A method and system for predicting engineering market prosperity by integrating bidding big data and macroeconomic data.

By integrating bidding big data with macroeconomic methods, and utilizing a hybrid time-series network of GRU-LSTM and a spatiotemporal graph neural network ST-GNN, the problems of low accuracy and unquantified transmission paths in engineering market prosperity forecasting were solved, achieving high-precision market trend forecasting and characterization of regional transmission relationships.

CN123071002APending Publication Date: 2026-11-10SHENZHEN QUANZHI DIGITAL TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611253539.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-18
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies lack a dedicated prosperity index for the engineering market. Bidding data utilizes superficial and poorly integrated macroeconomic factors, ignoring the spatial transmission effect between regions. This results in low prediction accuracy and an inability to accurately reflect the transmission path between regions.

Method used

A method for predicting the prosperity of the engineering market is constructed by integrating bidding big data with macroeconomics. By using a hybrid time series network of GRU-LSTM and a spatiotemporal graph neural network ST-GNN, multi-scale time series features and spatial transmission features are extracted to generate a comprehensive prosperity forecast value. The diffusion index method is then used to calculate the engineering market prosperity index.

Benefits of technology

It has enabled the construction of a dedicated prosperity index for the engineering market, improving forecast accuracy and the ability to depict regional transmission relationships, and accurately reflecting the transmission paths and market trends between regions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123071002A_ABST
    Figure CN123071002A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for predicting the prosperity of the engineering market by integrating bidding big data and macroeconomic data, belonging to the field of engineering market prosperity prediction technology. The method includes the following steps: collecting multi-source heterogeneous engineering data; constructing a regional spatiotemporal feature dataset; extracting global multi-scale time-series features; extracting inter-regional spatial transmission features; fusing spatiotemporal features to generate a prosperity prediction value; and calculating an engineering market prosperity index. This invention improves prediction accuracy and the ability to characterize regional transmission relationships by constructing a dedicated engineering market prosperity index, extracting multi-scale time-series features hierarchically, and quantitatively modeling inter-regional transmission paths. It solves the problems of low prediction accuracy and inability to accurately reflect inter-regional transmission paths in existing technologies.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of engineering market prosperity forecasting technology, and in particular to an engineering market prosperity forecasting method and system that integrates bidding big data with macroeconomic data. Background Technology

[0002] Fluctuations in the engineering market have a significant impact on investment and the industrial chain, but there is currently a lack of high-frequency business climate indices specifically for this sector. The industry's usual practice is to refer to the manufacturing PMI or fixed asset investment data. However, the manufacturing PMI is based on questionnaire surveys, which are highly subjective and have limited correlation with the policy cycles and seasonal patterns unique to the engineering sector. Fixed asset investment data is only released monthly or even quarterly, and by the time the data is available, market conditions may have already changed.

[0003] In recent years, some studies have begun to use bidding data to predict the market, and bidding announcements can indeed reflect the demand for projects that will be launched in the short term. However, most practices still remain at the stage of statistically analyzing the number and amount of bids, which makes predictions prone to deviation at economic cycle turning points. Another overlooked issue is that projects in central cities will spread outward along the industrial chain. For example, a large infrastructure project will drive demand for building materials, logistics, and labor in the surrounding areas, but this ripple effect is not quantitatively described in existing methods.

[0004] In terms of time series modeling, LSTM networks are good at capturing long-term trends, but react slowly to short-term data fluctuations; GRU networks are computationally efficient, but struggle to capture patterns across multiple time scales when used alone. More importantly, both types of networks process data as independent sequences, failing to consider spatial relationships between regions. A review of patent and non-patent literature reveals that there is currently no mature solution combining bidding big data, macroeconomic indicators, and graph neural networks for predicting engineering market conditions.

[0005] In existing technologies, the engineering market lacks a dedicated prosperity index, bidding data uses superficial methods and does not deeply integrate macroeconomic factors, and ignores the spatial transmission effect between regions. Pure time series models also cannot effectively capture multi-scale patterns and non-Euclidean spatial dependence, resulting in low prediction accuracy and inability to accurately reflect the transmission path between regions. Summary of the Invention

[0006] This application provides a method and system for predicting the prosperity of the engineering market by integrating bidding big data and macroeconomics. It solves the problems of low prediction accuracy and inability to accurately reflect the transmission path between regions in the existing technology. It realizes the construction of a special prosperity index for the engineering market, the hierarchical extraction of multi-scale time series features, and the quantitative modeling of the transmission path between regions, thereby improving the prediction accuracy and the ability to depict regional transmission relationships.

[0007] This application provides a method for predicting the prosperity of the engineering market by integrating bidding big data and macroeconomic data, including: S1, collect multi-source data, including bidding and tendering big data, macroeconomic indicator data, and project lifecycle data; S2, preprocess the multi-source data to construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region; S3, construct and train a GRU-LSTM hybrid temporal network, input the globally aggregated time series features, and obtain the global temporal feature vector; S4. Construct and train the spatiotemporal graph neural network ST-GNN, using regions as nodes, constructing an adjacency matrix based on the correlation between regions, inputting the temporal features of each region, and extracting spatial transmission features. S5, integrate the global temporal feature vector with the spatial transmission feature to generate a comprehensive economic forecast value; S6. Based on the comprehensive prosperity forecast value, the engineering market prosperity index is calculated using the diffusion index method.

[0008] Furthermore, the engineering market prosperity index includes the number of new bidding projects index, the project winning bid amount index, the participation rate of bidding enterprises index, the project commencement rate index, and the project contract performance progress index; The calculation of the engineering market prosperity index based on the comprehensive prosperity forecast value using the diffusion index method includes: Based on the comprehensive economic forecast, calculate the sample proportions of the increase in the number of new bidding projects, the winning bid amount, the participation rate of bidding enterprises, the project commencement rate, and the project contract performance progress in this period compared to the previous period, as well as the sample proportions of the increase and the decrease. For each sub-item, the increase in the sample proportion is summed with half of the unchanged sample proportion to obtain the diffusion index for each sub-item. The weights of the new bidding project quantity index, project winning bid amount index, bidding enterprise participation index, project commencement rate index, and project contract performance progress index are obtained, and the weighted sum of each sub-diffusion index is calculated to obtain the comprehensive engineering market prosperity index. The comprehensive engineering market prosperity index is output after seasonal adjustment.

[0009] Furthermore, the GRU-LSTM hybrid timing network adopts a serial structure, including at least one GRU layer and at least one LSTM layer; The input is the globally aggregated time series features, and the global time series feature vector is obtained, including: The globally aggregated time series features are input into the GRU layer for processing, and the GRU layer outputs short-term fluctuation features of the time series. The output sequence of the GRU layer is used as the input of the LSTM layer. The LSTM layer processes the output sequence of the GRU layer through a gating mechanism to output long-term dependency features on the time series. The output sequence of the LSTM layer is input into the fully connected layer, and a linear transformation is performed through the fully connected layer to obtain the global temporal feature vector.

[0010] Furthermore, the spatiotemporal graph neural network ST-GNN is composed of multiple spatiotemporal convolutional blocks stacked together, each spatiotemporal convolutional block including a temporally gated convolutional layer and a graph convolutional layer; The extraction of spatial transmission features includes: The temporal features and adjacency matrix of each region are input into the spatiotemporal graph neural network ST-GNN; In the time-gated convolutional layer, one-dimensional convolution is performed on the input data along the time axis to extract the local dependency features of each region in the time dimension; In the graph convolutional layer, based on the adjacency matrix, graph convolution operations are performed on the features of each region node and its neighboring region nodes to aggregate market status information of adjacent regions; Multiple stacked spatiotemporal convolutional blocks are processed sequentially to output the spatially aware feature vector of each node.

[0011] Furthermore, the graph convolutional layer employs Chebyshev graph convolution, using Chebyshev polynomial expansion to approximate the graph convolution kernel; The step of performing graph convolution operation on the features of each region node and its neighboring region nodes based on the adjacency matrix includes: The adjacency matrix is ​​symmetrically normalized to obtain the normalized Laplace matrix; Based on the normalized Laplace matrix, the Chebyshev polynomial is calculated to obtain the terms of each order of the polynomial; The feature vectors of each region node are multiplied by the terms of the Chebyshev polynomial, and the factorials are accumulated and processed by a nonlinear activation function to obtain the spatial aggregated features of each node output by graph convolution.

[0012] Furthermore, the step of fusing the global temporal feature vector with the spatial transmission feature to generate a comprehensive economic forecast value includes: The spatial transmission characteristics of each node are processed using an attention mechanism, and the contribution weight of each node's spatial transmission characteristics to the global trend is calculated. Based on the aforementioned contribution weights, the spatial transmission characteristics of each node are weighted and summed to obtain the national spatial summary vector. The national spatial aggregation vector is concatenated with the global temporal feature vector to obtain the fused feature vector; The fused feature vector is input into a multilayer perceptron for processing. The multilayer perceptron includes hidden layers with nonlinear activation functions, and the fused feature vector is mapped to a comprehensive economic forecast value.

[0013] Furthermore, the bidding and tendering big data includes the project name, project type, investment scale, tender amount, number of bidding companies, winning bid price, region, and timestamp fields in the tender announcement; The macroeconomic indicators include the amount of fixed asset investment completed, GDP growth rate, the amount of local government special bonds issued and the progress of their use, and broad money supply. The full-cycle data of the project includes the time nodes and key indicators of each stage, such as project initiation time, design time, bidding time, construction time, and completion time, as well as the area of ​​new construction starts, the volume of projects under construction, and the output value of completed projects.

[0014] Furthermore, the preprocessing of the multi-source data to construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region includes: Named entity recognition processing is performed on the tender notice text to extract information such as project type, investment amount, and construction content; Data from different sources and frequencies are time-aligned and then aggregated to a monthly granularity. Standardize each data field; Linear interpolation was used to fill in missing values ​​in each data field; Box plots were used to identify and remove outliers in each data field. The processed multi-source data is organized into a spatiotemporal dataset with regions as nodes according to administrative division codes. The feature vector of each node is the result of splicing multi-source features under each monthly timestamp.

[0015] Furthermore, the construction of the adjacency matrix based on inter-regional correlation includes: The correlation values ​​between regions are determined by comprehensively and quantitatively based on historical bidding correlation, investment flow intensity, geographical proximity, and industrial chain correlation. Using each region as a graph node and the correlation value as the weight of the edges between nodes, a weighted adjacency matrix is ​​generated. The weighted adjacency matrix is ​​then subjected to symmetric normalization.

[0016] This application provides an engineering market prosperity forecasting system that integrates bidding big data and macroeconomic data, which is used to implement an engineering market prosperity forecasting method that integrates bidding big data and macroeconomic data. The system includes: a data acquisition module, a spatiotemporal feature construction module, a global time series feature module, a spatial transmission feature module, a feature fusion forecasting module, and a prosperity index calculation module. The data acquisition module is used to collect multi-source data, including bidding and tendering big data, macroeconomic indicator data, and project lifecycle data. The spatiotemporal feature construction module is used to preprocess the multi-source data and construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region. The global temporal feature module is used to construct and train a GRU-LSTM hybrid temporal network, inputting globally aggregated time series features to obtain a global temporal feature vector; The spatial transmission feature module is used to construct and train the spatiotemporal graph neural network ST-GNN, using regions as nodes, constructing an adjacency matrix based on the correlation between regions, inputting the temporal features of each region, and extracting spatial transmission features. The feature fusion prediction module is used to fuse the global temporal feature vector and the spatial transmission feature to generate a comprehensive economic forecast value. The prosperity index calculation module is used to calculate the engineering market prosperity index based on the comprehensive prosperity forecast value using the diffusion index method.

[0017] This application provides a method and system for predicting the prosperity of the engineering market by integrating bidding big data and macroeconomic data. It collects bidding big data, macroeconomic indicator data and engineering full-cycle data, and after preprocessing, constructs a spatiotemporal dataset containing time series features and spatial topological relationships, enabling the model to simultaneously perceive the operating status of the engineering market from three dimensions: project intention, external environment and implementation progress.

[0018] In this process, an engineering market prosperity index was constructed as a forward-looking comprehensive indicator. Five sub-indicators were selected, including the number of new bidding projects, the amount of project winning bids, the participation rate of bidding enterprises, the project commencement rate, and the progress of project contract performance. The diffusion index method was used to calculate the values ​​of each sub-indicator and synthesize the comprehensive index according to the preset weights. 50% was used as the dividing line between prosperity and decline to provide an intuitive quantitative judgment basis, filling the technical gap of the long-term lack of dedicated prosperity indicators in this field.

[0019] In the prediction process, a GRU-LSTM cascaded hybrid network is employed. The GRU layer is sensitive to short-term fluctuations and is highly efficient in training, while the LSTM layer memorizes long-term historical information through a fine-grained gating mechanism. The two layers work together to extract global multi-scale features from the aggregated time-series data. Spatially, the method constructs a graph structure with provinces or cities as nodes. Edge weights are quantified based on historical bidding correlation, investment flow intensity, and geographical adjacency. Local dependencies are extracted along the time axis through time-gated convolutional layers in the spatiotemporal graph neural network. The graph convolutional layers use Chebyshev multinomials to approximate the market state of adjacent areas converging in the spatial dimension, explicitly characterizing the ripple effect of central projects on surrounding areas.

[0020] Finally, the global temporal features and the spatial perception features of each node are fused across scales through an attention mechanism to output the economic forecast values ​​for the whole country and various regions, which effectively improves the prediction accuracy and the ability to identify the inflection point of the cycle. Attached Figure Description

[0021] Figure 1 A flowchart illustrating a method for predicting engineering market trends that integrates bidding big data with macroeconomic data, provided for embodiments of this application; Figure 2 This is a schematic diagram of the structure of an engineering market prosperity forecasting system that integrates bidding big data and macroeconomic data, provided as an embodiment of this application. Detailed Implementation

[0022] This application provides a method and system for predicting the prosperity of the engineering market by integrating bidding big data and macroeconomic data. It solves the problems of low prediction accuracy and inability to accurately reflect the transmission path between regions in the prior art. By constructing a special prosperity index for the engineering market, extracting multi-scale time series features in a hierarchical manner, and quantitatively modeling the transmission path between regions, it improves the prediction accuracy and the ability to depict regional transmission relationships.

[0023] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0024] like Figure 1 The diagram shown is a flowchart of a method for predicting the prosperity of the engineering market by integrating bidding big data and macroeconomic data, provided in an embodiment of this application. This method is applied to an engineering market prosperity prediction system that integrates bidding big data and macroeconomic data, and includes the following steps: S1, Collect multi-source heterogeneous engineering data: Collect multi-source data, including bidding and tendering big data, macroeconomic indicator data, and engineering full-cycle data.

[0025] The accuracy of engineering market forecasting highly depends on the quality and coverage of data sources. Bidding and tendering big data comes from public resource trading platforms across the country, collected incrementally daily, reflecting the real-time dynamics of project initiation and contracting. Macroeconomic indicators are regularly released by the National Bureau of Statistics and the Ministry of Finance, providing directional guidance from a macroeconomic perspective for forecasting. Full-cycle engineering data covers all key stages from project initiation to completion, supplementing information on construction transformation and output that bidding data cannot cover. The combination of these three elements allows for a comprehensive depiction of the engineering market's operational status from three dimensions: project intent, external environment, and implementation progress. When organizing data by administrative division codes, each province or city across the country is treated as an independent node. The feature vector of each node is composed of features from multiple sources at a unified monthly granularity, enabling the model to simultaneously handle comprehensive assessments at the national scale and differentiated analyses at the regional scale.

[0026] Furthermore, the bidding and tendering big data includes the project name, project type, investment scale, tender amount, number of bidding companies, winning bid price, region, and timestamp fields in the tender announcement; The macroeconomic indicators include the amount of fixed asset investment completed, GDP growth rate, the amount of local government special bonds issued and the progress of their use, and broad money supply. The full-cycle data of the project includes the time nodes and key indicators of each stage, such as project initiation time, design time, bidding time, construction time, and completion time, as well as the area of ​​new construction starts, the volume of projects under construction, and the output value of completed projects.

[0027] In this embodiment, the tender notice text contains numerous instances of inconsistent formatting and non-standard terminology, necessitating the extraction of key information such as project type, investment amount, and construction content using natural language processing technology. The region and timestamp fields are used for subsequent data organization by region and data alignment by time. The project type field distinguishes between subcategories of engineering projects, such as housing construction, municipal engineering, transportation, and water conservancy. Among macroeconomic indicators, the completed fixed asset investment and GDP growth rate are released by the National Bureau of Statistics, while the issuance and usage progress of local government special bonds are released by the Ministry of Finance. These data are typically monthly or quarterly and need to be uniformly aligned to the monthly granularity. Broad money supply reflects macroeconomic liquidity and serves as a leading indicator for medium- and long-term engineering investment. The time nodes at each stage in the full-cycle engineering data are used to calculate derived indicators such as commencement rate and contract performance progress. The newly started construction area and the volume of projects under construction directly reflect the intensity of construction activities, while the completed output value reflects the output results of engineering activities.

[0028] S2, Constructing a regional spatiotemporal feature dataset: Preprocess the multi-source data and construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region.

[0029] Furthermore, the preprocessing of the multi-source data to construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region includes: Named entity recognition processing is performed on the tender notice text to extract information such as project type, investment amount, and construction content; Data from different sources and frequencies are time-aligned and then aggregated to a monthly granularity. Standardize each data field; Linear interpolation was used to fill in missing values ​​in each data field; Box plots were used to identify and remove outliers in each data field. The processed multi-source data is organized into a spatiotemporal dataset with regions as nodes according to administrative division codes. The feature vector of each node is the result of splicing multi-source features under each monthly timestamp.

[0030] In this embodiment, the named entity recognition model needs to be fine-tuned based on annotated corpora in the engineering field to improve the accuracy of recognizing specific entities such as project type and investment amount. The specific process includes word segmentation and part-of-speech tagging of the announcement text, followed by the use of a sequence labeling model to identify entity categories in the text. The time alignment of data from different sources needs to be selected according to the intraday or intraweekly distribution characteristics of each indicator. For example, tender announcements are usually released on weekdays, while macroeconomic data is concentrated on specific dates. Linear interpolation is suitable for scenarios where the trend of changes before and after missing values ​​is relatively smooth. Missing values ​​near abrupt change points need to be carefully handled in conjunction with auxiliary information. Box plots using 3 times the IQR as the outlier threshold are suitable for approximately normally distributed data. For skewed data fields, a logarithmic transformation can be performed first. When organizing data by administrative division coding, the consistency of the coding system should be ensured to avoid regional nodes being unable to be correctly associated due to coding differences.

[0031] S3, Extract global multi-scale temporal features: Construct and train a GRU-LSTM hybrid temporal network, input the globally aggregated time series features, and obtain the global temporal feature vector.

[0032] Furthermore, the GRU-LSTM hybrid timing network adopts a serial structure, including at least one GRU layer and at least one LSTM layer; The input is the globally aggregated time series features, and the global time series feature vector is obtained, including: The globally aggregated time series features are input into the GRU layer for processing, and the GRU layer outputs short-term fluctuation features of the time series. The output sequence of the GRU layer is used as the input of the LSTM layer. The LSTM layer processes the output sequence of the GRU layer through a gating mechanism to output long-term dependency features on the time series. The output sequence of the LSTM layer is input into the fully connected layer, and a linear transformation is performed through the fully connected layer to obtain the global temporal feature vector.

[0033] In this embodiment, the GRU structure is simplified, including update and reset gates, resulting in higher computational efficiency and greater sensitivity to recent changes in the input sequence, making it suitable for capturing short-term fluctuations and abrupt changes in bidding data. The LSTM structure includes input, forget, and output gates, offering a more refined gating mechanism that effectively controls the retention or discarding of information in long sequences, making it suitable for capturing long-term dependent patterns such as macroeconomic cycles and seasonality. After concatenation, the GRU layer preprocesses and filters noise from the original sequence, and its output complete time series is then fed into the LSTM layer for deep modeling, demonstrating a clear division of labor. The fully connected layer compresses the high-dimensional time series representation output by the LSTM into a fixed-dimensional global feature vector, which serves as the carrier of global information in subsequent fusion steps, participating in cross-scale fusion. During training, a sliding time window is used to construct samples, predicting the next 1 to 3 months using data from the past 12 months. The Adam optimizer and mean squared error loss function, combined with an early stopping mechanism, prevent overfitting.

[0034] S4, Extracting Spatial Transmission Features Between Regions: Construct and train a spatiotemporal graph neural network ST-GNN, using regions as nodes, construct an adjacency matrix based on the inter-regional correlation, input the temporal features of each region, and extract spatial transmission features.

[0035] Furthermore, the spatiotemporal graph neural network ST-GNN is composed of multiple spatiotemporal convolutional blocks stacked together, each spatiotemporal convolutional block including a temporally gated convolutional layer and a graph convolutional layer; The extraction of spatial transmission features includes: The temporal features and adjacency matrix of each region are input into the spatiotemporal graph neural network ST-GNN; In the time-gated convolutional layer, one-dimensional convolution is performed on the input data along the time axis to extract the local dependency features of each region in the time dimension; In the graph convolutional layer, based on the adjacency matrix, graph convolution operations are performed on the features of each region node and its neighboring region nodes to aggregate market status information of adjacent regions; Multiple stacked spatiotemporal convolutional blocks are processed sequentially to output the spatially aware feature vector of each node.

[0036] In this embodiment, the regional graph structure constructed by this method uses provincial-level administrative regions across the country as nodes, with a fixed number of nodes. Each node has a feature vector associated with it at each time step, and the entire input constitutes a dynamic graph sequence unfolding along the time axis. Temporally gated convolutional layers independently perform one-dimensional convolutions on each node along the time dimension, extracting the temporal evolution features of each region. Graph convolutional layers propagate information along the spatial dimension at each time slice, enabling each region to receive state information from neighboring nodes. After stacking multiple spatiotemporal convolutional blocks, the model can capture the complex dependencies of spatiotemporal coupling at a deeper level. The final output spatially aware feature vector of each node contains both the region's own historical evolution information and the influence transmitted from related regions through the transmission path, allowing the model to explicitly characterize the ripple-like diffusion effect of large-scale projects in central cities on surrounding markets.

[0037] Furthermore, the graph convolutional layer employs Chebyshev graph convolution, using Chebyshev polynomial expansion to approximate the graph convolution kernel; The step of performing graph convolution operation on the features of each region node and its neighboring region nodes based on the adjacency matrix includes: The adjacency matrix is ​​symmetrically normalized to obtain the normalized Laplace matrix; Based on the normalized Laplace matrix, the Chebyshev polynomial is calculated to obtain the terms of each order of the polynomial; The feature vectors of each region node are multiplied by the terms of the Chebyshev polynomial, and the factorials are accumulated and processed by a nonlinear activation function to obtain the spatial aggregated features of each node output by graph convolution.

[0038] In the specific implementation process, a regional spatial map is constructed using provincial-level administrative regions as graph nodes. Edge weights are calculated comprehensively based on the number of inter-provincial bidding successes of construction companies, investment flow intensity, and geographical adjacency relationships, forming a weighted adjacency matrix and undergoing symmetric normalization. Historical data is divided into training and validation / test sets according to time. The GRU-LSTM hybrid network is configured with two GRU layers followed by two LSTM layers. The GRU has 128 hidden units and returns a complete time series, while the LSTM has 64 hidden units. After passing through fully connected layers, it outputs a 32-dimensional global temporal feature vector. The ST-GNN stacks two spatiotemporal convolutional blocks with a temporally gated convolutional kernel size of 3. The graph convolution is approximated using a Chebyshev polynomial with K=3. The 18-dimensional feature sequences and adjacency matrices of each province are input into the ST-GNN to obtain a 32-dimensional spatially perceived feature vector for each province. The feature fusion layer calculates the attention weights of spatial features for each province, concatenates the weighted national spatial summary vector with global temporal features, and inputs it into a hidden layer containing 64 neurons and a ReLU activation function. Finally, a linear mapping is used to obtain the national comprehensive economic climate value and the economic climate values ​​for each province for the next month. Training uses mean squared error as the loss function, with the Adam optimizer learning rate set to 0.001, a batch size of 64, and a total of 200 training epochs. An early stopping mechanism terminates training when the validation set loss does not decrease for 15 consecutive epochs.

[0039] In this embodiment, spectral domain graph convolution defines the convolution operation on the graph as filtering the graph signal in the frequency domain. However, directly calculating the eigenvalue decomposition of the graph Laplacian matrix is ​​computationally too expensive for large-scale graph structures. Chebyshev polynomial approximation calculates the polynomial function of the Laplacian matrix recursively, avoiding eigenvalue decomposition. The computational complexity is linearly related to the number of edges in the graph. Setting the order K=3 means that the feature aggregation range of each node covers its neighboring nodes within the third order. This range can effectively cover the multi-level transmission path of large projects in the provincial capital city, which first spreads to other cities in the province and then further affects border cities of adjacent provinces. In specific calculations, the symmetric normalized Laplacian matrix is ​​calculated first, and then the 0th to Kth order terms are calculated sequentially according to the recursive relationship of the Chebyshev polynomial. Each order term is multiplied by the learnable weight parameters and accumulated, and then output after nonlinear activation.

[0040] Furthermore, the construction of the adjacency matrix based on inter-regional correlation includes: The correlation values ​​between regions are determined by comprehensively and quantitatively based on historical bidding correlation, investment flow intensity, geographical proximity, and industrial chain correlation. Using each region as a graph node and the correlation value as the weight of the edges between nodes, a weighted adjacency matrix is ​​generated. The weighted adjacency matrix is ​​then subjected to symmetric normalization.

[0041] In this embodiment, the calculation of historical bidding correlation is based on the actual records of cross-provincial bidding behavior of construction enterprises, reflecting the business network and resource allocation patterns at the enterprise level in the engineering market. This cross-regional business flow is a micro-realization mechanism of spatial transmission effects. Investment flow intensity characterizes capital-level correlation through inter-regional capital flow data. Capital flow often precedes physical engineering activities, thus providing a leading signal for transmission effects. Geographical adjacency reflects the similarity of geographically adjacent regions in terms of climate conditions, policy environment, and infrastructure networks. This similarity leads to convergence in engineering market demand among adjacent regions. Industrial chain correlation is quantified by analyzing the interdependence of upstream and downstream links such as building material supply, professional subcontracting, and equipment leasing in various regions, enabling changes in demand in upstream regions to be transmitted to downstream regions along the industrial chain. When comprehensively quantifying the four dimensions, normalization processing is required to eliminate dimensional differences, ultimately generating a weighted adjacency matrix and performing symmetric Laplace normalization.

[0042] S5, Integrate spatiotemporal features to generate economic forecast value: Integrate the global temporal feature vector with the spatial transmission feature to generate a comprehensive economic forecast value.

[0043] Furthermore, the step of fusing the global temporal feature vector with the spatial transmission feature to generate a comprehensive economic forecast value includes: The spatial transmission characteristics of each node are processed using an attention mechanism, and the contribution weight of each node's spatial transmission characteristics to the global trend is calculated. Based on the aforementioned contribution weights, the spatial transmission characteristics of each node are weighted and summed to obtain the national spatial summary vector. The national spatial aggregation vector is concatenated with the global temporal feature vector to obtain the fused feature vector; The fused feature vector is input into a multilayer perceptron for processing. The multilayer perceptron includes hidden layers with nonlinear activation functions, and the fused feature vector is mapped to a comprehensive economic forecast value.

[0044] In this embodiment, the weights of different regions' influence on the national market sentiment are not fixed at different times. When large-scale infrastructure projects are concentrated in the eastern coastal areas, these regions may become the main driving force of the national market; while when urbanization accelerates in the central and western regions, the main driver of growth may shift. The attention mechanism can dynamically calculate the importance of each region at the current prediction time. Regions with large weight values ​​indicate that their market status has a strong indicative or driving effect on the overall national trend. The calculation method is to map the spatial transmission features of each node through a fully connected layer and then normalize them using softmax to obtain a weight distribution with a sum of 1. After weighted summation, a national spatial summary vector is obtained, which centrally reflects the comprehensive impact of the transmission effects of each region on the national market. The concatenation operation merges the global time-series feature vector and the national spatial summary vector in the feature dimension. The multilayer perceptron performs a nonlinear transformation on the fused feature and then performs a linear mapping to obtain the comprehensive market sentiment prediction value.

[0045] S6, Calculate the engineering market prosperity index: Based on the comprehensive prosperity forecast value, the engineering market prosperity index is calculated using the diffusion index method.

[0046] Furthermore, the engineering market prosperity index includes the number of new bidding projects index, the project winning bid amount index, the participation rate of bidding enterprises index, the project commencement rate index, and the project contract performance progress index; The calculation of the engineering market prosperity index based on the comprehensive prosperity forecast value using the diffusion index method includes: Based on the comprehensive economic forecast, calculate the sample proportions of the increase in the number of new bidding projects, the winning bid amount, the participation rate of bidding enterprises, the project commencement rate, and the project contract performance progress in this period compared to the previous period, as well as the sample proportions of the increase and the decrease. For each sub-item, the increase in the sample proportion is summed with half of the unchanged sample proportion to obtain the diffusion index for each sub-item. The comprehensive engineering market prosperity index is obtained by weighting and summing the diffusion indices of each sub-item according to the following weights: the number of new bidding projects index (30%), the project winning bid amount index (30%), the participation rate of bidding enterprises index (20%), the project commencement rate index (10%), and the project contract performance progress index (10%). The comprehensive engineering market prosperity index is output after seasonal adjustment.

[0047] In this embodiment, the core of the diffusion index method lies in measuring the direction of change of each sub-indicator on a cross-section rather than the magnitude of change, which makes the index naturally sensitive to trend turning points. The five sub-indicators correspond to the complete chain of the engineering market from demand release, transaction completion, competition participation, construction conversion to execution and delivery, all calculated based on objective data. The number of new bidding projects and the amount of winning bids directly reflect the incremental changes in market demand, each accounting for 30% of the weight. Bidding enterprise participation reflects the intensity of market competition, accounting for 20%. The project commencement rate and contract performance progress reflect the conversion efficiency from bidding to implementation, each accounting for 10%. The X-13 seasonal adjustment method can eliminate the cyclical interference of mobile holidays such as the Spring Festival and seasonal construction peaks and off-peak seasons on the index, ensuring that the output engineering PMI truly reflects fundamental changes rather than calendar effects. The 50% threshold, the dividing line between expansion and contraction, inherits the classic PMI compilation convention; a value above this indicates that the engineering market is in the expansion range, and a value below this indicates that it is in the contraction range.

[0048] The method for predicting engineering market prosperity by integrating bidding big data and macroeconomic data also includes: Based on the comprehensive prosperity forecast, the sub-diffusion index of each region is calculated, and the engineering market prosperity index of each region is synthesized according to the weight of each sub-item. Seasonally adjust the synthesized regional engineering market prosperity index; Visualized data is generated based on the engineering market prosperity index of each region. This visualized data is used to support filtering and viewing by region and viewing trend changes by time dimension.

[0049] In this embodiment, the calculation method for the engineering market prosperity index in each region is the same as that for the national index. Specifically, for each region, a sub-diffusion index is calculated for the number of new bidding projects, the amount of winning bids, the participation rate of bidding enterprises, the project commencement rate, and the progress of project contract performance. These sub-indices are then weighted and synthesized with equal weights and seasonally adjusted using X-13. For visualization, a heatmap is generated for each region's prosperity index according to a preset color mapping rule, with warm colors corresponding to high prosperity values ​​and cool colors corresponding to low prosperity values. The time series of each region's index is smoothed and displayed as a line chart, facilitating observation of the synchronicity, leading, or lagging relationship of prosperity trends between different regions. Market heat warnings are generated based on the comparison between the prosperity index and the 50% threshold (the boom-bust line) and the direction of their first-order difference: a downward warning is triggered when the prosperity index remains below the threshold for several consecutive months and the difference remains negative; a recovery warning is triggered when the prosperity index breaks through the threshold from below. The interactive dashboard provides controls for region filtering and time dimension switching, allowing users to view information as needed. like Figure 2The diagram shown is a structural schematic of an engineering market prosperity forecasting system that integrates bidding big data and macroeconomic data according to an embodiment of this application. The engineering market prosperity forecasting system that integrates bidding big data and macroeconomic data according to an embodiment of this application includes: a data acquisition module, a spatiotemporal feature construction module, a global time series feature module, a spatial transmission feature module, a feature fusion forecasting module, and a prosperity index calculation module. The data acquisition module is used to collect multi-source data, including bidding and tendering big data, macroeconomic indicator data, and project lifecycle data. The spatiotemporal feature construction module is used to preprocess the multi-source data and construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region. The global temporal feature module is used to construct and train a GRU-LSTM hybrid temporal network, inputting globally aggregated time series features to obtain a global temporal feature vector; The spatial transmission feature module is used to construct and train the spatiotemporal graph neural network ST-GNN, using regions as nodes, constructing an adjacency matrix based on the correlation between regions, inputting the temporal features of each region, and extracting spatial transmission features. The feature fusion prediction module is used to fuse the global temporal feature vector and the spatial transmission feature to generate a comprehensive economic forecast value. The prosperity index calculation module is used to calculate the engineering market prosperity index based on the comprehensive prosperity forecast value using the diffusion index method.

[0050] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0051] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0052] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0053] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0054] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0055] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the prosperity of the engineering market by integrating bidding big data and macroeconomic data, characterized in that, Includes the following steps: S1, collect multi-source data, including bidding and tendering big data, macroeconomic indicator data, and project lifecycle data; S2, preprocess the multi-source data to construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region; S3, construct and train a GRU-LSTM hybrid temporal network, input the globally aggregated time series features, and obtain the global temporal feature vector; S4. Construct and train the spatiotemporal graph neural network ST-GNN, using regions as nodes, constructing an adjacency matrix based on the correlation between regions, inputting the temporal features of each region, and extracting spatial transmission features. S5, integrate the global temporal feature vector with the spatial transmission feature to generate a comprehensive economic forecast value; S6. Based on the comprehensive prosperity forecast value, the engineering market prosperity index is calculated using the diffusion index method.

2. The method for predicting engineering market prosperity by integrating bidding big data and macroeconomics as described in claim 1, characterized in that, The engineering market prosperity index includes the number of new bidding projects, the project winning bid amount, the participation rate of bidding enterprises, the project commencement rate, and the project contract performance progress index. The calculation of the engineering market prosperity index based on the comprehensive prosperity forecast value using the diffusion index method includes: Based on the comprehensive economic forecast, calculate the sample proportions of the increase in the number of new bidding projects, the winning bid amount, the participation rate of bidding enterprises, the project commencement rate, and the project contract performance progress in this period compared to the previous period, as well as the sample proportions of the increase and the decrease. For each sub-item, the increase in the sample proportion is summed with half of the unchanged sample proportion to obtain the diffusion index for each sub-item. The weights of the new bidding project quantity index, project winning bid amount index, bidding enterprise participation index, project commencement rate index, and project contract performance progress index are obtained, and the weighted sum of each sub-diffusion index is calculated to obtain the comprehensive engineering market prosperity index. The comprehensive engineering market prosperity index is output after seasonal adjustment.

3. The method for predicting engineering market prosperity by integrating bidding big data and macroeconomics as described in claim 1, characterized in that, The GRU-LSTM hybrid time-series network adopts a serial structure, including at least one GRU layer and at least one LSTM layer; The input is the globally aggregated time series features, and the global time series feature vector is obtained, including: The globally aggregated time series features are input into the GRU layer for processing, and the GRU layer outputs short-term fluctuation features of the time series. The output sequence of the GRU layer is used as the input of the LSTM layer. The LSTM layer processes the output sequence of the GRU layer through a gating mechanism to output long-term dependency features on the time series. The output sequence of the LSTM layer is input into the fully connected layer, and a linear transformation is performed through the fully connected layer to obtain the global temporal feature vector.

4. The method for predicting engineering market prosperity by integrating bidding big data and macroeconomic data as described in claim 1, characterized in that, The spatiotemporal graph neural network ST-GNN is composed of multiple spatiotemporal convolutional blocks stacked together, each spatiotemporal convolutional block including a temporally gated convolutional layer and a graph convolutional layer; The extraction of spatial transmission features includes: The temporal features and adjacency matrix of each region are input into the spatiotemporal graph neural network ST-GNN; In the time-gated convolutional layer, one-dimensional convolution is performed on the input data along the time axis to extract the local dependency features of each region in the time dimension; In the graph convolutional layer, based on the adjacency matrix, graph convolution operations are performed on the features of each region node and its neighboring region nodes to aggregate market status information of adjacent regions; Multiple stacked spatiotemporal convolutional blocks are processed sequentially to output the spatially aware feature vector of each node.

5. The method for predicting engineering market prosperity by integrating bidding big data and macroeconomics as described in claim 4, characterized in that, The graph convolutional layer employs Chebyshev graph convolution, and uses Chebyshev polynomial expansion to approximate the graph convolution kernel; The step of performing graph convolution operation on the features of each region node and its neighboring region nodes based on the adjacency matrix includes: The adjacency matrix is ​​symmetrically normalized to obtain the normalized Laplace matrix; Based on the normalized Laplace matrix, the Chebyshev polynomial is calculated to obtain the terms of each order of the polynomial; The feature vectors of each region node are multiplied by the terms of the Chebyshev polynomial, and the factorials are accumulated and processed by a nonlinear activation function to obtain the spatial aggregated features of each node output by graph convolution.

6. The method for predicting engineering market prosperity by integrating bidding big data and macroeconomics as described in claim 1, characterized in that, The process of fusing the global temporal feature vector with the spatial transmission feature to generate a comprehensive economic forecast value includes: The spatial transmission characteristics of each node are processed using an attention mechanism, and the contribution weight of each node's spatial transmission characteristics to the global trend is calculated. Based on the aforementioned contribution weights, the spatial transmission characteristics of each node are weighted and summed to obtain the national spatial summary vector. The national spatial aggregation vector is concatenated with the global temporal feature vector to obtain the fused feature vector; The fused feature vector is input into a multilayer perceptron for processing. The multilayer perceptron includes hidden layers with nonlinear activation functions, and the fused feature vector is mapped to a comprehensive economic forecast value.

7. The method for predicting engineering market prosperity by integrating bidding big data and macroeconomics as described in claim 1, characterized in that, The bidding and tendering big data includes the project name, project type, investment scale, tender amount, number of bidding companies, winning bid price, region, and timestamp fields in the tender announcement; The macroeconomic indicators include the amount of fixed asset investment completed, GDP growth rate, the amount of local government special bonds issued and the progress of their use, and broad money supply. The full-cycle data of the project includes the time nodes and key indicators of each stage, such as project initiation time, design time, bidding time, construction time, and completion time, as well as the area of ​​new construction starts, the volume of projects under construction, and the output value of completed projects.

8. The method for predicting engineering market prosperity by integrating bidding big data and macroeconomics as described in claim 1, characterized in that, The preprocessing of the multi-source data to construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region includes: Named entity recognition processing is performed on the tender notice text to extract information such as project type, investment amount, and construction content; Data from different sources and frequencies are time-aligned and then aggregated to a monthly granularity. Standardize each data field; Linear interpolation was used to fill in missing values ​​in each data field; Box plots were used to identify and remove outliers in each data field. The processed multi-source data is organized into a spatiotemporal dataset with regions as nodes according to administrative division codes. The feature vector of each node is the result of splicing multi-source features under each monthly timestamp.

9. The method for predicting engineering market prosperity by integrating bidding big data and macroeconomics as described in claim 1, characterized in that, The construction of the adjacency matrix based on inter-regional correlation includes: The correlation values ​​between regions are determined by comprehensively and quantitatively based on historical bidding correlation, investment flow intensity, geographical proximity, and industrial chain correlation. Using each region as a graph node and the correlation value as the weight of the edges between nodes, a weighted adjacency matrix is ​​generated. The weighted adjacency matrix is ​​then subjected to symmetric normalization.

10. A system for predicting the engineering market sentiment by integrating bidding big data and macroeconomic data, used to implement the engineering market sentiment prediction method for integrating bidding big data and macroeconomic data as described in any one of claims 1-9, characterized in that, include: Data acquisition module, spatiotemporal feature construction module, global time series feature module, spatial transmission feature module, feature fusion prediction module, and business climate index calculation module; The data acquisition module is used to collect multi-source data, including bidding and tendering big data, macroeconomic indicator data, and project lifecycle data. The spatiotemporal feature construction module is used to preprocess the multi-source data and construct a spatiotemporal dataset containing temporal features and spatial topological relationships by region. The global temporal feature module is used to construct and train a GRU-LSTM hybrid temporal network, inputting globally aggregated time series features to obtain a global temporal feature vector; The spatial transmission feature module is used to construct and train the spatiotemporal graph neural network ST-GNN, using regions as nodes, constructing an adjacency matrix based on the correlation between regions, inputting the temporal features of each region, and extracting spatial transmission features. The feature fusion prediction module is used to fuse the global temporal feature vector and the spatial transmission feature to generate a comprehensive economic forecast value. The prosperity index calculation module is used to calculate the engineering market prosperity index based on the comprehensive prosperity forecast value using the diffusion index method.