, which can be used for the mapping calculation of floating impact factor scores without special adjustment requirements;

[0026] The second mapping relationship calculation formula is:

[0027] The low-density adjustment formula makes the final score biased towards the low range:

[0028]

[0029] Where m is the density adjustment parameter, which is a natural number greater than 1, and is generally 2. The larger m is, the more the final calculated score tends to be in the low range. Indicates w i The mth power, Indicates w max，j m to the power of;

[0030] The second mapping relationship calculation formula is used for the case where the floating impact factor is adjusted at a low density based on the score range given by the floating impact factor evaluation criteria, so that the final score is biased towards the low range of the score range given by the evaluation criteria:

[0031] The third mapping relationship calculation formula is:

[0032]

[0033]

[0034] Wherein a, b, ... are the boundary values ​​in each level of the given data of the floating impact factor, and s′ is the intermediate variable, i.e., the value calculated according to the basic mapping relationship;

[0035] The third mapping relationship calculation formula is used for the case where the floating impact factor is adjusted within an interval to deal with the problem of gaps in the given floating impact factor interval.

[0036] Select historical data that has been manually evaluated, and according to the corresponding screening method, extract the corresponding information from the feasibility study report for each influencing factor evaluation criterion.

[0037] The step S3 extracts the corresponding information from the feasibility study report that has been reviewed, which means that the feasibility study report is deformatted using the Python program, and the text information associated with the impact factor is screened out according to the text screening method and corresponding paragraphs and keywords formulated in S22. At the same time, the corresponding review results are collected to construct a data set.

[0038] The step S4 comprises:

[0039] S41: Split the dataset into training set, validation set and test set;

[0040] S42: Using the training set to train the NLP semantic understanding model, to obtain a trained vertical domain model for intelligent review of digital projects in the power field;

[0041] S43: using the verification set to verify the trained vertical domain big model of intelligent review of digital projects in the power field, and obtain a verification result;

[0042] S44: Testing the verification result using the test set to obtain a test result;

[0043] S45: Determine whether the test result meets the preset conditions. If so, output the trained vertical domain big model for intelligent review of digital projects in the power field as a new vertical domain big model for intelligent review of digital projects in the power field; otherwise, return to step S41.

[0044] The NLP semantic understanding model includes: an input module, a feature calculation module, a fusion module, and an output module;

[0045] The input module is used to receive input text data and convert the text into a numerical vector representation through word embedding. The vector of each word is usually obtained through a pre-trained word embedding model. At the same time, since the Transformer itself does not have the ability to process sequence order, the input module helps the model identify the position of each word in the input sequence by adding position encoding;

[0046] The feature calculation module is used to perform multi-layer calculations based on the self-attention mechanism of Transformer. The module includes multiple calculation layers, each of which is composed of a self-attention layer and a feedforward neural network, and all use the self-attention mechanism to extract contextual relationships and high-level semantic features until a deep semantic representation of the text is finally generated;

[0047] The fusion module is used to fuse the features of each layer obtained from the feature calculation module. The fusion module includes multiple fusion layers, which gradually fuse the features of each layer to ensure that the final feature representation covers all key information of the input text;

[0048] The output module processes the final output of the fusion module to generate the final result of the task. The result output by the model consists of two parts: the review result, i.e., the score of the impact factor, and the review basis.

[0049] Through the above design scheme, the present invention can bring the following beneficial effects:

[0050] The present invention proposes a solution to the problem that the scores of influencing factors with floating score intervals cannot be accurately assessed in the actual application of intelligent review using models. The present invention automatically analyzes project data through technical means such as big data analysis and machine learning algorithms, accurately predicts costs, promptly discovers potential risks, effectively avoids human errors, and optimizes resource allocation, providing a more scientific and objective cost basis for the decision-making of power companies, thereby promoting the in-depth development of digital transformation in the power sector and achieving a dual improvement in economic and social benefits. The present invention not only improves the objectivity, scientificity, and accuracy of the cost review results of digital projects in the power sector, but also improves the efficiency of the cost review work for digital projects.

[0051] The experimental results show that the intelligent evaluation method for the digital project cost in the power field provided by the present invention can accurately and efficiently complete the scientific and accurate evaluation of the digital project cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The present invention is further described below with reference to the accompanying drawings and specific embodiments:

[0053] Figure 1 It is a flow chart of the method for evaluating the cost of digital projects in the electric power field based on influencing factors of the present invention;

[0054] Figure 2 It is the overall operation diagram of the system used in the digital project cost review method in the electric power field based on the influencing factor of the present invention;

[0055] Figure 3 The present invention is a flowchart of S2 for clarifying the order of classification and refinement operations in the digital project cost review method in the power field based on influencing factors of the present invention. DETAILED DESCRIPTION

[0056] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0057] Example

[0058] The present invention provides a method for evaluating the cost of digital projects in the electric power field based on influencing factors. Figure 1 As shown, the specific steps are as follows:

[0059] S1: Select a pre-trained NLP semantic understanding model to learn the text data related to the power system to build a special model in the field of power knowledge; this embodiment uses the ERNIE model as the pre-trained NLP semantic understanding model. The ERNIE model is optimized based on the BERT model, and performs particularly well in Chinese natural language processing tasks. In order to adapt the pre-trained NLP model to the needs of the power knowledge field, it needs to be fine-tuned, that is, continue to be trained on professional text data related to the power system. The input data should include various documents of the power system, such as feasibility study reports, equipment manuals, technical specifications, operating procedures, scientific research papers, etc. These texts contain a large number of professional terms, operation descriptions, and technical knowledge, which are an important source for the model to understand the semantics of the power system.

[0060] In the embodiment, it is first necessary to collect the original data of the text related to the power system, that is, to collect the feasibility study reports, equipment manuals, technical specifications, operating procedures, scientific research papers, etc. in the field of power systems; then, according to various text formats, the above texts are subjected to data cleaning processing, and some paragraphs with poor relevance to the knowledge field of power systems are deleted. The above operations can be implemented either manually or using program code. The processed data is input as training data into the pre-trained semantic understanding model for learning, and a semantic understanding model that is more accurate for the analysis and understanding of power system copywriting can be obtained.

[0061] S2: Manual preliminary processing of the given impact factor evaluation criteria, including: selecting appropriate review methods, determining the method and corresponding parameters for extracting effective information from the feasibility study report for each criterion.

[0062] For fixed impact factors, the model can easily understand this content and only needs to determine the screening method and screening parameters for it.

[0063] Taking the "disaster recovery level" evaluation standard (see the table below for details) as an example, this influencing factor is a fixed influencing factor and can be directly input into the model:

[0064]

[0065] For floating impact factors, the evaluation results are evaluation indicators of floating intervals, with relatively brief explanations and vague evaluation criteria. The results are often inaccurate and unstable if only these standards are input into the model. In order to obtain more valuable results, it is necessary to manually refine such impact factors, refer to the actual situation of the project, and split the impact factors into refined impact factors based on past review experience.

[0066] The refined impact factor can be considered as a more specific expression of a floating impact factor in the project. The floating impact factor score obtained according to the mapping relationship calculation formula in step S5 is used to determine the degree or level of the floating impact factor to which these refined impact factors belong in the evaluation criteria, so that the evaluation results of the refined impact factors that are easy to evaluate after refinement can effectively map the evaluation results of the floating impact factors.

[0067] The refinement process also requires manual determination of the order of classification and refinement operations based on the refinement operation analysis results, and the formulation of evaluation criteria for each refinement influencing factor and the assignment of corresponding scores. The specific flow chart is as follows: Figure 3 shown.

[0068] Taking the evaluation standard of "system construction difficulty" (see the table below for specific information) as an example, this embodiment refines this influencing factor into 6 refined influencing factors: system scale, functional complexity, data structure complexity, technical novelty, resource investment, and integration complexity. As for the classification and refinement operation sequence, "new construction", "improvement", and "reconstruction" in this standard are relatively easy to quantify or judge information, and can be directly judged from part of the original text using the model. The operation of first grading and then refining should be selected.

[0069]

[0070] For each refined impact factor of "system construction difficulty", a text screening method, screening keywords, evaluation criteria and corresponding scores are formulated. The selected text screening method should be able to filter out text information in the project description that is highly related to the evaluation criteria based on the screening keywords. The evaluation criteria and corresponding score formulation can be formulated and adjusted according to the importance of the refined impact factor in the floating impact factor.

[0071] A feasible refinement of the floating impact factor of "system construction difficulty" is as follows:

[0072]

[0073]

[0074] Since "system construction difficulty" is a floating impact factor that needs to be classified and refined, the evaluation criteria of the floating impact factor are written in the first column. In actual applications, the order of various types of information can be changed according to the program reading method.

[0075] Taking "user activity" (see the table below for specific information) as an example, this embodiment refines this influencing factor into 9 refined influencing factors: the number of registered users of the system, the number of concurrent users of the system, the duration of use, the frequency of use, the frequency of interaction, the complexity of system functions / processes, the user coverage rate, the low-frequency function points, and the high-frequency function points. For the classification and refinement operation sequence, the value selection standard is relatively vague and contradictory (there is no strong correlation between the system user activity and the number of low-frequency function points, and it is not easy to divide the two into different levels). Performing the grading operation first may constrain the final result to an inappropriate interval, resulting in inaccurate results. For the judgment of "user activity", first refine the score, then select the file according to the score, and the final judgment result is more reliable.

[0076] A feasible refinement of the floating impact factor of "user activity" is as follows:

[0077]

[0078] A text screening method, screening keywords, evaluation criteria and corresponding scores are proposed for each detailed influencing factor of "user activity". A feasible processing result of the floating influencing factor of "user activity" is as follows:

[0079]

[0080] Similar to the "difficulty of system construction", in actual applications, the order of various types of information can be changed according to the way the program reads.

[0081] The scores described in this section are currently manually drafted because, although the semantic understanding model has strong language processing capabilities, it has a certain lag in responding to rapid changes in review conditions and real-time policy releases. Model training and fine-tuning take time, so it cannot immediately adapt to new policy requirements or changes in scoring criteria. Keeping the manual scoring operation ensures the accuracy and flexibility of the scoring so that it can respond to policy updates in a timely manner and make effective assessments of new conditions.

[0082] S3: Select historical data that has been manually evaluated, and according to the corresponding screening method, for each impact factor evaluation standard, extract the corresponding information from the feasibility study report; according to the corresponding screening method, for each impact factor evaluation standard, extract the corresponding information from the feasibility study report, including: extracting the corresponding information from the feasibility study report that has been reviewed, which means using Python and other programs to deformat the feasibility study report, and formulating text screening methods and corresponding paragraphs, keywords, etc. to filter out text information associated with the impact factor, and appropriately supplement the review basis based on the historical evaluation results to construct a data set.

[0083] The structure of the constructed data set has two columns: the first column is the evaluation criteria and evaluation text of the impact factor, and the second column is the evaluation result (i.e. score) and evaluation basis of the impact factor.

[0084] S4: The above dataset is constructed as a training set, and the NLP semantic understanding model is trained using the training set to build a vertical field model for intelligent review of digital projects in the power field;

[0085] The step S4 comprises:

[0086] S41: combining the evaluation criteria given or proposed in step S21 with the relevant documents and their corresponding answers extracted from the feasibility study report in step S3 to form a data set;

[0087] S42: Split the dataset into training set, validation set and test set;

[0088] S43: Using the training set to train the basic semantic understanding model, to obtain a trained vertical domain model for intelligent review of digital projects in the power field;

[0089] S44: using the verification set to verify the trained vertical domain big model of intelligent review of digital projects in the power field, and obtain a verification result;

[0090] S45: Testing the verification result using the test set to obtain a test result;

[0091] S46: Determine whether the test result meets the preset conditions. If so, output the trained vertical domain big model for intelligent review of digital projects in the power field as a new vertical domain big model for intelligent review of digital projects in the power field; otherwise, return to step S42.

[0092] Optionally, the data segmentation result includes a training set, a validation set and a test set. In the present invention, the first 70% of the new power field related data is mainly used as a training set, the middle 20% as a validation set, and the last 10% as a test set.

[0093] S5: Based on the input project feasibility report and processed assessment criteria, an intelligent review is conducted using the vertical field big model of the intelligent review of digital projects in the power sector to obtain a series of assessment scores and related basis.

[0094] Through program coding, the feasibility study report to be reviewed is processed in step S3 to construct a review reference data set, which is input into the trained model in step S4, and the model will output the review results and review basis. According to the review results, the score of the impact factor can be indexed.

[0095] Optionally, the mapping relationship calculation formulas for integrating the refined impact factors into floating impact factor scores include the following three:

[0096] The first mapping relationship is the basic mapping relationship, and the calculation formula is:

[0097]

[0098] Among them, s is the score calculated by the reference standard in the digital project cost, S max is the highest score for this evaluation criterion, S min is the minimum score of the evaluation criterion, n is the number of refined impact factors of the evaluation criterion, and w i is the score of the i-th refined impact factor, w max，j is the highest score of the jth refined impact factor;

[0099] The first mapping relationship is a basic mapping relationship, which can be used for the mapping calculation of floating impact factor scores without special adjustment requirements;

[0100] The second mapping relationship calculation formula is:

[0101] The low-density adjustment formula makes the final score biased towards the low range:

[0102]

[0103] Where m is the density adjustment parameter, which is a natural number greater than 1, and is generally 2. The larger m is, the more the final calculated score tends to be in the low range. Indicates w i The mth power, Indicates w max，j m to the power of;

[0104] The second mapping relationship calculation formula is used for the case where the floating impact factor is adjusted at a low density based on the score range given by the floating impact factor evaluation criteria, so that the final score is biased towards the low range of the score range given by the evaluation criteria:

[0105] The third mapping relationship calculation formula is:

[0106]

[0107]

[0108] Wherein a, b, ... are the boundary values ​​in each level of the given data of the floating impact factor, and s′ is the intermediate variable, i.e., the value calculated according to the basic mapping relationship;

[0109] The third mapping relationship calculation formula is used for the case where the floating impact factor is adjusted within an interval to deal with the problem of gaps in the given floating impact factor interval.

[0110] Select historical data that has been manually evaluated, and according to the corresponding screening method, extract the corresponding information from the feasibility study report for each influencing factor evaluation criterion.

[0111] After the above steps, for each influencing factor, its score can be obtained. The cost reference of the project on the influencing factor can be obtained by taking the product of the score and the unit score cost reference value of the given evaluation standard. The final cost reference value given by the intelligent evaluation system constructed in this paper for the project can be obtained by summing up all the cost references.

[0112] Of course, in the subsequent specific application process, the new vertical model of intelligent review of digital projects in the power field will not remain static and be applied all the time, but will be dynamically adjusted according to different user input data, including but not limited to adjusting the refined review criteria, the screening methods and parameters of the review criteria, the scores of the review criteria, etc., so that the new vertical field model of intelligent review can more accurately recommend relevant content in the power field, thereby meeting user needs.

[0113] In a case of application of the present invention, the final result obtained by intelligent review was 93% consistent with the manual evaluation standard. For some evaluation standards with floating evaluation results, the method of the present invention can also give scientific and reasonable results and correspondingly give reasonable and objective explanations.

[0114] The beneficial results of intelligent review of the cost of digital projects in the power sector are mainly reflected in significantly improving the efficiency and accuracy of the review. Through big data analysis, machine learning algorithms and other technical means, project data can be automatically analyzed, costs can be accurately predicted, potential risks can be discovered in a timely manner, and human errors can be effectively avoided. At the same time, resource allocation can be optimized to provide a more scientific and objective cost basis for the decision-making of power companies, thereby promoting the in-depth development of digital transformation in the power sector and achieving a dual improvement in economic and social benefits. The experimental results show that the intelligent review method for the cost of digital projects in the power sector provided by the present invention can accurately and efficiently complete the scientific and accurate evaluation of the cost of digital projects.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The cost evaluation method of digital projects in the power sector based on impact factors is characterized by: The process includes the following steps, which are performed in sequence: S1: Select a pre-trained NLP semantic understanding model to learn the vocabulary, grammar, semantics, and syntax of the power system for text data related to the power system, and form a dedicated model in the power knowledge field; S2: Manually preliminarily process the impact factor evaluation standard given by the power system. The evaluation standard includes the impact factor name, impact factor description, impact factor evaluation standard, and specific evaluation level of the impact factor. At the same time, each level is also accompanied by the conditions for the project to meet the level and the score value of the level; The impact factor is divided into fixed impact factor and floating impact factor according to the scoring method of the evaluation criteria: if the score given in the evaluation criteria has only one definite value, then the impact factor is a fixed impact factor; if at least one of the evaluation criteria gives a score in the form of an interval, then the impact factor is a floating impact factor; Since the score of the floating impact factor contains an interval form, the evaluation score outputted by directly inputting it into the NLP semantic understanding model is inaccurate and of little reference value. Therefore, the floating impact factor is manually refined into multiple refined impact factors, and evaluation criteria and corresponding scores are formulated for each refined impact factor. At the same time, the text screening method and corresponding paragraphs and keywords of each fixed impact factor and each refined impact factor are formulated, and the proposed content is formed into a document and then input into the NLP semantic understanding model; The specific steps for the impact factor score review are: S21: If the impact factor is a fixed impact factor, directly execute S22; if the impact factor is a floating impact factor, select a refined impact factor, and manually determine the order of classification and refinement operations. At the same time, formulate evaluation criteria for each refined impact factor, give a corresponding score, and then execute S22; S22: If the impact factor is a fixed impact factor, a corresponding search is performed on the evaluation criteria of the fixed impact factor to obtain the text screening method and corresponding paragraphs and keywords of the corresponding information of the impact factor, which are used to extract relevant text information from the feasibility study report and index the impact factor score; If the impact factor is a floating impact factor, the evaluation criteria of each refined impact factor of the floating impact factor are searched accordingly to obtain the text screening method and corresponding paragraphs and keywords of the corresponding information of the impact factor, which are used to extract relevant text information from the feasibility study report and index the scores corresponding to each refined impact factor; S3: Select the historical data that has been manually evaluated, and according to the corresponding screening method, for each impact factor evaluation standard, extract the corresponding information from the feasibility study report that has been reviewed. The impact factors include fixed impact factors and various detailed impact factors; S4: The evaluation criteria processed in S3, their corresponding information, and the corresponding cost scores obtained through manual review and calculation are used to construct a data set. The data set has two columns: the first column is the evaluation criteria and review text of the impact factor, and the second column is the review result of the impact factor, i.e., the score and review basis. This data set is used to train the NLP semantic understanding model to build a vertical field model for intelligent review of digital projects in the power field. S5: Based on the project feasibility report input by the user and the corresponding proposed evaluation criteria, an intelligent review is conducted using the vertical field big model of the intelligent review of digital projects in the power field to obtain the score of each influencing factor. The detailed influencing factors are integrated into the corresponding floating influencing factor scores through the mapping relationship calculation formula to obtain the score calculated for this reference standard in the digital project cost. The score is multiplied by the unit score cost reference value of the given evaluation standard to obtain the cost reference of the project on this influencing factor. All cost references are summed to obtain the final cost reference value given by the constructed intelligent review system for the project.

2. The method for evaluating the cost of digital projects in the electric power field based on impact factors according to claim 1 is characterized by: The specific method of manually determining the order of grading and refining the review steps is as follows: If a floating impact factor is directly classified through the evaluation criteria given by the power system, the order of classification first and then refinement and grading is adopted, that is, classification is first carried out according to the given classification, and then the refined impact factors are graded, and corresponding evaluation criteria and scores are formulated for each grade of refined impact factors; for the floating impact factor with ambiguous and contradictory value standards in the operation order of classification first and then refinement and grading, the order of refinement first and then grading is adopted; A floating impact factor is derived from different degrees or grades through the evaluation criteria given by the power system. The order of first refinement and then grading is adopted, that is, each refined impact factor is graded first, and corresponding evaluation criteria and scores are formulated for each grade of refined impact factors. The floating impact factor scores obtained by the mapping relationship calculation formula in step S5 are used to determine which degree or grade of the floating impact factors to which these refined impact factors belong in the evaluation criteria.

3. The method for evaluating the cost of digital projects in the electric power field based on impact factors according to claim 1 or 2 is characterized in that: The mapping relationship calculation formulas include the following three: The first mapping relationship is the basic mapping relationship, and the calculation formula is: Among them, s is the score calculated by the reference standard in the digital project cost, S max is the highest score for this evaluation criterion, S min is the minimum score of the evaluation criterion, n is the number of refined impact factors of the evaluation criterion, and w i is the score of the i-th refined impact factor, w max,j is the highest score of the jth refined impact factor; The first mapping relationship is a basic mapping relationship, which is used for the mapping calculation of floating impact factor scores without special adjustment requirements; The second mapping relationship calculation formula is: The low-density adjustment formula makes the final score biased towards the low range: Where m is the density adjustment parameter, which is a natural number greater than 1, and is generally 2. The larger m is, the more the final calculated score tends to be in the low range. Indicates w i The mth power, Indicates w max,j m to the power of; The second mapping relationship calculation formula is used for the case where the floating impact factor is adjusted at a low density based on the score range given by the floating impact factor evaluation criteria, so that the final score is biased towards the low range of the score range given by the evaluation criteria: The third mapping relationship calculation formula is: Where a, b, ... are the boundary values ​​of each level of the given data of the floating impact factor, and s′ is an intermediate variable, that is, a value calculated according to the basic mapping relationship; The third mapping relationship calculation formula is used for the case where the floating impact factor is adjusted within an interval to deal with the problem of gaps in the given floating impact factor interval; Select historical data that has been manually evaluated, and according to the corresponding screening method, extract the corresponding information from the feasibility study report for each influencing factor evaluation criterion.

4. The method for evaluating the cost of digital projects in the electric power field based on impact factors according to claim 1 is characterized by: The step S3 extracts the corresponding information from the feasibility study report that has been reviewed, which means that the feasibility study report is deformatted using the Python program, and the text information associated with the impact factor is screened out according to the text screening method and corresponding paragraphs and keywords formulated in S22. At the same time, the corresponding review results are collected to construct a data set.

5. The method for evaluating the cost of digital projects in the electric power field based on impact factors according to claim 1 is characterized by: The step S4 comprises: S41: Split the dataset into training set, validation set and test set; S42: Using the training set to train the NLP semantic understanding model, to obtain a trained vertical domain model for intelligent review of digital projects in the power field; S43: using the verification set to verify the trained vertical domain big model of intelligent review of digital projects in the power field, and obtain a verification result; S44: Testing the verification result using the test set to obtain a test result; S45: Determine whether the test result meets the preset conditions. If so, output the trained vertical domain big model for intelligent review of digital projects in the power field as a new vertical domain big model for intelligent review of digital projects in the power field; otherwise, return to step S41.

6. The method for evaluating the cost of digital projects in the electric power field based on impact factors according to claim 1 is characterized by: The NLP semantic understanding model includes: an input module, a feature calculation module, a fusion module, and an output module; The input module is used to receive input text data and convert the text into a numerical vector representation through word embedding. The vector of each word is usually obtained through a pre-trained word embedding model. At the same time, since the Transformer itself does not have the ability to process sequence order, the input module helps the model identify the position of each word in the input sequence by adding position encoding; The feature calculation module is used to perform multi-layer calculations based on the self-attention mechanism of Transformer. The module includes multiple calculation layers, each of which is composed of a self-attention layer and a feedforward neural network, and all use the self-attention mechanism to extract contextual relationships and high-level semantic features until a deep semantic representation of the text is finally generated; The fusion module is used to fuse the features of each layer obtained from the feature calculation module. The fusion module includes multiple fusion layers, which gradually fuse the features of each layer to ensure that the final feature representation covers all key information of the input text; The output module processes the final output of the fusion module to generate the final result of the task. The result output by the model consists of two parts: the review result, i.e., the score of the impact factor, and the review basis.

Images