Search

CN-122022727-A - Intelligent analysis system for adaptive multidimensional bidding quotation of rail transit projects

CN122022727ACN 122022727 ACN122022727 ACN 122022727ACN-122022727-A

Abstract

The invention discloses an intelligent analysis system for adaptive multidimensional bidding quotation of rail transit projects, which relates to the technical field of intelligent bidding quotation, and comprises a multidimensional data acquisition module, an intelligent quotation module and an intelligent quotation module, wherein the multidimensional data acquisition module is used for acquiring historical bidding data of the rail transit projects, analyzing attributes and quotation structures of the rail transit projects, carrying out group division and behavior label definition on competitors to generate high-dimensional composite feature vectors of historical bidding associated context-owner tendency-competition tendency of the rail transit projects, the probability prediction module is used for establishing a bid winning probability prediction model of the rail transit projects by taking external market environment parameters as compensation factors and evaluating bid winning enterprises of the rail transit projects, and the intelligent quotation module is used for constructing a multi-objective optimization function and dynamically generating an optimal quotation scheme set by combining multi-dimensional resource constraint conditions of the enterprises and expected profit targets of the projects based on the evaluation results of the bid winning enterprises of the rail transit projects. The invention improves the comprehensiveness of the bidding analysis of the rail transit project.

Inventors

  • ZHU RUYI
  • Tang Xunan

Assignees

  • 中车新型基础设施投资开发有限公司

Dates

Publication Date
20260512
Application Date
20260129

Claims (10)

  1. 1. An intelligent analysis system for adapting to multidimensional bidding offers of rail transit projects, comprising: The system comprises a multidimensional data acquisition module, a probability prediction module and an intelligent quotation module; the intelligent quotation module is electrically connected with the probability prediction module; The multidimensional data acquisition module acquires historical bidding data of the track traffic project, analyzes the attribute and quotation structure of the track traffic project, performs group division and behavior label definition on competitors, and generates a high-dimensional composite feature vector of the historical bidding associated context-owner trend-competition trend of the track traffic project; The probability prediction module is used for establishing a winning probability prediction model of the track traffic project based on a high-dimensional composite feature vector of historical bidding associated context-owner tendency-competition tendency of the track traffic project, taking external market environment parameters as compensation factors, and evaluating winning enterprises of the track traffic project in high probability; The intelligent quotation module is used for constructing a multi-objective optimization function based on the high-probability bid-winning enterprise evaluation result of the track traffic project and combining the multi-dimensional resource constraint condition of the enterprise and the expected income target of the project, and dynamically generating an optimal quotation scheme set.
  2. 2. The intelligent analysis system for adaptive multidimensional rail transit project bid quotation according to claim 1, wherein the multidimensional data acquisition module comprises: The data set establishing unit is used for extracting the past 5-10 years track traffic project bidding historical data through an API (application program interface) connecting tool based on the enterprise ERP system, the bidding management platform and the public bidding platform, cleaning the data, removing repeated items, processing missing values, unifying data formats and aligning time stamps to obtain the track traffic project bidding historical data set after data preprocessing; The track traffic project bid historical data set comprises project basic attributes such as line types, mileage, construction period requirements and technical standard grades, owner information such as unit properties and historical project preferences, and competitor bid records such as quotation, technical scheme and bid winning conditions.
  3. 3. The intelligent analysis system for adaptive multidimensional rail transit project bid quotation of claim 2, wherein the multidimensional data acquisition module further comprises: The enterprise bidding behavior feature matrix building unit is used for calculating the average bidding discount rate, the bidding fluctuation rate and the budget deviation degree of an enterprise in the time of historical data based on a rail transit project bidding historical data set, extracting the annual bidding times, the bidding density and the project type concentration degree of the enterprise, calculating the adoption rate of an enterprise high-technology scheme and the application frequency of a patent technology, analyzing the frequent competitor set and the bidding time selection preference of the enterprise, carrying out Z-score standardization processing on the frequent competitor set and the bidding time selection preference, and building the enterprise bidding behavior feature matrix; The abnormal bidding behavior enterprise list acquisition unit is used for calculating the distance from each behavior data point of the matrix to the kth nearest neighbor point of the matrix by using a DBSCAN (density-based spatial clustering algorithm) based on an enterprise bidding behavior feature matrix, arranging the k-distance graph of the enterprise bidding behavior according to descending order, selecting the distance value at the curve inflection point of the k-distance graph of the enterprise bidding behavior as the optimal field radius, setting the minimum point h in the neighborhood, calculating the number of neighbors in the radius of each enterprise field, marking the enterprises with the neighbor number smaller than the minimum point in the neighborhood as abnormal points, judging that the bidding anomaly is lower than the high similarity of a plurality of enterprise bidding behaviors and the space aggregation is the abnormal type, and obtaining the abnormal bidding behavior enterprise list; The low-dimensional feature vector generation unit calculates cosine similarity among enterprises of abnormal bidding behaviors, establishes an enterprise similarity matrix, obtains an enterprise adjacency matrix and a degree matrix, performs normalization processing on the enterprise adjacency matrix and the degree matrix to obtain an enterprise Laplace matrix, performs feature value decomposition on the enterprise Laplace matrix, calculates feature values and feature vectors of the enterprise Laplace matrix, screens feature vectors corresponding to the first A minimum feature values, and generates low-dimensional space feature vectors of the enterprise; The label unit takes the enterprise low-dimensional space feature vector as input, sets the competition behavior label category as aggressive type, conservative type and robustness, randomly selects K initial centroids, distributes each enterprise low-dimensional space feature data point to the nearest centroids to generate K clusters, calculates the new centroids of each cluster, and repeatedly updates until the centroids are not changed any more, and divides the competition groups and the corresponding behavior labels of each enterprise.
  4. 4. The intelligent analysis system for adaptive multidimensional rail transit project bid quotation of claim 3, wherein the multidimensional data acquisition module further comprises: the generation unit of the document-level semantic feature vector of the bidding text in the track traffic field divides the bidding text in the track traffic field into segments with the length not exceeding 512token as input, substitutes the segments into a BERT pre-training model, acquires the semantic coding vector of each segment of the bidding text in the track traffic field, aggregates all segment representations through an attention pooling layer, and generates the document-level semantic feature vector of the bidding text in the track traffic field; The track traffic project relation diagram establishing unit is used for establishing a track traffic project relation diagram by taking enterprises, track traffic projects and owners as nodes and taking bidding relations, historical cooperation and attribution relations as edges, calculating historical bidding rates, average discount rates and technical score averages of the enterprise nodes in the bidding historical data of the track traffic projects, calculating the scores of the mileage, budget amount and technical complexity of the track projects, and evaluating the scores of the fund adequacy of the nodes of the owners and risk preference indexes; The technical complexity score comprises a technical standard grade, a construction period urgent coefficient and a novel technical application duty ratio, the fund adequacy score comprises owner unit properties, a historical project payment record and annual investment budget, and the risk preference index comprises historical bid number, tendency to bid at the lowest price and technical scheme weight; The method comprises the steps of designing different edge type message transfer functions, carrying out internal aggregation on similar nodes in a track transaction project relation diagram, aggregating heterogeneous nodes, carrying out iterative propagation and aggregation through a 3-layer diagram convolution neural network to obtain an embedded representation of each node obtained by fusing multi-hop neighborhood information, and extracting all nodes to be finally embedded to serve as track traffic project owner trend-competition trend feature vectors; The high-dimensional composite feature vector acquisition unit is used for splicing the bidding text document-level semantic feature vector and the rail transit project owner trend-competition trend feature vector in the rail transit project field, and performing dimension reduction and fusion through a full-connection layer to obtain the high-dimensional composite feature vector of the history bidding associated context-owner trend-competition trend of the rail transit project.
  5. 5. The intelligent analysis system for adaptive multidimensional rail transit project bid quotation of claim 4, wherein the probability prediction module comprises: The bidding decision depth coding feature extraction unit takes a high-dimensional composite feature vector of historical bidding association context-owner tendency-competition tendency of a rail transit project as an input feature, designs a Transformer encoder framework, encodes the input feature through an embedded layer to obtain dense representation of the input feature, adds position codes to capture position information of the input feature, constructs a multi-head self-attention mechanism, calculates attention head query matrixes, key matrixes and value matrixes, captures complex dependency relations in the input feature through zooming dot product attention, performs layer normalization and residual connection after each attention sub-layer, performs nonlinear transformation through a feedforward neural network, stacks N layers structurally, and takes the last layer as bidding decision depth coding feature to output.
  6. 6. The intelligent analysis system for adapting multi-dimensional rail transit project bid offers of claim 5, wherein the probability prediction module further comprises: The external market environment parameter definition unit is used for collecting material price data sequences of past 6-month track traffic projects, calculating rolling standard deviation to obtain material price fluctuation rate, acquiring contemporaneous LPR interest rate and industry financing cost, carrying out weighted average calculation on the material price fluctuation rate and the industry financing cost, and defining the material price fluctuation rate and the material price fluctuation rate as external market environment parameters; The weighted external market environment factor vector generation unit takes the enterprise node historical bid rate in the track traffic project bid history data as a reference, compares the enterprise node historical bid rate with external market environment parameters, performs standardization processing, calculates correlation coefficients of the enterprise node historical bid rate and the external market environment parameter, takes average correlation coefficients of the two data points to perform descending arrangement, and gives the external market environment parameter weight coefficient to generate the weighted external market environment factor vector.
  7. 7. The intelligent analysis system for adapting multi-dimensional rail transit project bid offers of claim 6, wherein the probability prediction module further comprises: The bid winning probability prediction unit takes bid decision depth coding as input, takes weighted external market environment factor vectors as compensation factors, takes actual bid winning results in historical projects as supervision signals, trains gradient lifting decision trees, optimizes super parameters of learning rate, maximum tree depth and leaf node number, establishes a bid winning probability prediction model of track traffic projects, and takes bid winning probability predicted by each bidding enterprise as output; The evaluation unit is used for setting a high probability threshold value of 0.7, screening to obtain a winning bid enterprise of the track traffic item high probability, calculating a winning bid enterprise confidence interval of the track traffic item high probability according to cross verification, and obtaining enterprise names, winning bid probability of track traffic item prediction and corresponding confidence intervals.
  8. 8. The intelligent analysis system for adaptive multidimensional rail transit project bid quotation of claim 7, wherein the intelligent quotation module comprises: the enterprise resource constraint matrix building unit is used for calculating the maximum bearable yield of a new bidding project as productivity constraint according to the current production capacity and resource allocation of an enterprise, analyzing cash flow and dominant capital of the enterprise, combining project cushioning requirements, determining the maximum bearable quotation upper limit of a single project, taking the maximum bearable quotation upper limit as capital constraint, evaluating the prior art capacity and personnel allocation of the enterprise, ensuring that the technical requirements of the bidding project are met, generating a technical feasibility identifier, taking the technical feasibility identifier as technical resource constraint, integrating the productivity constraint, the capital constraint and the technical resource constraint, and building an enterprise resource constraint matrix; the risk tolerance coefficient calculation unit takes enterprise names, the winning probability of track traffic project prediction, corresponding confidence intervals, quotations and risk coefficients as project expected income targets, obtains each enterprise-project pair multi-target decision feature package, combines an enterprise resource constraint matrix, establishes an enterprise risk tolerance model, and takes the risk tolerance coefficient of an enterprise under the current project condition as output; the quotations comprise enterprise historical quotation discount rates and cost structure estimation, and the risk coefficients comprise project technology complexity scores and owner risk preferences.
  9. 9. The intelligent analysis system for adapting to multi-dimensional rail transit project bid offers of claim 8, wherein the intelligent offer module further comprises, internally: And the pareto optimal front calculation unit is used for randomly generating a group of quotation schemes as initial population individuals in a quotation feasible interval of each winning enterprise with high probability, taking maximized winning probability, maximized expected benefit and maximized enterprise risk tolerance coefficient as multi-objective optimization functions, calculating objective function values corresponding to each individual as fitness, comparing the performances of all individuals in the population on three optimization targets, layering the winning enterprise individuals according to non-dominant levels, calculating the crowding degree distance of each winning enterprise individual, identifying and dividing pareto non-dominant levels, layering all individuals according to dominant relations, and enabling the first layer to be the pareto optimal front which is not subjected to the dominant of any other individuals.
  10. 10. The intelligent analysis system for adapting to multi-dimensional rail transit project bid offers of claim 9, wherein the intelligent offer module further comprises, internally: The crowding degree distance calculating unit is used for calculating the crowding degree distance of each individual in the target space based on the inside of each non-dominant level, evaluating the distribution density of the individual in the target space, wherein the higher the crowding degree distance is, the lower the solution density of the area where the individual is located is; The optimal quotation scheme set establishing unit is used for comprehensively sequencing individuals based on the non-dominant level and the crowding degree distance, preferentially selecting individuals with higher levels and larger crowding degree distances as father generation, carrying out crossing and mutation operation on the individuals to generate child new individuals, combining the father generation individuals and the child new individuals, carrying out non-dominant sequencing and crowding degree calculation again, reserving the optimal individuals to enter next generation iteration, and dynamically generating the optimal quotation scheme set comprising winning probability, expected benefits and risk bearing.

Description

Intelligent analysis system for adaptive multidimensional bidding quotation of rail transit projects Technical Field The invention relates to the technical field of intelligent bidding quotation, in particular to an intelligent analysis system for adaptive multidimensional rail transit project bidding quotation. Background In the conventional bidding quotation system for the rail transit project, the defects of single data dimension and split analysis process generally exist, the traditional technology usually only depends on historical quotation or simple cost data to conduct linear prediction and lacks deep association and fusion analysis on project context, dynamic trends of owners and competitor group behaviors, meanwhile, market environment changes and internal resource constraints of enterprises are usually processed statically or in isolation, so that bid probability evaluation deviation is large, bid decision making is difficult to cooperatively optimize enterprise resources, expected benefits and risk bearing capacity, and finally a generation scheme lacks dynamic adaptability and global optimality. Disclosure of Invention In order to solve the technical problems, the intelligent analysis system for bidding quotation of the adaptive multidimensional rail transit project is provided, and the technical scheme solves the key problems of inaccurate prediction, insufficient resource coordination and non-global optimal quotation scheme caused by single data dimension and split decision process in the conventional bidding quotation system. In order to achieve the above purpose, the invention adopts the following technical scheme: An intelligent analysis system for adapting to multidimensional rail transit project bid quotation, comprising: The system comprises a multidimensional data acquisition module, a probability prediction module and an intelligent quotation module; the intelligent quotation module is electrically connected with the probability prediction module; The multidimensional data acquisition module acquires historical bidding data of the track traffic project, analyzes the attribute and quotation structure of the track traffic project, performs group division and behavior label definition on competitors, and generates a high-dimensional composite feature vector of the historical bidding associated context-owner trend-competition trend of the track traffic project; The probability prediction module is used for establishing a winning probability prediction model of the track traffic project based on a high-dimensional composite feature vector of historical bidding associated context-owner tendency-competition tendency of the track traffic project, taking external market environment parameters as compensation factors, and evaluating winning enterprises of the track traffic project in high probability; The intelligent quotation module is used for constructing a multi-objective optimization function based on the high-probability bid-winning enterprise evaluation result of the track traffic project and combining the multi-dimensional resource constraint condition of the enterprise and the expected income target of the project, and dynamically generating an optimal quotation scheme set. Preferably, the multidimensional data acquisition module specifically comprises: The data set establishing unit is used for extracting the past 5-10 years track traffic project bidding historical data through an API (application program interface) connecting tool based on the enterprise ERP system, the bidding management platform and the public bidding platform, cleaning the data, removing repeated items, processing missing values, unifying data formats and aligning time stamps to obtain the track traffic project bidding historical data set after data preprocessing; The track traffic project bid historical data set comprises project basic attributes such as line types, mileage, construction period requirements and technical standard grades, owner information such as unit properties and historical project preferences, and competitor bid records such as quotation, technical scheme and bid winning conditions. Preferably, the multidimensional data acquisition module further comprises: The enterprise bidding behavior feature matrix building unit is used for calculating the average bidding discount rate, the bidding fluctuation rate and the budget deviation degree of an enterprise in the time of historical data based on a rail transit project bidding historical data set, extracting the annual bidding times, the bidding density and the project type concentration degree of the enterprise, calculating the adoption rate of an enterprise high-technology scheme and the application frequency of a patent technology, analyzing the frequent competitor set and the bidding time selection preference of the enterprise, carrying out Z-score standardization processing on the frequent competitor set and the bidding time selection preferen