A dynamic risk assessment method and system for unbalanced pricing in bill of quantities

CN122779640APending Publication Date: 2026-09-18QINGDAO FULAI YITONG SOFTWARE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611228937.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-13
Publication Date
2026-09-18

AI Technical Summary

Technical Problem

审核人员通过逐项比对清单单价与招标控制价,或计算简单的单价算术平均值偏离度来筛查异常,然而,上述方法维度单一且缺乏量化标准;随着电子招投标的普及,部分系统尝试引入数据分析手段进行辅助预警,但仍停留在基于固定阈值的简单比对层面,无法根据项目特征动态调整判断标准

Benefits of technology

首先,通过获取目标评估对象的招投标基础数据,并通过预训练的动态评估模型根据招投标基础数据中每个清单项的清单编码从时间维度和变更概率维度对清单项进行分类预测,得到每个清单项的维度分组和维度权重,可以消除权重赋值的主观随意性,并且可以保证分类结果与工程造价规律的吻合度,为后续偏离率聚合和风险评分提供精准的数据基础;接着,计算每个清单项的造价占比偏离率,可以精准捕捉分享造价结构的隐性转移,并且按照每个数据维度分组聚合得到的分组造价占比偏离率数据,将零散的清单项数据转化为反映施工资金流的连续特征,可以为后续策略倾向识别提供可靠的输入数据;随后,利用预训练的动态评估模型,根据每个数据维度分组的分组造价占比偏离率判断目标评估对象的目标策略倾向标签,以精准区分不同的隐蔽报价意图,使得判断结果具备较高的一致性和抗干扰能力;之后,结合各清单项维度权重对偏离率进行融合,可以得到不同维度的冒烟指数,可以体现不同工程分项在造价结构中的重要性差异;最后,基于不同维度的冒烟指数以及目标策略倾向标签共同确定风险等级,将不同维度的风险纳入统一决策框架,既可以避免单一维度判断的片面性,又可以通过量化分级输出为评标流程的差异化处置提供客观依据,从而实现从数据获取到风险决策的全链路自动化闭环。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122779640A_ABST
    Figure CN122779640A_ABST
Patent Text Reader

Abstract

This application discloses a dynamic risk assessment method and system for unbalanced pricing in bill of quantities, relating to the field of construction project bidding information processing technology. The method includes: acquiring basic bidding data; extracting features from the bill of quantities codes using a pre-trained dynamic assessment model; predicting the dimensional grouping and weights of each bill of quantities item from multiple dimensions; calculating the cost ratio deviation rate based on the ratio of the bid price to the tender control price of each bill of quantities item; aggregating the cost ratio deviation rates by different dimensions; analyzing the group cost ratio deviation rate for each data dimension group using the pre-trained dynamic assessment model to obtain a target strategy tendency label; and determining the risk level of the target assessment object by combining the dimensional weights with the cost ratio deviation rates of each bill of quantities item to obtain different dimensional smoke indices and target strategy tendency labels. In this way, hidden unbalanced pricing can be accurately identified, thereby significantly improving bid evaluation efficiency and settlement risk control capabilities.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of construction project bidding information processing technology, and in particular to a dynamic risk assessment method and system for unbalanced pricing in bill of quantities. Background Technology

[0002] In the bidding process for construction projects, unbalanced pricing is a common bidding strategy employed by bidders. The core practice involves intentionally increasing the unit price of items whose projected workload might increase, while simultaneously decreasing the unit price of items whose projected workload is reduced or those in later stages of construction, all while maintaining a constant or competitive total bid price. This allows them to obtain excess profits through project changes after winning the bid, or to recoup funds early. This behavior severely damages the economic interests of the tendering party and creates the potential for cost overruns in the final settlement.

[0003] Currently, the industry primarily relies on the manual review of cost engineers to identify unbalanced pricing. Reviewers screen for anomalies by comparing unit prices in the bill of quantities with the tender control price item by item, or by calculating the deviation of the simple arithmetic mean of unit prices. However, these methods are limited in scope and lack quantitative standards. With the popularization of electronic bidding, some systems have attempted to introduce data analysis methods for auxiliary early warning, but they still remain at the level of simple comparison based on fixed thresholds and cannot dynamically adjust the judgment criteria according to project characteristics. Summary of the Invention

[0004] This application provides a dynamic risk assessment method and system for unbalanced pricing in bill of quantities, which aims to solve the following technical problem: how to accurately identify hidden unbalanced pricing in order to significantly improve bid evaluation efficiency and settlement risk control capabilities.

[0005] In a first aspect, embodiments of this application provide a dynamic risk assessment method for unbalanced pricing in a bill of quantities, the method comprising: Obtain the basic bidding data of the target evaluation object; By using a pre-trained dynamic evaluation model, feature extraction is performed on the list code of each list item in the bidding basic data. Each list item is classified and regressed from multiple data dimensions to obtain the dimension grouping and dimension weight corresponding to each list item. Based on the total bid price and the total tender control price for each of the aforementioned items, determine the cost percentage deviation rate for each of the aforementioned items; For each data dimension, the cost percentage deviation rate of each item in the list is aggregated according to the dimension group corresponding to the data dimension, so as to obtain the group cost percentage deviation rate of each dimension group in the data dimension. The pre-trained dynamic evaluation model identifies the target strategy tendency label of the target evaluation object based on the deviation rate of the group cost ratio of each group in each data dimension. For each data dimension, based on the dimension weight of each item in the list within the data dimension, the cost ratio deviation rate of each item in the list is fused to determine the smoke index of the data dimension; Based on the smoke index of each data dimension and the target strategy tendency label of the target assessment object, the risk level of the target assessment object is determined.

[0006] Secondly, embodiments of this application also provide a dynamic risk assessment system for unbalanced bill of quantities pricing. The system includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a dynamic risk assessment method for unbalanced bill of quantities pricing as described above.

[0007] Thirdly, embodiments of this application also provide a computer storage medium storing computer-executable instructions, which, when executed, implement a dynamic risk assessment method for unbalanced pricing of bill of quantities as described above.

[0008] The dynamic risk assessment method and system for unbalanced pricing in bill of quantities provided in this application have the following beneficial effects: First, by acquiring the bidding and tendering basic data of the target evaluation object, and using a pre-trained dynamic evaluation model to classify and predict the list items from the time dimension and change probability dimension based on the list code of each list item in the bidding and tendering basic data, the dimensional grouping and dimensional weight of each list item are obtained. This can eliminate the subjective arbitrariness of weight assignment and ensure the consistency of classification results with the engineering cost pattern, providing an accurate data foundation for subsequent deviation rate aggregation and risk scoring. Next, the cost proportion deviation rate of each list item is calculated, which can accurately capture the implicit transfer of the shared cost structure. Furthermore, the grouped cost proportion deviation rate data obtained by grouping and aggregating according to each data dimension transforms the scattered list item data into continuous features reflecting the construction fund flow, providing reliable input data for subsequent strategy tendency identification. Subsequently, using a pre-trained dynamic evaluation model, the target strategy tendency label of the target evaluation object is determined based on the deviation rate of the group cost proportion of each data dimension group, so as to accurately distinguish different hidden bidding intentions and make the judgment results have high consistency and anti-interference ability. Then, the deviation rate is fused by combining the weight of each item dimension to obtain the smoke index of different dimensions, which can reflect the difference in importance of different engineering sub-items in the cost structure. Finally, the risk level is determined based on the smoke index of different dimensions and the target strategy tendency label, and the risks of different dimensions are incorporated into a unified decision-making framework. This can not only avoid the one-sidedness of single-dimensional judgment, but also provide an objective basis for differentiated handling in the bidding process through quantitative and graded output, thereby realizing a closed loop of full-link automation from data acquisition to risk decision-making. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a dynamic risk assessment method for unbalanced pricing in a bill of quantities, provided as an embodiment of this application; Figure 2 A schematic diagram illustrating the comprehensive risk analysis results of unbalanced pricing provided in this application embodiment; Figure 3 This is a schematic diagram of the internal structure of a dynamic risk assessment system for unbalanced pricing in a bill of quantities, provided as an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] It is understood that in the embodiments of this application, data such as user information are involved. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with relevant laws, regulations and standards.

[0012] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0013] In the following description, the terms “first, second, ...” are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that “first, second, ...” may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0014] In the bidding process for construction projects, unbalanced pricing is a common bidding strategy employed by bidders. The core practice involves intentionally increasing the unit price of items whose projected workload might increase, while simultaneously decreasing the unit price of items whose projected workload is reduced or those in later stages of construction, all while maintaining a constant or competitive total bid price. This allows them to obtain excess profits through project changes after winning the bid, or to recoup funds early. This behavior severely damages the economic interests of the tendering party and creates the potential for cost overruns in the final settlement.

[0015] Currently, the industry primarily relies on the manual review of cost engineers to identify unbalanced pricing. Reviewers screen for anomalies by comparing unit prices in the bill of quantities with the tender control price item by item, or by calculating the deviation of the simple arithmetic mean of unit prices. However, these methods are limited in scope and lack quantitative standards. With the popularization of electronic bidding, some systems have attempted to introduce data analysis methods for auxiliary early warning, but they still remain at the level of simple comparison based on fixed thresholds and cannot dynamically adjust the judgment criteria according to project characteristics.

[0016] Based on this, this application provides a dynamic risk assessment method for unbalanced pricing in bill of quantities, which can accurately identify hidden unbalanced pricing, thereby significantly improving bid evaluation efficiency and settlement risk control capabilities. This process can be executed by computing devices in the relevant field (e.g., smart mobile terminals), and certain input parameters or intermediate results in the process can be manually adjusted to help improve accuracy.

[0017] This application provides a dynamic risk assessment method for unbalanced pricing in bill of quantities. It should be noted that the execution entity in these embodiments can be a server or any terminal device with data processing capabilities. For example, the server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal device can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, in-vehicle terminal, etc., but is not limited to these.

[0018] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0019] Figure 1 This document presents a flowchart illustrating a dynamic risk assessment method for unbalanced pricing in a bill of quantities, as provided in an embodiment of this application. This method can be applied to various construction project bidding scenarios, such as automatic bid clearing on electronic oversubscription platforms, bid awarding decisions by large real estate companies / owners, auditing and review services by cost consulting firms, and self-inspection and pricing strategy verification by bidders.

[0020] like Figure 1 As shown in the embodiment of this application, a dynamic risk assessment method for unbalanced pricing in a bill of quantities specifically includes the following steps: Step 101: Obtain the basic bidding data of the target evaluation object.

[0021] It should be noted that the target evaluation object refers to the bidders who have entered the bid evaluation stage in the construction project bidding activities and whose bid price documents require unbalanced risk analysis. It can be a single bidder or a collection of all bidders that meet the review requirements in the same bidding project. The bidding basic data refers to the structured data extracted from the bidding control price document and the bid price documents of the target evaluation object, which is used to calculate the cost economic indicators. It includes at least the list code, project name, quantity, comprehensive unit price of each list item, the bid price of the list item obtained by multiplying the comprehensive unit price and the quantity, the total bid price obtained by summing the bid prices of all list items, the bidding control price of the list items, and the total bidding price obtained by summing the bidding control prices of all list items.

[0022] As an example, the target evaluation object is bidder A. The list code for a certain item in the bidding data is 040101001001, the project name is site leveling, and the quantity is 2000 m². 2 The comprehensive unit price is 5.20 yuan, the total tender control price for the items in the bill of quantities is 10,400 yuan, the total tender cost is 5,000,000 yuan, the comprehensive unit price of bidder A is 4.80 yuan, and the quantity of work is 2000 m². 2 The bid price for each item in the bill of quantities is 9,600 yuan, and the total bid price is 4,980,000 yuan.

[0023] Step 102: Using a pre-trained dynamic evaluation model, extract features from the list code of each list item in the bidding basic data, classify and regress each list item from multiple data dimensions, and obtain the dimension grouping and dimension weight corresponding to each list item.

[0024] Here, multiple data dimensions can include time dimension and change probability dimension; in addition, they can also include dimensions such as engineering quantity flexibility dimension and market price fluctuation sensitivity dimension. Among them, the time dimension and change probability dimension have a significant impact on the accuracy of unbalanced pricing identification, so the following embodiments will focus on explaining the time dimension and change probability dimension in detail.

[0025] It should be noted that a "list item" refers to the smallest pricing unit in the bill of quantities that constitutes the physical entity of the proposed project and is measured by a single unit of measurement; a "list code" refers to a hierarchical numerical code uniformly compiled according to national or local construction engineering quantity calculation standards, used to uniquely identify the professional attributes and sub-item affiliation of a list item, typically represented by twelve Arabic numerals, with the first two digits being the professional code, the middle two the sub-section code, the fifth to eighth digits the item code, and the last four digits the specific sub-item code; the "time dimension" refers to the classification perspective that reflects the sequence of construction stages to which each list item belongs, with reference to the construction project schedule; and the "change probability dimension" refers to the probability of design changes occurring in each sub-item of the project in the historical engineering database. Based on the frequency of changes and the statistical regularity of the increase in the quantity of work after the changes, this reflects the classification perspective of the possibility that the quantity of work in the bill of quantities will be adjusted during future construction. The quantity elasticity dimension refers to the coefficient used to characterize the transmission multiple of quantity changes to settlement amount changes, which is determined by the statistical regression relationship between the quantity change rate and the settlement amount change rate of similar bill of quantities items in historical projects. It is used to amplify or suppress the contribution of the deviation rate of the unit price of changed weights. The market price fluctuation sensitivity dimension refers to the sensitivity level determined by the weighted product of the cost weight of the main materials in the comprehensive unit price of the bill of quantities item and the historical market price fluctuation rate of the material. It is used to identify the irrational low-price or profit-locking high-price behavior of bidders on high-fluctuation material items.

[0026] In some embodiments, multiple data dimensions include a time dimension and a change probability dimension; the pre-trained dynamic evaluation model includes a first prediction submodule, which includes a first feature encoding layer, a first output head, a second output head, and a third output head; step 102 above can be implemented in the following way: for each list item in the bidding basic data, the following processing is performed respectively: through the first feature encoding layer, feature extraction is performed on the list code of the list item to obtain list item features; through the first output head, based on the list item features, the list item is classified in the time dimension to obtain the time dimension grouping of the list item; through the second output head, based on the list item features, the list item is classified in the change probability dimension to obtain the change probability dimension grouping of the list item; through the third output head, based on the list item features, regression prediction is performed on the list item to obtain the time dimension weight and change probability dimension weight of the list item.

[0027] Thus, by transforming the hierarchical structure of the bill of quantities coding into a dense vector representation through the first feature encoding layer, the model can automatically capture the deep-seated patterns implicit in the national standard bill of quantities coding, such as professional attributes, construction sequence, and change sensitivity, avoiding the limitations of manually formulating matching rules. Simultaneously, the three output heads share the same underlying coding features, completing the prediction tasks of time-dimension grouping, change probability-dimension grouping, and two-dimensional weights in parallel during a single forward propagation. This significantly reduces inference latency and meets the batch real-time processing requirements of electronic bidding scenarios. Furthermore, the joint output of time-dimension grouping and change probability-dimension grouping enables each bill of quantities... Each item can carry two types of labels simultaneously: construction sequence and change risk, providing a structured input basis for the subsequent parallel deviation rate fusion. The two-dimensional weights obtained from the regression of the third output head can adapt to the differences in the actual cost impact of various sub-items in different project types, thereby improving the generalization ability of the scoring system in different engineering scenarios. In addition, the entire first prediction sub-module compresses the originally scattered steps of table lookup matching, manual weighting, and classification judgment into an end-to-end reasoning process, which greatly simplifies the system architecture while ensuring that the output results conform to the logic of cost engineering, and realizes automatic mapping from raw coding to structured classification results.

[0028] As an example, taking the bill of quantities code 010101001001 (site leveling) in a building construction project as an example, firstly, the twelve-digit code (bill of quantities code) is split into professional code 01, section code 01, item code 0100, and sub-item code 1001 according to the hierarchy. This sequence is then input into the first feature encoding layer. The first feature encoding layer uses a pre-trained embedding matrix to map each segment of the code into a dense vector of fixed dimension, and then concatenates and fuses them to output the bill of quantities item features that represent the process attributes and cost characteristics of the item. Subsequently, the characteristics of the list items will be determined. The system simultaneously inputs three parallel output heads. The first output head uses a fully connected layer and a Softmax activation function to classify the feature vectors along the time dimension, grouping them into the early stages of the project. The second output head uses the same structure to classify the feature vectors along the change probability dimension, grouping them into the high-occurrence group. The third output head uses a linear activation function to classify the list item features. Regression prediction is performed, outputting the weight values ​​of the list items in the time dimension (9.2) and the change probability dimension (8.7). These weight values ​​are then normalized to the range of 0 to 10 after being mapped using the Sigmoid function, thus obtaining the time dimension weights of the list items. and change probability dimension weights .

[0029] In some embodiments, the first prediction submodule may be optimized end-to-end using a multi-task joint training approach.

[0030] In the training dataset construction phase, bill of quantities data of historical completed projects were collected, and the 12-digit code of each bill of quantities item was extracted as model input. Simultaneously, based on the construction organization design and actual progress records of historical projects, the construction stage to which each bill of quantities item belongs was labeled, and time-dimensional grouping labels were obtained. Based on the statistical results of the number of change orders and the increase in quantity after changes for each sub-item of the historical projects, the product of the change occurrence rate and the average change magnitude was used as the sorting criterion, and bill of quantities items were labeled as high-occurrence or low-occurrence groups in the change probability dimension. Furthermore, the proportion of each bill of quantities item's total price in its respective project to the total project cost was used as an importance reference. Combined with the cost engineer's experience in risk rating of key sub-items, regression labels for time-dimensional weights and change probability dimension weights were generated. The label values ​​were normalized and mapped to the range of 0 to 10.

[0031] During the training phase, the labeled list-encoded sequences are first batch-inputted into the first feature encoding layer. After being converted into dense vectors by the embedding layer, they are fed into a shared feature extraction network. This network can employ a multilayer perceptron or a shallow Transformer encoder structure, responsible for capturing the hierarchical semantic relationships between segments of numbers in the encoding. Subsequently, the output of the feature extraction network is fed into three task heads. The first and second output heads both use fully connected layers with Softmax activation to minimize the cross-entropy loss for time-dimension classification and the cross-entropy loss for change probability-dimension classification, respectively. The third output head uses a fully connected layer with linear activation to minimize the mean squared error loss between the predicted weights and the labeled weights. The loss functions of the three tasks are weighted and summed according to preset coefficients to obtain the total loss, and the network parameters of the shared encoding layer and each output head are updated synchronously through the backpropagation algorithm.

[0032] Furthermore, an early stopping mechanism and a learning rate decay strategy can be introduced during training. Training is terminated when the multi-task comprehensive index on the validation set no longer improves, ultimately resulting in a first prediction submodule that can simultaneously output dimension grouping and dimension weights in a single forward propagation. After training is complete, the parameters of the first prediction submodule can be fixed for online evaluation inference.

[0033] In some embodiments, in addition to predicting the dimension grouping and dimension weight corresponding to the list item through the pre-trained dynamic evaluation model in step 102, a mapping table can also be pre-set to determine the corresponding dimension grouping and dimension weight through mapping. For example, the list item with the list code prefix 0116 or 0410 corresponds to the time dimension grouping as the early stage of the project and the dimension weight is 10. The list item with the list code prefix 0112 corresponds to the change probability dimension grouping as high change risk and the dimension weight is 8.75.

[0034] Step 103: Based on the bid price and tender control price of each item in the list, determine the cost deviation rate of each item in the list.

[0035] It should be noted that the bid total price of a list item refers to the total price calculated based on the comprehensive unit price and the quantity of work for a single list item in the bid price list of the target evaluation object, that is, the absolute monetary contribution of the list item to the total bid cost; the tender control price total price of a list item refers to the total price of the same list item corresponding to the bid list item code in the tender control price list issued by the tenderer, which is obtained by multiplying the comprehensive unit price and the quantity of work specified in the tender control price, and its value represents the expected cost benchmark of the sub-item project when the tenderer prepares the control price.

[0036] In some embodiments, step 103 described above can be implemented as follows: for each of the list items, the following processing is performed: the ratio between the bid price of the list item and the total bid cost of the target evaluation object is taken as the first proportion of the list item in the total bid cost; the ratio between the tender control price of the list item and the total tender cost is taken as the second proportion of the list item in the total tender cost; the ratio between the first proportion and the second proportion is subtracted from the value 1 to obtain the cost proportion deviation rate of each of the list items.

[0037] In this way, by unifying the cost data of the bidding side and the tendering side into a proportion of the total cost before performing the ratio calculation, the interference of differences in the total investment scale of different projects on the deviation calculation can be effectively eliminated, making the quotation data across projects and types comparable. The ratio relationship between the first proportion and the second proportion can penetrate the disguise of total price balance and accurately capture the implicit adjustments made by bidders to the cost structure of each item without changing the total price. This facilitates subsequent integration and risk level determination. Furthermore, this calculation method can ensure the consistency and traceability of the evaluation process, while providing a standardized input basis for subsequent weighted average integration by dimension grouping.

[0038] As an example, the cost deviation rate of the i-th item in the bill of quantities can be calculated using formula (1). .

[0039] (1) in, For the bid price of the i-th item in the list, The total bid cost, Let i be the total tender control price for the i-th item in the list. This represents the total cost of the tender.

[0040] Step 104: For each data dimension, aggregate the cost ratio deviation rate of each item in the list according to the dimension group corresponding to the data dimension to obtain the group cost ratio deviation rate of each dimension group in the data dimension.

[0041] Here, aggregation can be direct summation, weighted summation, or weighted average; no specific limitation is made here.

[0042] As an example, taking a certain project as an example, the system calculates through the above steps that the cost percentage deviation rates for preliminary items such as foundation pit support and earthwork are +18.3% and +15.7%, respectively; for intermediate items such as main structure and masonry, the rates are -2.1% and +1.4%; and for post-project items such as roof waterproofing and interior decoration, the rates are -12.6% and -9.8%. During the aggregation process, the system weights the deviation rates of items within the same data dimension group according to their respective bid prices as a percentage of the total cost of that group, thus obtaining the cost percentage deviation rate for each dimension group in the time dimension. For example, in the preliminary group, the cost of foundation pit support accounts for 65% of the total cost of the group, and earthwork accounts for 35%. Therefore, the weighted average can be calculated to obtain a cost percentage deviation rate of approximately 17.4% for the preliminary group. Similarly, the cost percentage deviation rate for the intermediate group is approximately 0.6%, and the cost percentage deviation rate for the post-project group is approximately -11.7%.

[0043] Step 105: Using the pre-trained dynamic evaluation model, strategy identification is performed based on the deviation rate of the group cost ratio of each dimension group in each data dimension to determine the target strategy tendency label of the target evaluation object.

[0044] In some embodiments, the pre-trained dynamic evaluation model includes a second prediction submodule, which includes a second feature encoding layer and a fully connected layer. Step 105 described above can be implemented as follows: through the second feature encoding layer, feature extraction is performed on the deviation rate of the group cost ratio of each group in each of the data dimensions to obtain the deviation rate feature; through the fully connected layer, based on the deviation rate feature, the strategy tendency prediction is performed on the target evaluation object to obtain the confidence level of the target evaluation object for each strategy tendency label, and the strategy tendency label corresponding to the highest confidence level is taken as the target strategy tendency label of the target evaluation object.

[0045] Thus, the second prediction submodule can analyze the distribution information implicit in the deviation rates of different groups, and the second feature encoding layer can capture the relative high and low relationships and gradient change trends between different group deviation rates, thereby identifying imbalanced strategies with features. Furthermore, the fully connected layer maps the encoded deviation rate features to the confidence distribution of each strategy tendency label, enabling the system to not only output qualitative judgments but also quantitatively express the degree of confidence in the judgments, providing a quantifiable adjustment basis for the dynamic calibration of the smoke index. Moreover, the second prediction submodule is functionally decoupled from the first prediction submodule, which can reduce the complexity of a single model, thereby improving the maintainability and scalability of the overall framework.

[0046] As an example, taking bidder A as an example, the above steps can obtain the cost ratio deviation rate of bidder A in each dimension group in each data dimension; then, the second feature encoding layer receives it and converts it into a low-dimensional dense vector through a fully connected embedding network. During the feature extraction process, the encoding layer automatically captures the gradient change difference of the deviation rate and encodes it into a vector representation with a specific direction in the feature space, thus obtaining the deviation rate feature. Then, the deviation rate feature The input is a fully connected layer, which contains neurons that correspond one-to-one with various policy preference labels. After Softmax activation, the output is the confidence distribution of each label. For example, the confidence of the "heavy before light" type is 0.91, the confidence of the "light before heavy" type is 0.06, the confidence of the "change arbitrage" type is 0.02, and the confidence of the "no significant preference" type is 0.01. Finally, the "heavy before light" type corresponding to the maximum confidence value of 0.91 is selected as the target policy preference label, while the complete confidence vector is retained for use in the subsequent smoke index calibration step.

[0047] In some embodiments, the second prediction submodule may use a sample of bidding strategies from historical projects that have been manually audited as the labeled data source.

[0048] During the training sample construction phase, the cost proportion deviation rate of each historical bidder in each dimension group across different data dimensions was extracted to form multidimensional input features. At the same time, cost engineers combined actual project settlement data with bidder performance behavior to label each bid document with its true strategic tendency label. The labeling categories included heavy at the beginning and light at the end, light at the beginning and heavy at the end, change arbitrage, and no significant tendency. To improve the model's ability to distinguish ambiguous samples, samples with indistinct strategic tendencies were assigned lower labeling confidence weights during the labeling process, thereby reducing their contribution to the loss function.

[0049] During the training phase, firstly, the multidimensional input features are fed into the second feature encoding layer. This layer employs a two-layer fully connected network structure, mapping the spatial distribution features of the input into a discriminative dense representation through nonlinear transformation. A Dropout mechanism is introduced in the intermediate layers of the network to prevent overfitting. Subsequently, the deviation rate features output by the second feature encoding layer are fed into the subsequent fully connected layer. The fully connected layer has four output nodes corresponding to four policy preference categories, and after Softmax activation, it outputs the probability distribution of each category. The training objective can adopt a cross-entropy loss function with class weights, assigning higher loss weights to the two categories of samples, namely, those with heavy initial weights followed by light initial weights and those with light initial weights followed by heavy initial weights, to enhance the model's sensitivity to time-imbalanced policies and its ability to identify change-based arbitrage. The optimizer can use the Adam algorithm, with an initial learning rate set to 0.001, and dynamically adjust the learning rate using a cosine annealing scheduling strategy.

[0050] Furthermore, during training, five-fold cross-validation can be used to evaluate the model's generalization performance. Training is terminated when the classification accuracy on the validation set reaches a preset threshold and no longer improves after ten consecutive epochs. This ultimately yields a second prediction submodule capable of accurately identifying bidder strategy tendencies based on time-dimension grouped deviation rate sequences. The parameters of the second prediction submodule are fixed after training, allowing it to be deployed in parallel with the first prediction submodule, sharing the same batch of deviation rate calculation results but independently performing inference tasks.

[0051] In some embodiments, a trend curve can be generated based on the cost percentage deviation rate of each item in the list, and the target strategy tendency label of the target evaluation object can be identified through the curve fluctuation characteristics.

[0052] Step 106: For each data dimension, based on the dimension weight of each item in the data dimension, merge the cost ratio deviation rate of each item in the data dimension to determine the smoke index of the data dimension.

[0053] It should be noted that the smoke index is a dimensionless risk score that measures the degree of imbalance in bid pricing within a given data dimension, calculated by weighting the deviation rate of each item's cost proportion according to the dimensional weight of each item. A higher score for a given data dimension indicates a greater risk for the bidder in that dimension.

[0054] In some embodiments, the step 106 above, which involves fusing the cost percentage deviation rates of each item in the data dimension based on the dimension weight of each item in the data dimension to determine the smoke index of the data dimension, can be implemented as follows: Based on the dimension weight of each item in the data dimension, a weighted summation of the cost percentage deviation rates of each item in the data dimension is performed to obtain a summation result; based on the confidence level corresponding to the target strategy tendency label, an index calibration factor corresponding to the data dimension is determined; the index calibration factor corresponding to the data dimension is multiplied by the summation result to obtain the smoke index of the data dimension.

[0055] Thus, by weighting the deviation rate through dimensional weights, the items with large monetary proportions and high risk sensitivity dominate the final score. This avoids the drawback of diluting the abnormal signals of key sub-items with simple arithmetic averaging, and aligns with the engineering logic of the actual impact differences of different sub-items in the cost structure. Furthermore, by introducing the confidence level corresponding to the target strategy tendency label to determine the index calibration factor, a dynamic feedback channel can be established between the strategy identification result and the risk quantification score. When the model is more confident in its judgment of a certain strategy tendency, the calibration factor amplifies the smoke index of that dimension accordingly, and conversely, it converges to the neutral value. This allows the score result to reflect both the deviation magnitude and the credibility of the strategy intention. In addition, by using the independent calculation of calibration factors for each dimension and applying them to the summation result of each dimension, the decoupled assessment of time-related and change-related imbalance risks can be achieved, avoiding the problem of anomalies in one dimension being masked by the smooth performance of another dimension.

[0056] As an example, suppose the list includes item A, item B, and item C, and the data dimensions include a time dimension and a change probability dimension. The time dimension weight for item A is... The weight of the probability of change dimension is The deviation rate of the cost ratio is The time dimension weight corresponding to item B in the list is The weight of the probability of change dimension is The deviation rate of the cost ratio is The time dimension weight corresponding to item C in the list is The weight of the probability of change dimension is The deviation rate of the cost ratio is The confidence level corresponding to the target policy preference label T is Subsequently, based on the dimensional weight of each item in the time dimension, the cost percentage deviation rate of each item is weighted and summed to obtain the summation result in the time dimension. Meanwhile, the weight of each item in the change probability dimension is calculated by weighting and summing the cost percentage deviation rates of each item to obtain the summation result of the change probability dimension. Then, based on the confidence level corresponding to the target policy preference label T, The mapping yields the exponential calibration factor for the time dimension. The index calibration factor obtained from the change probability dimension is: Finally, the time dimension exponential calibration factor is adjusted. The summation result with the time dimension Multiplying them together yields the smoke index in the time dimension; similarly, the index calibration factor in the probability of change dimension is calculated. The summation result with the change probability dimension Multiplying them together yields the smoke index for the probability of change.

[0057] In some embodiments, the above-mentioned determination of the index calibration factor corresponding to the data dimension based on the confidence level corresponding to the target policy tendency label can also be achieved by: calculating the difference between the confidence level corresponding to the target policy tendency label and a preset confidence level to obtain a confidence deviation; determining a first dimension mapping rule that is compatible with the data dimension; and performing mapping processing on the confidence deviation based on the first dimension mapping rule to obtain the index calibration factor corresponding to the data dimension.

[0058] In some embodiments, the multiple data dimensions include a time dimension and a change probability dimension. The aforementioned determination of the index calibration factor corresponding to the data dimension based on the confidence level corresponding to the target policy tendency label can be achieved as follows: Calculate the difference between the confidence level corresponding to the target policy tendency label and a preset confidence level to obtain a confidence deviation; In response to the data dimension being the time dimension, perform a mapping process on the confidence deviation based on a time dimension mapping rule to obtain the index calibration factor corresponding to the time dimension; In response to the data dimension being the change probability dimension, perform a second mapping process on the confidence deviation based on a change probability dimension mapping rule to obtain the index calibration factor corresponding to the change probability dimension.

[0059] Thus, by configuring differentiated mapping rules for the two dimensions, the time-based and change-based imbalance risks are asymmetrically amplified in the scoring. This aligns with the different mechanisms of harm caused by the earlier-heavy and later-lighter risk directly impacting cash flow and the delayed settlement overspending due to change arbitrage. Furthermore, the confidence bias is transformed into a calibration factor through monotonic mapping. This not only preserves the model's ability to adjust the final score based on its own confidence level, but also avoids score instability caused by extreme confidence values ​​by setting a pre-set confidence anchor baseline. This overall design can seamlessly embed the semantic output of strategy identification into the quantitative scoring framework, achieving a coupled assessment of risk intent and deviation magnitude without disrupting the weighted summation core logic. This makes the smoke index both statistically significant and engineering interpretable in terms of numerical value.

[0060] It should be noted that the first and second mapping processes can be implemented using simple linear mapping functions or mapping tables, or through prediction by pre-trained models; no specific limitations are made here.

[0061] As an example, assuming the confidence level corresponding to the target tendency policy label is 0.91 and the preset confidence level is 0.8, the confidence deviation can be calculated to be 0.11. If the data dimension is the time dimension, the index calibration factor corresponding to the time dimension can be obtained by mapping the data dimension using the pre-trained first prediction model, for example, the index calibration factor is 1.066. If the data dimension is the change probability dimension, the index calibration factor corresponding to the change probability dimension can be obtained by mapping the change probability dimension using the pre-trained second prediction model, for example, the index calibration factor is 0.982.

[0062] In some embodiments, step 106 above can also be used to directly calculate the smoke index for each data dimension using formula (2). .

[0063] (2) in, Let be the cost percentage deviation rate of the i-th item in the list. This represents the dimension weight of the i-th item in the corresponding data dimension.

[0064] Step 107: Determine the risk level of the target assessment object based on the smoke index of each data dimension and the target strategy tendency label of the target assessment object.

[0065] In some embodiments, step 107 described above can be implemented as follows: determining a first weight corresponding to the target strategy tendency label of the target evaluation object; based on the first weight, weighted summing of the smoke indices of each data dimension to obtain the comprehensive smoke index of the target evaluation object; in response to the number of target evaluation objects in the current batch being greater than a preset quantity threshold, performing statistical analysis on the comprehensive smoke indices of all target evaluation objects in the current batch to obtain the mean and standard deviation, and determining a first risk threshold and a second risk threshold based on the mean and the standard deviation; in response to the comprehensive smoke index of the target evaluation object being less than the first risk threshold, determining the risk level of the target evaluation object to be low risk; in response to the comprehensive smoke index of the target evaluation object being greater than or equal to the first risk threshold and the comprehensive smoke index of the target evaluation object being less than or equal to the second risk threshold, determining the risk level of the target evaluation object to be medium risk; in response to the comprehensive smoke index of the target evaluation object being greater than or equal to the second risk threshold, determining the risk level of the target evaluation object to be high risk.

[0066] Thus, by configuring differentiated weights for the target strategy preference labels and asymmetrically fusing the two-dimensional smoke index, strategies with greater potential harm, such as change arbitrage, receive higher weights in the comprehensive score, aligning with the practical need for differentiated treatment of different risk types in engineering risk management. Furthermore, by introducing an adaptive threshold mechanism for batch statistics, absolute scores are transformed into relative rankings, ensuring that risk assessment is always anchored to the overall distribution of the current competitive environment, avoiding the high misjudgment rate of fixed thresholds when dealing with cross-project types. The entire adaptive threshold mechanism, while maintaining the simplicity of the scoring framework, combines relative and absolute assessments to consider both structured risk identification within a single project and horizontal comparability among multiple bidders, providing a stable and reliable decision-making basis for differentiated handling in the bidding process.

[0067] As an example, the first weight is determined based on the target strategy preference label. For instance, when the label is "change arbitrage," the change probability dimension weight is 0.6 and the time dimension weight is 0.4 to highlight the dominant role of change risk. When the label is "heavy at the beginning and light at the end" or "light at the beginning and heavy at the end," the time dimension weight is 0.6 and the change probability dimension weight is 0.4 to strengthen the assessment of abnormal cash flow. For "no significant preference," both dimensions are weighted at 0.5. Then, the first weight is multiplied by the smoke index of the time dimension and the change probability dimension, and the results are summed to obtain the comprehensive smoke index of the target evaluation object. After that, the comprehensive smoke index of all bidders in the current batch is calculated. When the number of batches is greater than 5 (a preset threshold), the average of all indices is calculated. with standard deviation and with As the first risk threshold, As the second risk threshold; finally, for the current bidder (target assessment object), if its comprehensive smoke index is less than the first risk threshold, it is judged as low risk; if it is between the first and second risk thresholds, it is judged as medium risk; if it is greater than or equal to the second risk threshold, it is judged as high risk.

[0068] In some embodiments, the following processing may also be performed: in response to the number of target evaluation objects in the current batch being less than or equal to a preset quantity threshold, a second preset value is used as a third risk threshold, and the third preset value is used as a fourth risk threshold; in response to the comprehensive smoke index of the target evaluation object being less than the third risk threshold, the risk level of the target evaluation object is determined to be low risk; in response to the comprehensive smoke index of the target evaluation object being greater than or equal to the third risk threshold, and the comprehensive smoke index of the target evaluation object being less than or equal to the fourth risk threshold, the risk level of the target evaluation object is determined to be medium risk; in response to the comprehensive smoke index of the target evaluation object being greater than or equal to the fourth risk threshold, the risk level of the target evaluation object is determined to be high risk.

[0069] In this way, by backing down to a fixed threshold, the problem of statistical distortion in small sample scenarios is properly solved. This avoids the risk of threshold drift caused by an excessively large standard deviation when there are only a few bidders, which could lead to the omission of high-risk bids. Furthermore, the second and third preset values ​​follow the empirical boundaries verified by the industry over a long period of time. Even when there is a lack of batch distribution reference, the basic reliability of risk assessment can still be maintained, thereby ensuring the availability of the system under extreme conditions.

[0070] Following the example above, when the batch quantity is less than or equal to 5 (preset quantity threshold), assuming the third risk threshold is 5 and the fourth risk threshold is 10, these risk thresholds can be derived from the statistical analysis of historical large-scale project data and the empirical boundaries verified by long-term practice in the cost engineering industry. For the current bidder (target evaluation object), if its comprehensive smoke index is less than 5, it is judged as low risk; if the comprehensive smoke index is between 5 and 10, it is judged as medium risk; if the comprehensive smoke index is greater than or equal to 10, it is judged as high risk.

[0071] See Figure 2 , Figure 2 This is a schematic diagram illustrating the comprehensive risk analysis results of unbalanced pricing provided in an embodiment of this application, such as... Figure 2 As shown, the overall risk scores of "Unit a" and "Unit d" are high, exceeding the set upper limit of 10 for high-risk scores. The bids for these two units are listed as significantly unbalanced bids, with a risk level of high risk. The overall risk score of "Unit c" is in the range [5,10], with a risk level of medium risk. The overall risk scores of "Unit b" and "Unit e" are less than 5, with a risk level of low risk.

[0072] In some embodiments, after performing step 107 above, the following processes may also be performed: In response to the target assessment object having a low risk level, an automatic approval instruction is generated, the bid compliance check result of the target assessment object is marked as positive, and the original bid evaluation score of the target assessment object is entered into the bid evaluation report; In response to the target assessment object having a medium risk level, a manual review instruction is generated, the top N risk list items contributing to the comprehensive smoke index are selected from the target assessment object, a local risk description including the list code, dimension grouping, cost ratio deviation rate, and dimension weight of the risk list items is generated, and pushed to manual processing for targeted verification; In response to the target assessment object having a high risk level, a mandatory review instruction is generated, the automatic bid evaluation process of the target assessment object is terminated, and a complete risk analysis report is generated.

[0073] In this way, by accurately mapping the three-level risk level with differentiated handling actions, the abstract risk quantification results are transformed into technical instructions that can directly drive the bidding process, thus achieving closed-loop control from risk identification to business execution.

[0074] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a dynamic risk assessment system for unbalanced pricing in bill of quantities, the structure of which is as follows: Figure 3 As shown.

[0075] Figure 3 This is a schematic diagram of the internal structure of a dynamic risk assessment system for unbalanced pricing in a bill of quantities, provided as an embodiment of this application. Figure 3 As shown, the system includes: At least one processor 201; And a memory 202 that is communicatively connected to at least one processor; The memory 202 stores instructions executable by at least one processor, which are executed by at least one processor 201 to enable at least one processor 201 to: Obtain the basic bidding data of the target evaluation object; By using a pre-trained dynamic evaluation model, feature extraction is performed on the list code of each list item in the bidding basic data. Each list item is classified and regressed from multiple data dimensions to obtain the dimension grouping and dimension weight corresponding to each list item. Based on the total bid price and the total tender control price for each of the aforementioned items, determine the cost percentage deviation rate for each of the aforementioned items; For each data dimension, the cost percentage deviation rate of each item in the list is aggregated according to the dimension group corresponding to the data dimension, so as to obtain the group cost percentage deviation rate of each dimension group in the data dimension. The pre-trained dynamic evaluation model identifies the target strategy tendency label of the target evaluation object based on the deviation rate of the group cost ratio of each group in each data dimension. For each data dimension, based on the dimension weight of each item in the list within the data dimension, the cost ratio deviation rate of each item in the list is fused to determine the smoke index of the data dimension; Based on the smoke index of each data dimension and the target strategy tendency label of the target assessment object, the risk level of the target assessment object is determined.

[0076] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows: Obtain the basic bidding data of the target evaluation object; By using a pre-trained dynamic evaluation model, feature extraction is performed on the list code of each list item in the bidding basic data. Each list item is classified and regressed from multiple data dimensions to obtain the dimension grouping and dimension weight corresponding to each list item. Based on the total bid price and the total tender control price for each of the aforementioned items, determine the cost percentage deviation rate for each of the aforementioned items; For each data dimension, the cost percentage deviation rate of each item in the list is aggregated according to the dimension group corresponding to the data dimension, so as to obtain the group cost percentage deviation rate of each dimension group in the data dimension. The pre-trained dynamic evaluation model identifies the target strategy tendency label of the target evaluation object based on the deviation rate of the group cost ratio of each group in each data dimension. For each data dimension, based on the dimension weight of each item in the list within the data dimension, the cost ratio deviation rate of each item in the list is fused to determine the smoke index of the data dimension; Based on the smoke index of each data dimension and the target strategy tendency label of the target assessment object, the risk level of the target assessment object is determined.

[0077] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0078] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0079] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0080] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0081] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0082] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0083] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0084] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0085] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0086] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0087] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for dynamic risk assessment of bill of quantities imbalance bidding, characterized in that, The method includes: Obtain the basic bidding data of the target evaluation object; By using a pre-trained dynamic evaluation model, feature extraction is performed on the list code of each list item in the bidding basic data. Each list item is classified and regressed from multiple data dimensions to obtain the dimension grouping and dimension weight corresponding to each list item. Based on the total bid price and the total tender control price for each of the aforementioned items, determine the cost percentage deviation rate for each of the aforementioned items; For each data dimension, the cost percentage deviation rate of each item in the list is aggregated according to the dimension group corresponding to the data dimension, so as to obtain the group cost percentage deviation rate of each dimension group in the data dimension. The pre-trained dynamic evaluation model identifies the target strategy tendency label of the target evaluation object based on the deviation rate of the group cost ratio of each group in each data dimension. For each data dimension, based on the dimension weight of each item in the list within the data dimension, the cost ratio deviation rate of each item in the list is fused to determine the smoke index of the data dimension; Based on the smoke index of each data dimension and the target strategy tendency label of the target assessment object, the risk level of the target assessment object is determined.

2. The method of claim 1, wherein, The multiple data dimensions include a time dimension and a change probability dimension; the pre-trained dynamic evaluation model includes a first prediction submodule, which includes a first feature encoding layer, a first output head, a second output head, and a third output head; The pre-trained dynamic evaluation model extracts features from the list code of each list item in the bidding basic data, classifies and regresses each list item from multiple data dimensions, and obtains the dimension grouping and dimension weights corresponding to each list item, including: For each item in the aforementioned bidding and tendering basic data, the following processing shall be performed: The list item features are obtained by extracting features from the list code of the list item through the first feature encoding layer; Using the first output head, based on the features of the list items, the list items are classified in the time dimension to obtain the time dimension grouping of the list items; Using the second output head, based on the features of the list items, the list items are classified according to the change probability dimension to obtain the change probability dimension grouping of the list items; By using the third output head, regression prediction is performed on the list item based on the list item features to obtain the time dimension weight and change probability dimension weight of the list item.

3. The method of claim 1, wherein, The determination of the cost deviation rate for each item in the bill of quantities based on the combined bid price and the combined tender control price includes: For each of the aforementioned list items, perform the following processing: The ratio between the bid price of the listed items and the total bid cost of the target evaluation object shall be used as the first proportion of the listed items in the total bid cost; The ratio between the total tender control price of the listed items and the total tender cost shall be used as the second proportion of the listed items in the total tender cost; Subtracting the value 1 from the ratio between the first percentage and the second percentage yields the cost percentage deviation rate for each item in the list.

4. The method of claim 1, wherein, The pre-trained dynamic evaluation model includes a second prediction submodule, which includes a second feature encoding layer and a fully connected layer. The pre-trained dynamic evaluation model identifies the deviation rate of the group cost proportion for each group in each data dimension to obtain the target strategy tendency label of the target evaluation object, including: Through the second feature encoding layer, the deviation rate of the group cost ratio of each group in each of the data dimensions is extracted to obtain the deviation rate feature. Through the fully connected layer, based on the deviation rate feature, the strategy tendency prediction of the target evaluation object is performed to obtain the confidence level of the target evaluation object for each strategy tendency label, and the strategy tendency label corresponding to the highest confidence level is taken as the target strategy tendency label of the target evaluation object.

5. The method according to claim 4, characterized in that, The step of determining the smoke index of the data dimension by fusing the cost ratio deviation rate of each item in the list based on the dimension weight of each item in the data dimension includes: Based on the dimension weight of each item in the data dimension, the cost ratio deviation rate of each item in the list is weighted and summed to obtain the summation result; Based on the confidence level corresponding to the target strategy preference label, determine the index calibration factor corresponding to the data dimension; Multiply the index calibration factor corresponding to the data dimension by the summation result to obtain the smoke index of the data dimension.

6. The method according to claim 5, characterized in that, The multiple data dimensions include a time dimension and a change probability dimension; determining the index calibration factor corresponding to the data dimension based on the confidence level corresponding to the target strategy tendency label includes: The difference between the confidence level corresponding to the target policy tendency label and the preset confidence level is calculated to obtain the confidence deviation; In response to the data dimension being the time dimension, the confidence deviation is mapped based on the time dimension mapping rule to obtain the exponential calibration factor corresponding to the time dimension. In response to the data dimension being the change probability dimension, the confidence deviation is mapped based on the change probability dimension mapping rule to obtain the index calibration factor corresponding to the change probability dimension.

7. The method according to claim 1, characterized in that, The process of determining the risk level of the target assessment object based on the smoke index of each data dimension and the target strategy tendency label of the target assessment object includes: Determine the first weight corresponding to the target strategy tendency label of the target evaluation object; Based on the first weight, the smoke index of each data dimension is weighted and summed to obtain the comprehensive smoke index of the target evaluation object; In response to the number of target assessment objects in the current batch being greater than a preset number threshold, a statistical analysis is performed on the comprehensive smoke index of all target assessment objects in the current batch to obtain the mean and standard deviation, and a first risk threshold and a second risk threshold are determined based on the mean and the standard deviation; In response to the fact that the comprehensive smoke index of the target assessment object is less than the first risk threshold, the risk level of the target assessment object is determined to be low risk. In response to the fact that the comprehensive smoke index of the target assessment object is greater than or equal to the first risk threshold and the comprehensive smoke index of the target assessment object is less than or equal to the second risk threshold, the risk level of the target assessment object is determined to be medium risk; In response to the comprehensive smoke index of the target assessment object being greater than or equal to the second risk threshold, the risk level of the target assessment object is determined to be high risk.

8. The method according to claim 7, characterized in that, The method further includes: In response to the number of target evaluation objects in the current batch being less than or equal to a preset number threshold, the second preset value is used as the third risk threshold, and the third preset value is used as the fourth risk threshold; If the overall smoke index of the target assessment object is less than the third risk threshold, the risk level of the target assessment object is determined to be low risk. In response to the fact that the comprehensive smoke index of the target assessment object is greater than or equal to the third risk threshold, and the comprehensive smoke index of the target assessment object is less than or equal to the fourth risk threshold, the risk level of the target assessment object is determined to be medium risk; In response to the comprehensive smoke index of the target assessment object being greater than or equal to the fourth risk threshold, the risk level of the target assessment object is determined to be high risk.

9. The method according to claim 1, characterized in that, The method further includes: In response to the target evaluation object having a low risk level, an automatic approval instruction is generated, the bid compliance check result of the target evaluation object is marked as positive, and the original bid evaluation score of the target evaluation object is entered into the bid evaluation report; In response to the target assessment object having a risk level of medium risk, a manual review instruction is generated. The top N risk items contributing to the comprehensive smoke index are selected from the target assessment object. A local risk description containing the list code, dimension grouping, cost ratio deviation rate, and dimension weight of the risk list items is generated and pushed to manual processing for targeted verification. In response to the target assessment object being classified as high-risk, a mandatory review instruction is generated, terminating the automatic bidding process for the target assessment object, and generating a complete risk analysis report.

10. A dynamic risk assessment system for unbalanced pricing in a bill of quantities, characterized in that, The system includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1-9.