Effect index layered modeling and early warning method based on cost index fluctuation prediction

Abstract

The invention relates to an effect index layered modeling and early warning method based on cost index fluctuation prediction, and relates to the field of cost index monitoring and early warning. The method comprises the following steps: predicting various target cost indexes by adopting a CNN-LSTM hybrid model; dividing each target cost effect evaluation index into a long-term reference value and a short-term offset; constructing a probability density function of a target cost effect index based on a historical project set most similar to the target project, and determining a long-term reference value of the target cost effect index based on a target cost effect index distribution estimation function; establishing a mapping relation among the target cost index, the target cost index fluctuation and the short-term offset of the target cost effect index in combination with linear regression and support vector regression models so as to predict the short-term offset of the target cost effect index; and evaluating the risk level of each index through a risk classifier by using the long-term reference value and the short-term offset of the target cost effect index.

Description

Technical Field

[0001] This invention relates to the field of cost index monitoring and early warning technology, and in particular to a hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction. Background Technology

[0002] With the continuous expansion of my country's power system construction, power transmission and transformation projects have become key infrastructure for promoting high-quality development of the power grid. At the same time, power grid companies face multiple challenges, including high investment intensity, difficulty in cost control, and strict policy supervision. Cost management has gradually become a crucial factor affecting project benefits and operational quality. Especially under the background of "dual-carbon" goals and the construction of a new power system, how to improve the accurate prediction and dynamic early warning response capabilities of power transmission and transformation project costs has become a critical challenge that urgently needs to be addressed in the field of engineering construction management. Currently, the pace of investment in power grid projects is accelerating, and project costs are characterized by complex structures, frequent price fluctuations, and variable construction conditions. This makes traditional cost control methods, which are mainly based on historical experience, somewhat outdated and one-sided, unable to meet the needs of full-cycle, refined, and dynamic cost management. Therefore, it is urgent to establish an intelligent cost indicator monitoring and early warning technology system oriented towards future trends to improve the initiative in risk identification, the objectivity of performance evaluation, and the scientific nature of management decisions.

[0003] Currently, most engineering cost prediction methods rely on regression analysis or expert scoring based on historical data, lacking dynamic modeling and causal mechanism explanations for real-time data. Some systems introduce machine learning methods for trend prediction, but these are mostly limited to time series modeling of single variables, failing to identify the transmission of cost fluctuations to engineering performance evaluation indicators and to link risk warnings. Furthermore, risk identification is often based on threshold settings, ignoring the structural biases and uncertainties of indicators. Summary of the Invention

[0004] To address the aforementioned technical problems, or at least partially them, and to overcome the aforementioned shortcomings, this invention provides a hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction.

[0005] In a first aspect, the present invention provides a hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction, including: S1. Based on the actual characteristics of the project, representative target cost indices are selected. Then, a CNN-LSTM hybrid model is used to capture the time-series dependencies of the target cost index data and its influencing factors in order to predict various target cost indices. Among them, the target cost indices include: unit capacity cost, unit capacity item costs, and transformer price. S2, Select target cost performance indicators that are closely related to the target cost index, and divide each target cost performance evaluation indicator into two levels, including long-term benchmark value and short-term deviation. S3. Construct the probability density function of the target cost performance indicator based on the set of historical projects most similar to the target project, and determine the long-term benchmark value of the target cost performance indicator by means of the expected value or mode method based on the distribution estimation function of the target cost performance indicator. S4. Combining linear regression and support vector regression models, a mapping relationship is established between the target cost index, the fluctuation of the target cost index, and the short-term deviation of the target cost effectiveness indicator, in order to predict the short-term deviation of the target cost effectiveness indicator. S5 utilizes the long-term benchmark value and short-term offset of the target cost performance indicators, assesses the risk level of each indicator through a risk classifier, calculates the risk level of each target cost performance indicator, outputs the results based on the risk level, and triggers corresponding early warning prompts, strategy recommendations, or automatic correction intervention operations.

[0006] Furthermore, S1 includes: The quantitative influencing factors of the time series impact target cost index are obtained. These factors include material prices, labor costs, equipment procurement and installation costs, construction period extension, energy prices, and transportation costs. Factors affecting unit capacity cost: Unit capacity cost is affected by material prices, labor costs, equipment procurement and installation costs, construction period, energy prices, and geographical location; Factors affecting unit capacity itemized costs: Unit capacity itemized costs consist of civil engineering costs, equipment procurement, installation and commissioning costs; civil engineering costs depend on the geological conditions of the project site, construction difficulty and environmental protection policy requirements; Factors affecting transformer prices: Transformer prices fluctuate due to raw material costs, technical specifications, supplier market competitiveness, production costs, and transportation expenses. A convolutional neural network is used to extract local features of the time series data of the factors influencing the target cost index, and LSTM is combined to capture the time series dependency between the factors influencing the target cost index and the target cost index, so as to predict various types of target cost indices in the future. The formula for the convolution operation in a convolutional neural network is as follows: ; in, The specific data on the factors influencing the target construction cost index at time ti. Represents the convolution kernel weights. Represents the convolution bias term. Represents the size of the convolution kernel. This represents the output of the convolutional layer; the convolutional neural network extracts local features that help predict the target cost index based on influencing factor data through convolution operations, and the local features are used as the input of the LSTM part. The implementation process for each LSTM unit is as follows: The forget gate output is obtained based on the hidden state of the previous time step and the local features of the current input: ;in, This represents the activation function. Indicates the forget gate weight. Indicates the hidden state at the previous moment. and local features of the current input The splicing result, Indicates the forget gate bias; The input gate output is obtained based on the hidden state of the previous time step and the local features of the current input: ;in, Indicates the input gate weights. Indicates input gate bias; The candidate layer output is obtained based on the hidden state of the previous time step and the local features of the current input: ;in, Indicates the candidate layer weights. Indicates the candidate layer bias; The cell state is updated based on the previous cell state, input gate, forget gate, and candidate layer output: ;in, This indicates the cell state at the previous moment; The output gate outputs based on the hidden state from the previous time step and the local features of the current input: ; The current hidden state is obtained based on the current cell state and the output gate output: ,in: This represents the tanh activation function; The CNN-LSTM hybrid structure first extracts local features from the time-series data of factors influencing the target construction cost index using a CNN. Then, these extracted local features are fed into an LSTM network to further explore temporal dependencies and predict future target construction cost indices. The prediction formula is expressed as follows: ; in: The factors influencing the target construction cost index over time include all factors affecting the target construction cost index. This indicates the target construction cost index for the forecast.

[0007] Furthermore, the target cost performance indicators cover the project from the feasibility study stage, design stage to construction and final acceptance stage. The target cost performance indicators include the accuracy of the feasibility study investment estimate, the change rate of the feasibility study estimate before and after review, the change rate of the preliminary estimate before and after review, the change rate of the contract signing amount compared with the bidding limit price, and the change rate of the settlement before and after review.

[0008] Furthermore, based on the set of historical projects most similar to the target project, a probability density function of the target cost effectiveness index is constructed. The long-term benchmark value of the target project is determined using the probability density function of the target cost effectiveness index, including: using a combination of K-means clustering algorithm and weighted distance metric to aggregate engineering projects with similar cost structures and characteristics. The final clustering results are used to identify the set of historical projects most similar to the target project; Based on the clustering results, a set of projects that are similar to the target project in terms of key characteristics is selected; Assign a weight based on comprehensive similarity to any sample item in the item set. Satisfies the normalization condition; Construct arbitrary target cost-effectiveness indicators for target projects using weights and kernel functions. The probability density function: ; in: Let h be the kernel function with bandwidth h. Let represent the target cost-effectiveness index value of the i-th similar project. This represents the distribution estimation function for the constructed target cost-effectiveness index; The target cost effectiveness index is determined using the expected value or mode method based on the distribution estimation function of the target cost effectiveness index. The long-term benchmark value.

[0009] Furthermore, the process of aggregating engineering projects with similar cost structures and characteristics by combining the K-means clustering algorithm with weighted distance metrics includes: constructing project clustering feature vectors, which include scheme type, main variable capacity, and topography; K-means clustering is performed using the weighted Euclidean distance between clustering features of the projects to measure the similarity between them. The formula for the weighted Euclidean distance is as follows: ; in: Indicates project With the project Weighted distance between them; The weight coefficients of the clustering features of the v-th dimension items satisfy the following conditions: , Indicates project With the project Clustering characteristics of any items between them; The number of clusters in K-means clustering can be determined using the elbow method.

[0010] Furthermore, S4 includes: Construct an eigenvector representing the target cost index and its fluctuation at any time t: ; in, This represents the cost per unit capacity index. This represents the cost index per unit capacity. This represents the transformer price index. These represent the historical volatility of the unit capacity cost index, the unit capacity component cost index, and the unit capacity component cost index, respectively. This indicates the phased change gradient of the unit capacity cost index, the unit capacity sub-item cost index, and the unit capacity sub-item cost index; In the first stage, the dominant influence of the target cost index feature vector on the deviation trend of the target cost effectiveness index is captured by linear regression, and linear prediction results are given: In the second stage, a support vector regression model is introduced to perform nonlinear modeling on the residuals of linear regression. By combining the two-stage forecasts, the short-term offset is obtained. Based on the short-term offset and the long-term benchmark, the final target cost-effectiveness index forecast result is obtained.

[0011] Furthermore, the constraints followed by the support vector regression model include: ; in, This represents the nonlinear feature mapping function in the support vector regression model. It is the tolerance error boundary of the regression residual term. It is a slack variable; The output expression of the support vector regression model is: ; in, Represents the feature vector of the target cost index for the test sample. The prediction result, where I represents the number of training samples. Let represent the Lagrange multipliers, and represent the support vector weights of the sample in the support vector regression model. It is the kernel function value between the i-th support vector and the sample to be predicted. The kernel function formula is as follows: ; in, Represent two target cost index eigenvectors in the target cost index eigenvector space; This represents the bandwidth parameter of the kernel function, which controls the smoothness of the Gaussian distribution; It is the bias term of the support vector regression model, which is the linear translation part of the output of the support vector regression model; The short-term offset obtained from the combined two-stage forecasts is as follows: ; The final target cost-effectiveness index prediction results are obtained based on short-term deviation and long-term benchmark values: ; in, The short-term offset is obtained by combining the two-stage predictions at time t. This represents the long-term benchmark value for the target cost-effectiveness indicator. This represents the predicted value of the final target cost-effectiveness indicator.

[0012] Furthermore, the risk classifier assigns cost-effectiveness indicators to each objective. Constructing the multidimensional risk feature vector required for risk assessment: ; in: It is the rth performance indicator The long-term benchmark value, This corresponds to the r-th performance indicator. short-term offset Represents the r-th performance indicator Uncertainty of the prediction interval Reflecting the impact of the volatility of the construction cost index on the r-th performance indicator Sensitivity score; Based on risk feature vectors, a logistic regression model is used to output probabilities, and the following function is established: ; in, Indicators of performance The predicted probability of belonging to a "high-risk" state. Indicators of performance The value of the j-th feature dimension, This represents the bias term in the logistic regression model. The regression coefficient of the j-th feature is represented by the bias term and the regression coefficient, which are obtained by training with historical risk-labeled samples. Based on predicted probability Risk levels are classified based on the established risk management thresholds.

[0013] Secondly, the present invention provides a hierarchical modeling and early warning system for performance indicators based on cost index fluctuation prediction, comprising: multiple interconnected network devices, each network device comprising: at least one processing unit, the processing unit being connected to a storage unit via a bus unit, the storage unit storing a computer program, and the processing unit implementing the hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction by running the computer program stored in the storage unit.

[0014] Thirdly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed, implements the method for hierarchical modeling and early warning of performance indicators based on cost index fluctuation prediction.

[0015] The technical solutions provided in the embodiments of the present invention have the following advantages compared with the prior art: This invention constructs a multi-dimensional cost index system, integrating indicators such as unit capacity cost, itemized expenses, and key equipment prices, and utilizes a CNN-LSTM model to achieve high-precision prediction. Simultaneously, support vector regression is introduced to establish a mapping relationship between the target cost index and the deviation of the target cost effectiveness indicator, making the causes of risk clear and quantifiable, and providing data support for subsequent management and response.

[0016] This invention constructs a hierarchical modeling framework by dividing the target cost performance index into long-term benchmark values ​​and short-term offsets. This effectively distinguishes between historical stability and current volatility, thereby improving the robustness and sensitivity of cost evaluation results. The long-term benchmark values ​​are obtained based on cluster analysis of similar projects, while the short-term offsets are modeled in real time based on cost index fluctuations, enhancing the ability to detect abnormal changes.

[0017] By integrating logistic regression and multi-expert scoring mechanisms, this invention constructs a risk level classifier capable of classifying the risk level of target cost-effectiveness indicators, enabling intelligent early warning and dynamic management of cost risks. The method has good versatility and can be extended to various power transmission and transformation engineering cost management scenarios, improving the scientific nature and proactiveness of enterprise project management. This technology exhibits good portability and scalability in complex engineering scenarios, significantly enhancing the intelligence level and risk control capabilities of power engineering cost management. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction, provided in an embodiment of the present invention; Figure 2 The flowchart below shows how to construct a probability density function for a target cost performance indicator based on a set of historical projects most similar to the target project, and how to determine the long-term benchmark value of the target cost performance indicator using the expected value or mode method based on the distribution estimation function of the target cost performance indicator. Figure 3 A flowchart illustrating the method of combining K-means clustering algorithm with weighted distance metric for item clustering provided in this embodiment of the invention; Figure 4 A flowchart illustrating the mapping relationship between the target cost index, the fluctuation of the target cost index, and the short-term offset of the target cost effectiveness indicator, provided by an embodiment of the present invention, combining regression analysis and support vector regression models; Figure 5 This is a flowchart provided by an embodiment of the present invention, which uses a risk classifier to evaluate the risk level of each target cost effectiveness indicator by utilizing the long-term benchmark value and short-term offset of the target cost effectiveness indicator. Figure 6 This is a schematic diagram of a hierarchical modeling and early warning device for performance indicators based on cost index fluctuation prediction, provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be noted that, in this document, 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.

[0023] Example 1 like Figure 1 As shown, this invention provides a hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction, including: S1. Based on the actual characteristics of the project, a representative target cost index is selected; combined with data from multiple sources, the target cost index is normalized according to a unified base period; then, a CNN-LSTM hybrid model is used to capture the temporal dependency of the target cost index data in order to predict various target cost indices.

[0024] This patent takes substation engineering as an example, and the target cost index includes: unit capacity cost, unit capacity itemized costs, and transformer price.

[0025] Unit capacity cost: As a core indicator for measuring the construction cost of power engineering projects, the unit capacity cost directly reflects the overall cost level of the project and provides important information about the input and output of the entire project. The selection of unit capacity cost is closely related to the project's scale, design, material costs, and labor costs, and is a key factor in measuring the economic benefits of power grid construction projects.

[0026] Unit Capacity Itemized Costs: The cost of a power project is determined by multiple itemized costs, among which civil engineering costs, equipment costs, and installation costs account for a significant proportion. By breaking down costs into unit capacity itemized costs, it is possible to analyze cost fluctuations at each stage in greater detail, providing a basis for identifying the specific reasons for cost changes in the project. Choosing this index helps to gain a deeper understanding of the impact of various costs on the total project cost, thereby achieving more precise cost control.

[0027] Transformer Prices: Transformers are crucial core equipment in power engineering, and their price fluctuations significantly impact the overall cost of power grid construction. Especially in large-scale power grid construction, transformer procurement costs typically account for a substantial proportion; therefore, changes in transformer prices directly affect the total project cost. Using transformer prices as a target cost index for forecasting can help accurately capture fluctuations in equipment procurement prices and their impact on project economics.

[0028] In summary, the selection of the target cost index is to comprehensively and meticulously grasp the cost structure and fluctuation trends of power grid engineering projects, ensuring that key factors that have a significant impact on project costs can be captured from different dimensions, thereby providing a scientific and accurate basis for cost forecasting, risk assessment, and management decisions.

[0029] Collect data on target cost indices: cost per unit capacity, itemized costs per unit capacity, and transformer prices. For each target cost index, collect relevant data over a certain historical period and standardize it to eliminate bias from data from different sources. ; in Let represent any target construction cost index at time t. This represents the average of the historical data for the target cost index. This represents the standard deviation of the historical data for the target cost index.

[0030] Standardization ensures that index data from different sources can be compared on the same benchmark.

[0031] This application employs a CNN-LSTM hybrid model to capture the temporal dependencies of target cost index data and its influencing factors, in order to predict various target cost indices.

[0032] The quantitative influencing factors of the target cost index are obtained by acquiring the time series. These factors include material prices, labor costs, equipment procurement and installation costs, construction period extension, energy prices, and transportation costs.

[0033] Factors affecting unit capacity cost: Unit capacity cost is affected by material prices, labor costs, equipment procurement and installation costs, construction period, energy prices, and geographical location. Fluctuations in material and equipment prices directly impact the project's basic costs, especially when the prices of construction materials (such as steel, cables, and transformers) rise, the unit capacity cost will increase significantly. Simultaneously, extended construction periods and rising labor costs also lead to increased costs, particularly when constructing in areas far from cities or with inconvenient transportation, where transportation costs will also increase significantly.

[0034] Factors affecting unit capacity item costs: Unit capacity item costs consist of civil engineering costs, equipment procurement, installation, and commissioning costs, and are influenced by a variety of factors. Civil engineering costs depend on the geological conditions of the project site, construction difficulty, and environmental policy requirements. Equipment costs are affected by equipment market prices, supplier competitiveness, technical specifications, and installation difficulty; price fluctuations may cause changes in item costs. Furthermore, fluctuations in material prices will also affect the overall cost of each item, especially with increased demand for equipment and installation materials.

[0035] Factors influencing transformer prices: Transformer price fluctuations are primarily affected by raw material costs, technical specifications, supplier market competitiveness, production costs, and transportation expenses. Fluctuations in the prices of raw materials used in transformers, such as copper, aluminum, and steel, directly impact their cost. High-specification transformers are typically more expensive, and fewer suppliers and more complex manufacturing processes can also drive up prices. Furthermore, differences in transportation distance and conditions, as well as fluctuations in international market demand, significantly affect transformer prices, especially during periods of global supply chain instability.

[0036] A convolutional neural network is used to extract local features of the time-series data of the factors influencing the target cost index, and a long short-term memory network is combined to capture the time-series dependency between the factors influencing the target cost index and the target cost index, thereby predicting various types of target cost indices in the future.

[0037] Convolutional neural networks (CNNs) extract local features from time-series data of a target cost index through convolution operations, enabling the discovery of the time dependency between influencing factors and the target cost index. The formula for the convolution operation in a CNN is as follows: ; in, The specific data on the factors influencing the target construction cost index at time ti. Represents the convolution kernel weights. Represents the convolution bias term. Represents the size of the convolution kernel. This represents the output of the convolutional layer. The convolutional neural network, through convolutional operations, extracts local features that help predict the target cost index based on influencing factor data. These local features then serve as the input to the LSTM part.

[0038] LSTM is a variant of RNN, specifically designed to handle long-term dependencies in time series data. The LSTM formula controls the information flow, prevents the vanishing gradient problem, and has a gating mechanism. The implementation process of each LSTM unit is as follows: The forget gate output is obtained based on the hidden state of the previous time step and the local features of the current input: ; in, This represents the activation function. Indicates the forget gate weight. Indicates the hidden state at the previous moment. and local features of the current input The splicing result, This indicates the forget gate bias.

[0039] The input gate output is obtained based on the hidden state of the previous time step and the local features of the current input: ; in, Indicates the input gate weights. This indicates the input gate bias.

[0040] The candidate layer output is obtained based on the hidden state of the previous time step and the local features of the current input: ; in, Indicates the candidate layer weights. This indicates the candidate layer bias.

[0041] The cell state is updated based on the previous cell state, input gate, forget gate, and candidate layer output: ; in, This indicates the cell state at the previous moment.

[0042] The output gate outputs based on the hidden state from the previous time step and the local features of the current input:

[0043] The current hidden state is obtained based on the current cell state and the output gate output:

[0044] in: This represents the tanh activation function.

[0045] The CNN-LSTM hybrid structure first extracts local features from the time-series data of factors influencing the target construction cost index using a CNN. Then, these extracted local features are fed into an LSTM network to further explore temporal dependencies and predict future target construction cost indices. The prediction formula is expressed as follows:

[0046] in: The input data represents the time series, including all factors influencing the target cost index. This indicates the target construction cost index for the forecast.

[0047] S2; Select target cost performance indicators closely related to the target cost index, and divide each target cost performance evaluation indicator into two levels: long-term benchmark value and short-term deviation. The long-term benchmark value is calculated by summarizing the performance of similar historical projects at similar cost levels. The short-term deviation reflects the additional fluctuations of the current project relative to the benchmark level under the influence of changes in the cost index, fluctuations in material prices, and adjustments in construction organization.

[0048] The target cost performance indicators include the accuracy of the feasibility study investment estimate, the rate of change in the feasibility study estimate before and after review, the rate of change in the preliminary budget before and after review, the rate of change in the contract amount compared to the bidding limit, and the rate of change in the settlement before and after review. These indicators cover the project's feasibility study, design, construction, and final acceptance stages, comprehensively reflecting the effectiveness of project cost management. By tracking and analyzing these indicators, it is possible to accurately predict project cost fluctuation trends, thus providing crucial information for cost control and risk warning.

[0049] The Correspondence Between Unit Capacity Cost and Performance Indicators: Unit capacity cost is directly related to the following indicators: Accuracy of Feasibility Study Investment Estimates: The accuracy of unit capacity cost directly affects investment estimates during the feasibility study stage. Accurate unit capacity cost helps in formulating a reasonable initial investment plan. Change Rate of Feasibility Study Estimates Before and After Review: When unit capacity cost fluctuates, the initial project estimate will be significantly affected, leading to an increase in the change rate between pre-review and post-review estimates. Change Rate of Contract Amount Compared to Tender Limit Price: Large fluctuations in unit capacity cost usually result in contract amounts exceeding the tender limit price, thus increasing the difference between the contract amount and the tender limit price.

[0050] The Correspondence Between Unit Capacity Itemized Costs and Performance Indicators: Unit capacity itemized costs are related to the following indicators: Feasibility Study Estimate Change Rate Before and After Review: Changes in itemized costs, especially civil engineering and equipment procurement costs, affect the estimation differences during the feasibility study stage, leading to fluctuations in the change rate before and after review. Settlement Change Rate Before and After Review: Increases or decreases in unit capacity itemized costs will affect adjustments during settlement, resulting in the magnitude of changes before and after settlement review.

[0051] The Correspondence Between Transformer Prices and Performance Indicators: Transformer prices are mainly related to the following indicators: Accuracy of Feasibility Study Investment Estimates: Fluctuations in transformer prices affect initial investment estimates, especially during the equipment procurement stage, where price changes directly lead to deviations in preliminary estimates. Rate of Change Between Contract Amount and Tender Limit Price: Fluctuations in transformer prices directly affect contract amounts, especially in the procurement of high-specification transformers, where contract amounts often exceed tender limits, increasing the difference between the two.

[0052] Furthermore, hierarchical modeling is conducted for each target cost performance indicator. The purpose of hierarchical modeling is to divide each target cost performance indicator into a long-term benchmark value and a short-term offset. These two components help to analyze the sources of change for each performance indicator in depth.

[0053] Long-term benchmarks represent the normal performance of project management over a long historical period and in similar projects, based on similar cost levels. Long-term benchmarks are calculated by summarizing the performance of historical projects and reflect the "normal" or expected level of management.

[0054] Short-term offsets reflect the additional fluctuations in the target cost-effectiveness indicators caused by external factors within a specific time period for the current project.

[0055] S3. Construct the probability density function of the target cost performance indicator based on the set of historical projects most similar to the target project, and determine the long-term benchmark value of the target cost performance indicator by means of the expected value or mode method based on the distribution estimation function of the target cost performance indicator.

[0056] In the specific implementation process, such as Figure 2 As shown, S3 includes: To improve the accuracy of project clustering analysis, S3 uses a combination of K-means clustering algorithm and weighted distance metric to more accurately group engineering projects with similar cost structures and characteristics together, such as... Figure 3 As shown, the specific process includes: For each historical engineering project, a project clustering feature vector is constructed. This feature vector includes the project type, main transformer capacity, and topography. The project clustering feature vector for the m-th engineering project is represented as: ; in, Indicates the first Clustering feature vectors of each engineering project; Indicates the first The types of solutions for each engineering project; Indicates the first The main transformer capacity of each engineering project Indicates the first The price index of topographic materials for each engineering project.

[0057] In K-means clustering, calculating the distance between each pair of items is a crucial step. To comprehensively consider the influence of multiple item clustering features on clustering, a weighted Euclidean distance is used to measure the similarity between items. The formula for the weighted Euclidean distance is as follows: ; in: Indicates project With the project Weighted distance between them; The weight coefficients of the clustering features of the v-th dimension items satisfy the following conditions: , Indicates project With the project Clustering characteristics of any items between them.

[0058] K-means clustering based on similarity.

[0059] The core task of clustering is to determine how many clusters to divide the project into. To determine the optimal number of clusters, the elbow rule is used. By plotting the sum of squared errors (SSE) under different numbers of clusters, the inflection point where the SSE drops sharply is selected, and the number of clusters corresponding to the inflection point is the ideal number.

[0060] The within-group sum of squares (WCSS) error was used as the evaluation metric. ; in, Indicates the first A collection of items within a cluster; Indicates the first The center vector of the nth cluster; the nth The formula for calculating the center vector of a cluster is as follows: .

[0061] Using the silhouette coefficient to measure cluster quality: ; in: This represents the profile coefficient of the m-th item; It is the average distance between the m-th item and other items in its own cluster; It is the average distance between the m-th item and the other most similar cluster items.

[0062] The final clustering results are used to identify the set of historical projects most similar to the target project, serving as the basis for extracting long-term benchmark values ​​and improving the accuracy and robustness of performance indicator modeling.

[0063] Based on the clustering results, select a set of projects that are similar to the target project in terms of key characteristics: ;in, For the target project, This refers to N projects that are similar to the target project in terms of key characteristics.

[0064] Any project sample Assign weights based on overall similarity. Weight Satisfy the normalization condition: .

[0065] Constructing target cost-effectiveness indicators for target projects using weights and kernel functions The probability density function: ; in: Let h be the kernel function with bandwidth h. Let represent the target cost-effectiveness index value of the i-th similar project. This represents the distribution estimation function for the constructed target cost-effectiveness index.

[0066] The target cost effectiveness index is determined using the expected value or mode method based on the distribution estimation function of the target cost effectiveness index. long-term benchmark value : Expected value: ; Mode method (applicable to skewed or multimodal distributions): .

[0067] To enhance the statistical robustness of long-term benchmark values ​​for performance indicators, a weighted kernel density regression method is introduced. This method uses nonparametric density estimation to model the distribution patterns of indicators from historically similar projects, accurately capturing central trends and fluctuation structures, and improving the model's adaptability to skewed and multimodal indicators. This method provides a solid foundation for determining the long-term reasonable range and assessing the fluctuation risk of target cost performance indicators.

[0068] S4, combining regression analysis and support vector regression models, establishes a mapping relationship between the target cost index, the fluctuation of the target cost index, and the short-term deviation of the target cost performance indicator, providing a precise basis for subsequent risk assessment and management decisions.

[0069] To characterize the relationship between the target cost index, its fluctuations, and short-term deviations in the target cost effectiveness indicator, this step employs a two-stage modeling strategy using linear regression and support vector regression to establish linear and nonlinear dynamic mapping relationships. In specific implementation, such as... Figure 4 As shown, S4 includes: First, construct an eigenvector representing the target cost index and its fluctuation at any time t: ; in, This represents the cost per unit capacity index at time t. This represents the unit capacity cost index at time t. The transformer price index represents time t. These represent the historical volatility of the unit capacity cost index, the unit capacity component cost index, and the unit capacity component cost index, respectively. This indicates the phased change gradient of the unit capacity cost index, the unit capacity sub-item cost index, and the unit capacity sub-item cost index.

[0070] In the first stage, the dominant influence of the target cost index feature vector on the deviation trend of the target cost effectiveness indicator is captured by linear regression, and the prediction results of the target cost effectiveness indicator deviation by linear regression are given: ; in, The linear forecast value representing the deviation of the phased target cost-effectiveness indicator. Represents the linear regression parameters. This represents the regression residual.

[0071] The second stage introduces a support vector regression model to nonlinearly model the residuals of linear regression. The support vector regression model uses a radial basis function kernel function, employing kernel tricks to fit the complex relationship between the residuals and the input in the high-dimensional feature space.

[0072] The constraints followed by the support vector regression model include: ; in, This represents the nonlinear feature mapping function in the support vector regression model. It is the tolerance error boundary of the regression residual term. It is a slack variable; The output expression of the support vector regression model is: ; in, Represents the feature vector of the target cost index for the test sample. The prediction result, where I represents the number of training samples. Let represent the Lagrange multipliers, and represent the support vector weights of the sample in the support vector regression model. It is the kernel function value between the i-th support vector and the sample to be predicted. This is the bias term of the support vector regression model, the linear translation part of the support vector regression model output, and the radial basis kernel function: ; in, Represent two target cost index eigenvectors in the target cost index eigenvector space; This represents the bandwidth parameter of the kernel function, which controls the smoothness of the Gaussian distribution.

[0073] The short-term offset obtained from the combined two-stage forecasts is as follows: ; Based on the short-term offset and the long-term benchmark, the prediction results of the final target cost-effectiveness index are as follows: ; in, To obtain the short-term offset by combining the two-stage forecasts, This represents the long-term benchmark value for the target cost-effectiveness indicator. This represents the predicted value of the final target cost-effectiveness indicator at time t.

[0074] S5 utilizes the long-term benchmark value and short-term offset of the target cost-effectiveness indicators to assess the risk level of each indicator using a risk classifier. The risk classifier integrates the long-term benchmark value, offset magnitude, and prediction uncertainty factors of the effectiveness indicators, and uses logistic regression or a multi-expert scoring system to calculate the risk level of each target cost-effectiveness indicator.

[0075] This step aims to identify the risk level of each target cost-effectiveness indicator based on the long-term benchmark value and short-term offset obtained from previous predictions, combined with the uncertainty information of the predictions, and to provide early warnings for potentially high-risk states. To achieve this goal, this application designs a multi-factor risk classifier based on logistic regression. The risk classifier includes three sub-modules: risk feature construction, risk probability prediction, and risk level classification.

[0076] like Figure 5 As shown, S5 includes: cost effectiveness indicators for each target. Constructing the multidimensional risk feature vector required for risk assessment: ; in: It is the rth performance indicator The long-term baseline value is determined by historical project statistics in S3. This corresponds to the r-th performance indicator. The short-term offsets are predicted by the linear regression and support vector regression models in S4. Represents the r-th performance indicator Uncertainty of the prediction interval Reflecting the impact of the volatility of the construction cost index on the r-th performance indicator Sensitivity scoring. Subsequent risk scoring is performed using risk feature vectors.

[0077] Based on risk feature vectors, a logistic regression model is used to output probabilities, and the following function is established: ; in, Indicators of performance The predicted probability of belonging to a "high-risk" state. Indicators of performance The value of the j-th feature dimension, This represents the bias term in the logistic regression model. Let represent the regression coefficient of the j-th feature. Both the bias term and the regression coefficient are obtained through training on historical risk-labeled samples.

[0078] Based on predicted probability Risk levels are classified according to the established risk management thresholds: ; in: As a high-risk threshold, one example has a value of 0.7. The risk threshold is 0.4 for one example.

[0079] Based on the risk level output, corresponding early warning prompts, strategy recommendations, or automatic correction interventions are triggered.

[0080] Example 2 like Figure 6 As shown, this embodiment of the invention provides a hierarchical modeling and early warning device for performance indicators based on cost index fluctuation prediction, comprising: at least one processing unit, the processing unit being connected to a storage unit via a bus unit, the storage unit serving as a computer-readable storage medium, which can be used to store software programs, computer-executable programs, and modules, such as the software program, computer-executable program, and module corresponding to the hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction in this embodiment of the invention. The processing unit implements the aforementioned hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction by running the software program, computer-executable program, and module stored in the storage unit, including: S1. Based on the actual characteristics of the project, representative target cost indices are selected. Then, a CNN-LSTM hybrid model is used to capture the time-series dependencies of the target cost index data and its influencing factors in order to predict various target cost indices. Among them, the target cost indices include: unit capacity cost, unit capacity item costs, and transformer price. S2, Select target cost performance indicators that are closely related to the target cost index, and divide each target cost performance evaluation indicator into two levels, including long-term benchmark value and short-term deviation. S3. Construct the probability density function of the target cost performance indicator based on the set of historical projects most similar to the target project, and determine the long-term benchmark value of the target cost performance indicator by means of the expected value or mode method based on the distribution estimation function of the target cost performance indicator. S4. Combining linear regression and support vector regression models, a mapping relationship is established between the target cost index, the fluctuation of the target cost index, and the short-term deviation of the target cost effectiveness indicator, in order to predict the short-term deviation of the target cost effectiveness indicator. S5 utilizes the predicted values ​​and offsets of the target cost performance indicators, assesses the risk level of each indicator through a risk classifier, calculates the risk level of each target cost performance indicator, outputs results based on the risk level, and triggers corresponding early warning prompts, strategy recommendations, or automatic correction intervention operations.

[0081] Of course, the computer program stored in the storage unit of the performance indicator hierarchical modeling and early warning system based on cost index fluctuation prediction provided in the embodiments of the present invention is not limited to the method operation described above, and can also execute related operations in the performance indicator hierarchical modeling and early warning method based on cost index fluctuation prediction provided in any embodiment of the present invention.

[0082] Example 3 This invention provides a computer-readable storage medium storing a computer program. When executed, the computer program implements the hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction, including: S1. Based on the actual characteristics of the project, representative target cost indices are selected. Then, a CNN-LSTM hybrid model is used to capture the time-series dependencies of the target cost index data and its influencing factors in order to predict various target cost indices. Among them, the target cost indices include: unit capacity cost, unit capacity item costs, and transformer price. S2, Select target cost performance indicators that are closely related to the target cost index, and divide each target cost performance evaluation indicator into two levels, including long-term benchmark value and short-term deviation. S3. Construct the probability density function of the target cost performance indicator based on the set of historical projects most similar to the target project, and determine the long-term benchmark value of the target cost performance indicator by means of the expected value or mode method based on the distribution estimation function of the target cost performance indicator. S4. Combining linear regression and support vector regression models, a mapping relationship is established between the target cost index, the fluctuation of the target cost index, and the short-term deviation of the target cost effectiveness indicator, in order to predict the short-term deviation of the target cost effectiveness indicator. S5 utilizes the predicted values ​​and offsets of the target cost performance indicators, assesses the risk level of each indicator through a risk classifier, calculates the risk level of each target cost performance indicator, outputs results based on the risk level, and triggers corresponding early warning prompts, strategy recommendations, or automatic correction intervention operations.

[0083] In the embodiments provided by this invention, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the structural embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, structures, or units, and may be electrical, mechanical, or other forms.

[0084] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0085] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0086] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction, characterized in that, include: S1. Based on the actual characteristics of the project, target cost indices are selected. Then, a CNN-LSTM hybrid model is used to capture the time-series dependencies of the target cost index data and its influencing factors in order to predict various target cost indices. Among them, the target cost indices include: unit capacity cost, unit capacity item costs, and transformer price. S2, Select target cost performance indicators related to the target cost index, and divide each target cost performance evaluation indicator into two levels, including long-term benchmark value and short-term deviation. S3. Construct the probability density function of the target cost performance indicator based on the set of historical projects most similar to the target project, and determine the long-term benchmark value of the target cost performance indicator by means of the expected value or mode method based on the distribution estimation function of the target cost performance indicator. S4. Combining linear regression and support vector regression models, a mapping relationship is established between the target cost index, the fluctuation of the target cost index, and the short-term deviation of the target cost effectiveness indicator, so as to use the target cost index to predict the short-term deviation of the target cost effectiveness indicator. S5 utilizes the long-term benchmark value and short-term offset of the target cost performance indicators, assesses the risk level of each indicator through a risk classifier, calculates the risk level of each target cost performance indicator, outputs the results based on the risk level, and triggers corresponding early warning prompts, strategy recommendations, or automatic correction intervention operations.

2. The hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction as described in claim 1, characterized in that, S1 includes: The quantitative influencing factors of the time series impact target cost index are obtained. These factors include material prices, labor costs, equipment procurement and installation costs, construction period extension, energy prices, and transportation costs. A convolutional neural network is used to extract local features of the time series data of the factors influencing the target cost index, and LSTM is combined to capture the time series dependency between the factors influencing the target cost index and the target cost index, so as to predict various types of target cost indices in the future. The formula for the convolution operation in a convolutional neural network is as follows: ； in, The specific data on the factors influencing the target construction cost index at time ti. Represents the convolution kernel weights. Represents the convolution bias term. Represents the size of the convolution kernel. This represents the output of the convolutional layer; the convolutional neural network extracts local features that help predict the target cost index based on influencing factor data through convolution operations, and the local features are used as the input of the LSTM part. The implementation process for each LSTM unit is as follows: The forget gate output is obtained based on the hidden state of the previous time step and the local features of the current input: ;in, This represents the activation function. Indicates the forget gate weight. Indicates the hidden state at the previous moment. and local features of the current input The splicing result, Indicates the forget gate bias; The input gate output is obtained based on the hidden state of the previous time step and the local features of the current input: ;in, Indicates the input gate weights. Indicates input gate bias; The candidate layer output is obtained based on the hidden state of the previous time step and the local features of the current input: ;in, Indicates the candidate layer weights. Indicates the candidate layer bias; The cell state is updated based on the previous cell state, input gate, forget gate, and candidate layer output: ;in, This indicates the cell state at the previous moment; The output gate outputs based on the hidden state from the previous time step and the local features of the current input: ; The current hidden state is obtained based on the current cell state and the output gate output: ,in: This represents the tanh activation function; The CNN-LSTM hybrid structure first extracts local features from the time-series data of factors influencing the target construction cost index using a CNN. Then, these extracted local features are fed into an LSTM network to further explore temporal dependencies and predict future target construction cost indices. The prediction formula is expressed as follows: ; in: The factors influencing the target construction cost index over time include all factors affecting the target construction cost index. This indicates the target construction cost index for the forecast.

3. The hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction as described in claim 1, characterized in that, The target cost performance indicators cover the feasibility study stage, design stage, construction and final acceptance stage of the project. The target cost performance indicators include the accuracy of the feasibility study investment estimate, the change rate of the feasibility study estimate before and after review, the change rate of the preliminary estimate before and after review, the change rate of the contract signing amount compared with the bidding limit price, and the change rate of the settlement before and after review.

4. The hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction according to claim 1, characterized in that, Constructing a probability density function for the target cost-effectiveness index based on a set of historical projects most similar to the target project, and using the probability density function of the target cost-effectiveness index to determine the long-term benchmark value of the target project includes: using a combination of K-means clustering algorithm and weighted distance metric to aggregate engineering projects with similar cost structures and characteristics together; Based on the clustering results, a set of projects that are similar to the target project in terms of key characteristics is selected; Assign a weight based on comprehensive similarity to any sample item in the item set. Satisfies the normalization condition; Construct arbitrary target cost-effectiveness indicators for target projects using weights and kernel functions. The probability density function: ； in: Let h be the kernel function with bandwidth h. Let represent the target cost-effectiveness index value of the i-th similar project. This represents the distribution estimation function for the constructed target cost-effectiveness index; Based on the distribution estimation function of the target cost effectiveness index, the long-term benchmark value of the target cost effectiveness index is determined by the expected value or mode method.

5. The hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction according to claim 1, characterized in that, The process of aggregating engineering projects with similar cost structures and characteristics by combining the K-means clustering algorithm with weighted distance metrics includes: constructing project cluster feature vectors, which include scheme type, main variable capacity, and topography; The similarity between items is determined by the weighted Euclidean distance between their clustering features; the formula for the weighted Euclidean distance is as follows: ； in: Indicates project With the project Weighted distance between them; The weight coefficients of the clustering features of the v-th dimension items satisfy the following conditions: , Indicates project With the project Clustering characteristics of any items between them; K-means clustering based on similarity; The number of clusters in K-means clustering can be determined using the elbow method.

6. The hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction according to claim 1, characterized in that, S4 includes: Construct an eigenvector representing the target cost index and its fluctuation at any time t: ； in, This represents the cost per unit capacity index. This represents the cost index per unit capacity. This represents the transformer price index. These represent the historical volatility of the unit capacity cost index, the unit capacity component cost index, and the unit capacity component cost index, respectively. This indicates the phased change gradient of the unit capacity cost index, the unit capacity sub-item cost index, and the unit capacity sub-item cost index; In the first stage, the dominant influence of the target cost index feature vector on the deviation trend of the target cost effectiveness index is captured by linear regression, and linear prediction results are given: In the second stage, a support vector regression model is introduced to perform nonlinear modeling on the residuals of linear regression. By combining the two-stage forecasts, the short-term offset is obtained. Based on the short-term offset and the long-term benchmark, the final target cost-effectiveness index forecast result is obtained.

7. The hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction according to claim 6, characterized in that, The constraints followed by the support vector regression model include: ； in, This represents the nonlinear feature mapping function in the support vector regression model. It is the tolerance error boundary of the regression residual term. It is a slack variable; The output expression of the support vector regression model is: ； in, Represents the feature vector of the target cost index for the test sample. The prediction result, where I represents the number of training samples. Let represent the Lagrange multipliers, and represent the support vector weights of the sample in the support vector regression model. It is the kernel function value between the i-th support vector and the sample to be predicted. The kernel function formula is as follows: ； in, Represent two target cost index eigenvectors in the target cost index eigenvector space; This represents the bandwidth parameter of the kernel function, which controls the smoothness of the Gaussian distribution; It is the bias term of the support vector regression model, which is the linear translation part of the output of the support vector regression model; The short-term offset obtained from the combined two-stage forecasts is as follows: ； The final target cost-effectiveness index prediction results are obtained based on short-term deviation and long-term benchmark values: ； in, The short-term offset is obtained by combining the two-stage predictions at time t. This represents the long-term benchmark value for the target cost-effectiveness indicator. This represents the predicted value of the final target cost-effectiveness indicator.

8. The hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction according to claim 1, characterized in that, The risk classifier provides cost-effectiveness indicators for each objective. Constructing the multidimensional risk feature vector required for risk assessment: ； in: It is the rth performance indicator The long-term benchmark value, This corresponds to the r-th performance indicator. short-term offset Represents the r-th performance indicator Uncertainty of the prediction interval Reflecting the impact of the volatility of the construction cost index on the r-th performance indicator Sensitivity score; Based on risk feature vectors, a logistic regression model is used to output probabilities, and the following function is established: ； in, Indicators of performance The predicted probability of belonging to a "high-risk" state. Indicators of performance The value of the j-th feature dimension, This represents the bias term in the logistic regression model. The regression coefficient of the j-th feature is represented by the bias term and the regression coefficient, which are obtained by training with historical risk-labeled samples. Based on predicted probability Risk levels are classified based on the established risk management thresholds.

9. A hierarchical modeling and early warning device for performance indicators based on cost index fluctuation prediction, comprising: At least one processing unit, wherein the processing unit is connected to a storage unit via a bus unit, wherein the storage unit stores a computer program, and the processing unit implements the hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction as described in any one of claims 1-8 by running the computer program stored in the storage unit.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed, it implements the hierarchical modeling and early warning method for performance indicators based on cost index fluctuation prediction as described in any one of claims 1-8.

Images