A method, system and device for intelligent pricing of a hydropower project, and a storage medium
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Filing Date
- 2026-05-11
- Publication Date
- 2026-07-14
Smart Images
Figure CN122390819A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data analysis technology, and in particular to a method, system, device and storage medium for intelligent pricing of hydropower projects. Background Technology
[0002] In the pricing process of hydropower projects, the pricing work is complicated and involves a very large amount of data, so it is very difficult to start pricing from scratch every time.
[0003] Meanwhile, although many historical projects have undergone similar pricing, the characteristics of hydropower project data—including rich technical parameters, strong correlation of engineering features, a high proportion of unstructured technical descriptions, large data scale, and strong constraints from engineering technical specifications—make it impossible for computer systems to automatically match and calculate. Therefore, existing technologies lack effective means of utilizing hydropower project data. The only option is to rely on manual experience to find similar projects and then manually verify and modify them, which is time-consuming and labor-intensive. There is a lack of a method for intelligent pricing based on existing hydropower project data. Summary of the Invention
[0004] The main purpose of this application is to provide a smart pricing method, system, device and storage medium for hydropower projects, so as to automatically price the current hydropower project based on data accumulated from historical hydropower projects.
[0005] To achieve the above objectives, firstly, this application proposes an intelligent pricing method for hydropower projects, comprising the following steps: Based on a pre-defined tagging system, LLM structured extraction is performed on the target project and the corresponding historical dataset to obtain the target project tag and historical project tag set with a triplet structure. Based on preset multidimensional filtering conditions, the historical project tag set is filtered to obtain a candidate project tag set; Based on time decay factor, semantic similarity factor and data quality factor, dynamic weight calculation is performed on each historical item tag in the candidate item tag set to obtain the dynamic weight set corresponding to the candidate item tag set. An improved moving weighted average model is adopted to price the target project based on the historical price set, price index correction coefficient, and dynamic weight set corresponding to the candidate project label set, thereby obtaining the target feature value corresponding to the target project.
[0006] In one embodiment, the triplet structure includes a tag type, a tag attribute, and an attribute value; The multidimensional filtering conditions include filtering out non-compliant data through hard constraints, threshold filtering through preset range intervals, and selecting a specified number of historical items based on semantic similarity.
[0007] In one embodiment, the method for constructing the tag system includes: Construct black and white labels as hard constraints; Set corresponding error fluctuation ranges for various indicators included in the project as the range range; The project's unstructured text descriptions are transformed into vectors using an embedding model.
[0008] In one embodiment, a method for filtering a historical item tag set based on preset multidimensional filtering conditions to obtain a candidate item tag set includes: First, through hard constraints, non-compliant historical project tags containing black tags are filtered out from the historical project tag set to obtain the first candidate project set; Then, the first candidate item set is filtered by threshold through a preset range, and historical item labels with index errors exceeding the error fluctuation range are removed to obtain the second candidate item set. Finally, the cosine similarity between the target semantic vector of the target project and the candidate semantic vectors corresponding to each project in the second candidate project set is calculated. In the second candidate project set, the top K historical project labels with the highest similarity are selected as the candidate project label set.
[0009] In one embodiment, the time decay factor is constructed based on an exponential decay model, and its expression is as follows: ; in, The difference between the current time and the project preparation time of the i-th project. The attenuation coefficient; The semantic similarity factor is constructed based on a power function nonlinear transformation, and its expression is as follows: ; in, Sensitivity index For the i-th historical project tag in the candidate project tag set, Tag the target project; The data quality factor is constructed based on linear mapping logic, and its expression is as follows: ; in, This is a preliminary assessment score based on the review status of historical projects and the authority of data sources, with a value range of 0 to 1; The dynamic weight set is obtained by calculating the normalized weights of the time decay factor, semantic similarity factor, and data quality factor corresponding to each historical project tag.
[0010] In one embodiment, the price index correction factor is the ratio of the price index of the target item label to the price index of the historical item label, wherein the price index is configured to characterize one or more attributes of the item.
[0011] In one embodiment, when performing LLM structured extraction on the historical dataset of the target item to obtain a set of historical item labels with a triplet structure, if there is a specific modification item in the target item that meets the preset conditions, the LLM estimates the linear adjustment item corresponding to the target item based on the specific modification item and the difference between each historical item. When using an improved moving weighted average model to price the target feature value corresponding to the target project based on the historical price set, price index correction coefficient, and dynamic weight set corresponding to the candidate project label set, the linear adjustment term is used to adjust the target feature value.
[0012] Secondly, to achieve the above objectives, this application also proposes an intelligent pricing system for hydropower projects, comprising: The intelligent input and indicator extraction module is used to perform LLM structured extraction on the target item and the historical dataset corresponding to the target item based on the preset label system, and obtain the target item label and historical item label set with triple structure. The alternative item filtering module is used to filter the historical item tag set based on preset multi-dimensional filtering conditions to obtain the alternative item tag set; The dynamic weight calculation module is used to perform dynamic weight calculation on each historical item tag in the candidate item tag set based on the time decay factor, semantic similarity factor and data quality factor, so as to obtain the dynamic weight set corresponding to the candidate item tag set. The target feature numerical calculation module is used to calculate the target feature value corresponding to the target project by using an improved moving weighted average model, based on the historical price set corresponding to the candidate project label set, the price index correction coefficient, and the dynamic weight set.
[0013] Thirdly, to achieve the above objectives, this application also proposes an intelligent pricing device for hydropower projects, characterized in that the device includes: a memory, a processor, and an intelligent pricing program for hydropower projects stored in the memory and executable on the processor, wherein the intelligent pricing program for hydropower projects is configured to implement the steps of the method as described in any one of the first aspects.
[0014] Fourthly, to achieve the above objectives, this application also proposes a storage medium, characterized in that the storage medium stores a computer program, which, when executed by a processor, implements the method as described in any one of the first aspects.
[0015] This application provides a method, system, device and storage medium for intelligent pricing of hydropower projects. The method achieves automatic pricing based on historical data of hydropower projects through intelligent input and index extraction, multi-dimensional filtering mechanism, dynamic weight calculation and intelligent correction and pricing, shortening the pricing work of several days to minutes and automating 90% of the regular bill of quantities pricing. Historical project data is transformed into the company's core assets. At the same time, by using time decay and price index correction, the problem of outdated historical data is solved, and the valuation is made in line with the current market conditions. The pricing results are interpretable, listing the top historical items with the highest contributions and their weights for cost engineers to review, thus increasing trust. Attached Figure Description
[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart illustrating an embodiment of the intelligent pricing method for hydropower projects in this application. Figure 2 This is a schematic diagram of the structure of an embodiment of the intelligent pricing system for hydropower projects in this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the intelligent pricing device for hydropower projects in this application.
[0019] Explanation of icon numbers: 10. Memory; 20. Processor.
[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0021] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0022] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0023] It should be noted that, due to the characteristics of hydropower project data, such as rich technical parameter dimensions, strong correlation of engineering features, high proportion of unstructured technical descriptions, large data scale and strong constraints of engineering technical specifications, this application adopts an improved moving weighted average calculation method based on standardized triplet label system, LLM structured extraction, multi-dimensional screening, and multi-factor dynamic weighting. This method overcomes the problems of low reuse rate of historical engineering data, low efficiency of manual pricing, insufficient feature matching accuracy, and poor consistency of calculation results in the existing technology, and greatly improves the processing efficiency and accuracy of engineering pricing. This method can also be applied to other engineering data processing scenarios with strong engineering attributes and multi-dimensional structured and unstructured characteristics. It is not a set of artificially formulated pricing rules or quotation methods, but rather a solution to the technical problem that machines cannot automatically process engineering entity feature data through standardized labeling systems, progressive filtering algorithms, and dynamic weighted calculation models. All calculation logic is based on objective natural laws. As long as the engineering entity data is generated based on natural laws, has strong technical constraints, is structured and quantifiable, and its feature associations follow engineering technical laws, the technical solution of this invention can be used to complete automated feature extraction, standardized mapping, accurate matching, and quantitative weighted calculation, demonstrating strong versatility and universality.
[0024] like Figure 1 As shown in the figure, this application provides a smart pricing method for hydropower projects, including the following steps: Step S1: Based on the preset tag system, perform LLM structured extraction on the target item and the historical dataset corresponding to the target item to obtain the target item tag and historical item tag set with triple structure. Step S2: Based on preset multi-dimensional filtering conditions, filter the historical project tag set to obtain a candidate project tag set; Step S3: Based on the time decay factor, semantic similarity factor and data quality factor, perform dynamic weight calculation on each historical item tag in the candidate item tag set to obtain the dynamic weight set corresponding to the candidate item tag set. Step S4: Using an improved moving weighted average model, pricing is performed based on the historical price set corresponding to the candidate item label set, the price index correction coefficient, and the dynamic weight set to obtain the target feature value corresponding to the target item.
[0025] Specifically, the target project includes text data such as core technical parameters and engineering characteristics, compliance and mandatory constraints, and implementation conditions and construction organization. Historical datasets are full-cycle documents of completed hydropower projects of the same type as the target project, including but not limited to final settlement documents, as-built drawings and design change documents, financial review reports, final audit reports, project acceptance certificates, full-process cost consulting documents, main material and equipment procurement settlement vouchers, quota application and pricing details, and other data in various formats. The target feature value is used to characterize the predicted price corresponding to the target project; The target project and its corresponding historical dataset can be input into the preset Prompt project constraint large language model. This model will perform homologous and standardized structured content extraction based on the triplet specification of the preset tag system for the full text data of the target project and the full project documents of the historical dataset. Among them, the Prompt project is configured to pre-inject a tag system dictionary, hydropower engineering industry technical specifications and standardized enumeration rules, which constrain the large language model to extract information only from the original text of the input text, prohibit the fabrication of content not mentioned in the original text, ensure that the extraction results strictly follow the triplet structure of tag type-tag attribute-attribute value / vector, and output a standardized tag set that can be directly used for subsequent multidimensional filtering and dynamic weight calculation; At the same time, for the target project and the historical dataset, the extraction rules and label dimensions are completely consistent to ensure that the two types of label sets are isomorphic and can be matched. Multidimensional filtering conditions can adopt a progressive screening logic from coarse to fine, successively completing the removal of non-compliant data, filtering of numerical indicator ranges, and precise matching based on semantic similarity; When performing dynamic weight calculation, the time decay factor, semantic similarity factor and data quality factor can be calculated separately, and then multiplied and normalized to obtain the dynamic weights corresponding to each candidate item label in the candidate item label set. The improved moving weighted average model first performs exponential subtraction on the historical prices corresponding to each candidate item label, and then sums them according to dynamic weights to obtain the target feature value.
[0026] By adopting this implementation method, the traditional manual pricing work of hydropower projects, which takes several days, can be shortened to minutes, and more than 90% of the routine bill of quantities pricing work can be completed automatically, greatly improving pricing efficiency. At the same time, it can make full use of the dormant data in the enterprise's historical project database, transforming historical projects into reusable core assets of the enterprise and avoiding ineffective work of repeated pricing. By employing multi-dimensional screening and composite weight calculation, this approach addresses industry pain points such as reliance on personal experience, significant deviations in pricing results, and low reuse rates of historical data in traditional manual pricing, thus ensuring the accuracy and consistency of pricing results.
[0027] Furthermore, in an optional implementation, when performing LLM structured extraction on the historical dataset of the target item to obtain a set of historical item labels with a triplet structure, if there is a specific modification item in the target item that meets preset conditions, then the LLM estimates the linear adjustment term corresponding to the target item based on the specific modification item and the difference between each historical item. When using an improved moving weighted average model to price the target feature value corresponding to the target project based on the historical price set, price index correction coefficient, and dynamic weight set corresponding to the candidate project label set, the linear adjustment term is used to adjust the target feature value.
[0028] By combining LLM with an improved moving weighted average model, this approach can solve the technical problem that traditional weighted average pricing methods cannot automatically handle non-standardized customized content such as special processes, geological conditions, and ecological protection constraints not covered by the historical alternative project set in the target project. This enables the solution to extract personalized project characteristics in specific scenarios. Specifically, the preset condition is that the target project's list description and technical requirements include non-standardized customized content such as special processes not covered by the historical alternative projects, differences in geological conditions, efficiency reduction due to altitude, ecological protection constraints, and construction period limitations, which will increase costs. When LLM extracts differences, it pre-binds the current quota library for hydropower projects and industry price adjustment specifications. It uses Prompt to constrain the price adjustment data based only on quota rules and historical differences of the same type, extracts the core differences of specific modification items and alternative historical items, and outputs standardized difference feature labels. When estimating linear adjustment items, priority is given to matching the actual settlement price adjustment coefficients of similar difference characteristics in the historical project database to calculate the adjustment ratio or fixed adjustment value of the corresponding list item; In the absence of matching historical data, calculate compliant adjustment values based on hydropower quota standards; During the final pricing, the linear adjustment item is superimposed on the weighted benchmark price of the corresponding list item to complete the final adjustment of the target feature value, and the calculation basis and source of the adjustment item are output simultaneously.
[0029] This implementation method can solve the problems that traditional weighted average pricing cannot cover the personalized customization needs of projects and that there is no basis for price adjustment for special processes. It realizes the automatic identification and compliant price adjustment of the differences between the target project and historical projects, replacing the complicated work of manually checking differences and calculating price adjustments. Meanwhile, all adjustments are linked to quota specifications and historical settlement data, ensuring the compliance, traceability, and verifiability of the price adjustment results, avoiding the subjectivity and arbitrariness of manual price adjustments, and further improving the accuracy of the pricing results.
[0030] Furthermore, in an optional implementation, the triplet structure includes a tag type, a tag attribute, and an attribute value; The multidimensional filtering conditions include filtering out non-compliant data through hard constraints, threshold filtering through preset range intervals, and selecting a specified number of historical items based on semantic similarity.
[0031] Specifically, the standardized expression for the triplet structure is Label={Type,Key,Value / Vector}, where Type is the label type, used for top-level business domain classification and isolation of labels, and is divided into three categories: black and white labels, range labels, and semantic labels; The Key is a tag attribute that corresponds to a specific feature dimension under the tag type and is used to define the tag's description object; Value / Vector are attribute values, where Value is a scalar value of Boolean, numeric or enumeration type, and Vector is a semantic vector value after unstructured text is converted; Multidimensional filtering conditions can be executed sequentially in the order of hard constraint filtering, range interval filtering, and semantic similarity selection. The output of the preceding filtering is used as the input of the subsequent filtering, forming a progressive filtering chain.
[0032] By adopting this implementation method, the scattered and non-standardized feature information of hydropower projects is transformed into structured tags that can be recognized, retrieved and calculated by machines through a standardized ternary structure, which solves the problems of inconsistent descriptions of different projects, low efficiency of manual verification and inability to automate processing. By using a three-tiered, multi-dimensional filtering approach, we achieved multi-dimensional screening based on compliance, interval matching degree, and semantic similarity. This not only ensured the compliance and matching degree of the candidate items but also provided a high-quality data foundation for subsequent weight calculation and pricing.
[0033] Furthermore, the method for constructing the aforementioned tagging system includes: Construct black and white labels as hard constraints; Set corresponding error fluctuation ranges for various indicators included in the project as the range range; The project's unstructured text descriptions are transformed into vectors using an embedding model.
[0034] By using black and white labels as hard constraints, it is possible to automatically identify and remove non-compliant historical project data containing blacklist labels (such as abolished quotas or abandoned projects that have not passed compliance review), while retaining compliant projects that meet the whitelist requirements. This mechanism enables rapid filtering of historical project data with a single veto, eliminating the need for manual item-by-item verification. It significantly reduces data noise in subsequent screening and calculation, and enhances the system's automated processing capabilities and data security. By embedding models, unstructured texts such as construction methods, geological descriptions, and ecological protection plans are transformed into fixed-dimensional semantic vectors, making natural language descriptions that could not be directly processed by computers into quantifiable and comparable mathematical objects. This transformation supports subsequent precise semantic matching based on cosine similarity, solving the technical challenge of automatically retrieving and matching historical projects due to the diverse and inconsistent technical text descriptions of hydropower projects.
[0035] Specifically, when constructing black and white labels, two types of hard constraints are pre-set: blacklist labels and whitelist labels. Blacklist labels include veto-type characteristics such as abolished quotas, different types of projects, failure to pass compliance review, and abandoned projects. Whitelist labels include compliance access characteristics such as currently valid quotas, projects of the same type in the same river basin, and projects that have passed financial review. When setting the range, for quantifiable core indicators such as installed capacity, dam height, engineering volume, investment scale, construction period, and dam type parameters of hydropower projects, based on the statistical characteristics of historical data of similar projects, set the allowable error fluctuation range for the corresponding indicators. The range can be adapted to a fixed fluctuation ratio or an unbiased range automatically calculated based on the interquartile range method. When transforming semantic vectors, an embedding model is used to convert unstructured text descriptions such as project construction methods, geological descriptions, fish passage facility types, cofferdam structures, and ecological protection plans into fixed-dimensional semantic vectors, which serve as attribute values for semantic tags.
[0036] Using this implementation method, a full-dimensional tagging system covering hard constraint compliance, numerical magnitude features, and unstructured semantic features was constructed, realizing the standardized and full-coverage mapping of the full life cycle features of hydropower projects; The three types of tags are adapted to different filtering and calculation scenarios, enabling both rapid compliance verification and accurate volume and semantic matching, providing a unified and standardized data foundation for the entire process of intelligent pricing.
[0037] Furthermore, methods for filtering historical item tag sets based on preset multi-dimensional filtering conditions to obtain candidate item tag sets include: First, through hard constraints, non-compliant historical project tags containing black tags are filtered out from the historical project tag set to obtain the first candidate project set; Then, the first candidate item set is filtered by threshold through a preset range, and historical item labels with index errors exceeding the error fluctuation range are removed to obtain the second candidate item set. Finally, the cosine similarity between the target semantic vector of the target project and the candidate semantic vectors corresponding to each project in the second candidate project set is calculated. In the second candidate project set, the top K historical project labels with the highest similarity are selected as the candidate project label set.
[0038] Specifically, when performing hard constraint filtering, the preset blacklist and whitelist verification function is called to traverse all items in the historical item tag set. If there is an intersection between the item tag and the blacklist tag, it is directly removed. Only items that do not hit the blacklist and meet the whitelist requirements are retained to form the first candidate item set. For example, if a historical item tag is {Type: "Blacklist", Key: "Status", Value: "Abandoned"}, it means that it belongs to the blacklist and is directly removed. When performing range filtering, for each preset range-type indicator, the indicator value of the corresponding project in the first candidate project set is checked one by one to see if it falls within the allowable error fluctuation range of the target project. Any project whose core indicator exceeds the range is eliminated to form the second candidate project set. For example, the installed capacity indicator has a corresponding fluctuation range of 1000MW±20%. When performing semantic similarity filtering, the target semantic vector of the target item is first generated using an embedding model, such as BGE-M3 or text-embedding-3-large. Then, the cosine similarity between the target semantic vector and the candidate semantic vectors of each item in the second candidate item set is calculated one by one. The similarity value ranges from [0,1]. The closer the value is to 1, the higher the semantic matching degree. The expression is as follows: ; in, For the target semantic vector, Here are the candidate semantic vectors, where pi represents the historical item and puser represents the target item; After sorting items by similarity from highest to lowest, the top K items are selected as the candidate item tag set, where K ranges from 3 to 5. A preset similarity threshold τ is also used. If the similarity of all items is below τ, a message indicating no suitable reference items is output, terminating the pricing process. The expression is: ; S2 is the second candidate item set.
[0039] This implementation adopts a progressive screening logic from coarse to fine and from low cost to high computing power. First, it performs two low-cost operations of hard constraints and range screening to significantly reduce the size of the candidate item set and eliminate items that have no reference value. Then, it performs high-computing semantic vector similarity calculation, which not only greatly reduces the computing power consumption of the system and improves the screening efficiency, but also ensures the matching accuracy between the final candidate items and the target items. Meanwhile, by setting a similarity threshold, price distortion caused by low-matching items being included in the pricing process is avoided, further ensuring the reliability of the pricing results.
[0040] The time decay factor is constructed based on the exponential decay model, and its expression is as follows: ; in, The difference between the current time and the project preparation time of the i-th project. The attenuation coefficient; The semantic similarity factor is constructed based on a power function nonlinear transformation, and its expression is as follows: ; in, Sensitivity index For the i-th historical project tag in the candidate project tag set, Tag the target project; The data quality factor is constructed based on linear mapping logic, and its expression is as follows: ; in, This is a preliminary assessment score based on the review status of historical projects and the authority of data sources, with a value range of 0 to 1; The dynamic weight set is obtained by calculating the normalized weights of the time decay factor, semantic similarity factor, and data quality factor corresponding to each historical project tag.
[0041] Specifically, when calculating the time decay factor, The unit is years, and the decay coefficient ranges from 0.1 to 0.3. It can be automatically fitted and adjusted based on the price volatility of the hydropower industry. The negative exponential function is used to achieve a smooth and continuous decay of the project's reference value over time. The closer the project is to the current time, the closer the time decay factor is to 1, and the higher the reference value. When calculating the semantic similarity factor, the cosine similarity value obtained in the previous steps is used. It is recommended that the sensitivity index be 2. By using power transformation, the weight of high similarity items is amplified and the weight of low similarity items is compressed, thereby enhancing the reference value of highly matched items. When calculating data quality factors, standardized scores in the range of 0 to 1 can be automatically calculated using the entropy weight method based on five quantifiable dimensions: the completion and settlement status of historical projects, the audit level, data integrity, the authority of data sources, and the compliance of quotas. The expression for normalized weights is: ; Among them, what needs to be ensured is .
[0042] This implementation method breaks through the limitation of the traditional moving weighted average model that only considers the time dimension, and incorporates three core dimensions: time decay, semantic matching degree, and data credibility. It constructs a dynamic weighting system with multiple factors, and realizes the accurate quantification of the reference value of historical projects. By using a multiplicative weighting logic, the pricing results are avoided from being affected by projects that are prominent in one dimension but deficient in other dimensions. At the same time, normalization processing ensures the compliance and stability of the weighted calculation, which greatly improves the accuracy of pricing prediction from the core algorithm level.
[0043] Furthermore, in an optional implementation, the price index correction factor is the ratio of the price index of the target item label to the price index of the historical item label, wherein the price index is configured to characterize one or more attributes of the item.
[0044] Specifically, price indices can include sub-indices such as the comprehensive cost index for hydropower construction projects, the labor cost index, the cement price index, the steel price index, the sand and gravel price index, and the machinery cost index; The expression for the price index correction coefficient is: ; Where Inow is the price index of the target project tag, and Ii is the price index of the historical project tag; Let the comprehensive unit price of a certain item in the i-th historical project be... Then the predicted benchmark price for this item in the list is: ; The final expression for the target feature value is: ; in, This is a linear adjustment term.
[0045] By adopting this implementation method, the historical project prices at different compilation times are reduced across periods through the price index correction coefficient. This eliminates the problem of historical price distortion caused by inflation, industry price fluctuations, and changes in the prices of labor and main materials. The historical prices of different periods are uniformly converted into the current price level of the target project, ensuring that the pricing results are fully in line with the current market conditions. At the same time, it can automatically adapt to the corresponding sub-index based on the material and equipment composition of the list items, realizing refined price correction and further improving the accuracy of target feature values.
[0046] like Figure 2 As shown, this application also provides an embodiment of an intelligent pricing system for hydropower projects, including: The intelligent input and indicator extraction module is used to perform LLM structured extraction on the target item and the historical dataset corresponding to the target item based on the preset label system, and obtain the target item label and historical item label set with triple structure. The alternative item filtering module is used to filter the historical item tag set based on preset multi-dimensional filtering conditions to obtain the alternative item tag set; The dynamic weight calculation module is used to perform dynamic weight calculation on each historical item tag in the candidate item tag set based on the time decay factor, semantic similarity factor and data quality factor, so as to obtain the dynamic weight set corresponding to the candidate item tag set. The target feature numerical calculation module is used to calculate the target feature value corresponding to the target project by using an improved moving weighted average model, based on the historical price set corresponding to the candidate project label set, the price index correction coefficient, and the dynamic weight set.
[0047] like Figure 3 As shown, this application also provides an embodiment of a smart pricing device for hydropower projects. The device includes: a memory, a processor, and a smart pricing program for hydropower projects stored in the memory and executable on the processor. The smart pricing program for hydropower projects is configured to implement the steps of the method as described in any of the above-described method embodiments.
[0048] In addition, this application also provides an embodiment of a storage medium storing a computer program that, when executed by a processor, implements the method as described in any of the above-described method embodiments.
[0049] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.
Claims
1. A smart pricing method for hydropower projects, characterized in that, Includes the following steps: Based on a pre-defined tagging system, LLM structured extraction is performed on the target project and the corresponding historical dataset to obtain the target project tag and historical project tag set with a triplet structure. Based on preset multidimensional filtering conditions, the historical project tag set is filtered to obtain a candidate project tag set; Based on time decay factor, semantic similarity factor and data quality factor, dynamic weight calculation is performed on each historical item tag in the candidate item tag set to obtain the dynamic weight set corresponding to the candidate item tag set. An improved moving weighted average model is adopted to price the target project based on the historical price set, price index correction coefficient, and dynamic weight set corresponding to the candidate project label set, thereby obtaining the target feature value corresponding to the target project.
2. The intelligent pricing method for hydropower projects as described in claim 1, characterized in that: The triplet structure includes tag type, tag attribute, and attribute value; The multidimensional filtering conditions include filtering out non-compliant data through hard constraints, threshold filtering through preset range intervals, and selecting a specified number of historical items based on semantic similarity.
3. The intelligent pricing method for hydropower projects as described in claim 2, characterized in that, The method for constructing the aforementioned tag system includes: Construct black and white labels as hard constraints; Set corresponding error fluctuation ranges for various indicators included in the project as the range range; The project's unstructured text descriptions are transformed into vectors using an embedding model.
4. The intelligent pricing method for hydropower projects as described in claim 3, characterized in that, Methods for filtering historical project tag sets based on preset multidimensional filtering conditions to obtain candidate project tag sets include: First, through hard constraints, non-compliant historical project tags containing black tags are filtered out from the historical project tag set to obtain the first candidate project set; Then, the first candidate item set is filtered by threshold through a preset range, and historical item labels with index errors exceeding the error fluctuation range are removed to obtain the second candidate item set. Finally, the cosine similarity between the target semantic vector of the target project and the candidate semantic vectors corresponding to each project in the second candidate project set is calculated. In the second candidate project set, the top K historical project labels with the highest similarity are selected as the candidate project label set.
5. The intelligent pricing method for hydropower projects as described in claim 1, characterized in that: The time decay factor is constructed based on the exponential decay model, and its expression is as follows: ； in, The difference between the current time and the project preparation time of the i-th project. The attenuation coefficient; The semantic similarity factor is constructed based on a power function nonlinear transformation, and its expression is as follows: ； in, Sensitivity index For the i-th historical project tag in the candidate project tag set, Tag the target project; The data quality factor is constructed based on linear mapping logic, and its expression is as follows: ； in, This is a preliminary assessment score based on the review status of historical projects and the authority of data sources, with a value range of 0 to 1; The dynamic weight set is obtained by calculating the normalized weights of the time decay factor, semantic similarity factor, and data quality factor corresponding to each historical project tag.
6. The intelligent pricing method for hydropower projects as described in claim 1, characterized in that: The price index correction factor is the ratio of the price index of the target project label to the price index of the historical project labels, wherein the price index is configured to represent one or more attributes of the project.
7. The intelligent pricing method for hydropower projects as described in claim 1, characterized in that: When performing LLM structured extraction on the historical dataset of the target project to obtain a set of historical project labels with a triplet structure, if there are specific modifications in the target project that meet the preset conditions, the LLM estimates the linear adjustment term corresponding to the target project based on the specific modifications and the differences between each historical project. When using an improved moving weighted average model to price the target feature value corresponding to the target project based on the historical price set, price index correction coefficient, and dynamic weight set corresponding to the candidate project label set, the linear adjustment term is used to adjust the target feature value.
8. A smart pricing system for hydropower projects, characterized in that, include: The intelligent input and indicator extraction module is used to perform LLM structured extraction on the target item and the historical dataset corresponding to the target item based on the preset label system, and obtain the target item label and historical item label set with triple structure. The alternative item filtering module is used to filter the historical item tag set based on preset multi-dimensional filtering conditions to obtain the alternative item tag set; The dynamic weight calculation module is used to perform dynamic weight calculation on each historical item tag in the candidate item tag set based on the time decay factor, semantic similarity factor and data quality factor, so as to obtain the dynamic weight set corresponding to the candidate item tag set. The target feature numerical calculation module is used to calculate the target feature value corresponding to the target project by using an improved moving weighted average model, based on the historical price set corresponding to the candidate project label set, the price index correction coefficient, and the dynamic weight set.
9. A smart pricing device for hydropower projects, characterized in that, The apparatus includes: a memory, a processor, and a smart pricing program for hydropower projects stored in the memory and executable on the processor, the smart pricing program for hydropower projects being configured to implement the steps of the method as described in any one of claims 1 to 7.
10. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.