Construction cost and material price list measuring and calculating method based on artificial intelligence

Abstract

The invention discloses a project cost and material price list calculation method based on artificial intelligence, and the method comprises the steps: obtaining resource conditions and project type information matched with project demands from historical data through querying a preset database, and obtaining an initial data set; according to the initial data set, grouping resource conditions and engineering types by adopting a clustering analysis method, and determining a grouped feature set; if the dynamic change in the grouped feature set exceeds a preset threshold value, training a use rule model through a random forest algorithm to obtain a trained use rule model; for the trained use rule model, inputting current market supply and demand data, judging the influence degree of market supply and demand on price prediction, and obtaining an influence weight; and aiming at the predicted price value, combining the use rule model output, generating a final use amount and price combination of the material list, judging whether the combination meets a cost control requirement, and obtaining optimized list data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engineering cost, and in particular to a method for measuring and calculating a material price list of engineering cost based on artificial intelligence. BACKGROUND

[0002] In the field of engineering construction, accurate measurement of engineering cost plays a crucial role in project decision-making and cost control. In particular, during the preparation of the material price list, accurate prediction of material consumption and price is directly related to the economic benefits and feasibility of the project. Research in this field is not only the core demand of industry development, but also an important direction to promote the digital transformation of the construction industry. However, traditional cost measurement methods often face many challenges, and innovative means are needed to break through existing bottlenecks.

[0003] Currently, the preparation of the material list in engineering cost relies on manual experience or simple statistical tools, which is not sufficient when faced with complex projects. Especially when dealing with price fluctuations in different regions and time periods, as well as the special needs of various engineering projects, existing methods are difficult to fully capture the deep rules affecting material selection and price. This limitation not only leads to deviations in measurement results, but also increases the risk of project costs due to the lack of adaptability to dynamic changes.

[0004] Further, the core technical difficulties in this field mainly focus on how to accurately grasp the regularity of material use. Material use regularity is influenced by a variety of factors, such as resource conditions in the project location, structural characteristics of engineering types, etc. These factors are intertwined, making it difficult to extract universal rules simply by relying on traditional analysis methods. The deeper problem is that the trend of material price changes is often closely related to the implicit patterns in historical data, but existing technologies often fail to combine the complex associations of historical data with the actual needs of specific projects when mining these patterns. For example, in a large cross-regional engineering project, the consumption of a certain type of material may vary due to regional resource differences, while the price is affected by market supply and demand fluctuations. If these factors cannot be accurately matched, the list preparation may have large errors, leading to budget overruns or resource waste. SUMMARY

[0005] The present application provides a method for measuring and calculating a material price list of engineering cost based on artificial intelligence to solve the above technical problems.

[0006] The technical solution of the present application is as follows: An engineering cost material price list calculation method based on artificial intelligence, comprising: step S101: obtaining resource conditions and engineering type information matched with project requirements from historical data by querying a preset database to obtain an initial data set; step S102: grouping resource conditions and engineering types according to the initial data set to determine a feature set after grouping; step S103: if dynamic changes in the feature set after grouping exceed a preset threshold, training a usage rule model to obtain a trained usage rule model; step S104: inputting current market supply and demand data into the trained usage rule model to judge the influence degree of market supply and demand on price prediction to obtain an influence weight; step S105: introducing a multi-factor price driving analysis based on an attention mechanism, combining the influence weight and time series decomposition to extract potential periodic characteristics and external impact effects of price fluctuations to generate a dynamic price adjustment factor; and step S106: using an adaptive prediction engine to model price prediction as a sequence decision problem, and the model autonomously learns an optimal prediction strategy according to market state and dynamic adjustment signals to output a dynamic price interval with risk perception capability.

[0007] Compared with the prior art, the present application has the following beneficial effects: By introducing multi-factor analysis, time series decomposition, attention mechanism and reinforcement learning, the present application can more comprehensively capture the dynamic changes of material prices, improve prediction accuracy, and more accurately predict material usage and prices compared with traditional methods. Through dynamic price adjustment factors and adaptive prediction engines, the prediction strategy can be adjusted in real time to adapt to market supply and demand changes and external shocks, making the prediction results more timely and reliable, and reducing prediction errors caused by market fluctuations. Through automatic and intelligent data processing and prediction processes, the present application can quickly generate the final usage and price combination of the material list, greatly improving decision-making efficiency, saving time and labor costs compared with traditional manual experience methods. Combined with the usage rule model output, the final usage and price combination of the material list is generated, and it is judged whether the cost control requirement is met to ensure that the finally generated list data can effectively control the cost and avoid budget overruns or resource waste. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 The flowchart of the engineering cost material price list calculation method based on artificial intelligence of the present application; Figure 2 The flowchart of the engineering cost material price list calculation method based on artificial intelligence of the present application; Figure 1 The sub-step flowchart of step S103. DETAILED DESCRIPTION

[0009] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. Example 1

[0010] like Figure 1 As shown in the figure, this embodiment provides a method for calculating the material price list of engineering costs based on artificial intelligence, which specifically includes the following steps: Step S101: By querying the preset database, obtain resource conditions and project type information that match the project requirements from historical data to obtain the initial dataset; The process of obtaining the initial dataset specifically includes: By querying a pre-defined database, resource conditions and project type information that match the project requirements are obtained from historical data, thus generating an initial dataset. Based on the initial dataset, extended resource distribution features are obtained from the matching information using conditional filtering. Association conditions are determined for the type associations in the extended resource distribution features. If the association conditions exceed a preset threshold, the optimized resource distribution is determined.

[0011] Specifically, the pre-set database stores various data of historical engineering projects, including resource conditions such as the allocation of manpower, materials and equipment, as well as classifications of project types such as civil engineering, installation or maintenance.

[0012] First, the system receives project requirements input by the user, such as specifying the project type as bridge construction, along with resource conditions such as budget constraints and time requirements. Furthermore, the query process is implemented based on keyword matching and condition filtering. For example, project requirements can be broken down into keyword groups, such as "bridge construction" and "budget less than 5 million yuan", and then historical data records can be searched in the database. It should be noted that resource condition matching involves comparing the resource allocation of historical projects with current needs. For example, if the need emphasizes the efficient use of materials, then priority is given to retrieving records that have optimized material consumption in past projects, thus forming a subset with a high degree of matching. In one possible implementation, the matching algorithm uses similarity calculations to assess the consistency between historical data and the requirements.

[0013] Specifically, the resource conditions can be quantitatively represented, for example, converting human resources into a numerical vector, and engineering types are classified as coded labels. By calculating the Euclidean distance between vectors, it is ensured that the selected historical data has the minimum deviation from the demand, and the initial data set obtained contains multiple matching items, such as resource allocation details of the past five similar bridge projects; Preferably, to enhance accuracy, a weight mechanism can be introduced; For example, a higher weight is given to the engineering type because it determines the core framework of the project, while the resource conditions such as equipment type serve as auxiliary filtering. In this way, the initial data set not only includes directly matched data, but also extends to approximately matched variants, demonstrating the universality of the technical solution in the field of engineering management; Illustratively, in the context of bridge construction, the query may retrieve a steel reinforced concrete installation project from the historical data, whose resource conditions include 50 workers and specific mechanical equipment; If the demand matches, these information is integrated into the initial data set for subsequent planning. Further, the process ensures data integrity by checking the update time of historical data to avoid using outdated information; For example, a time threshold is set to only select records from the past three years, thereby improving the timeliness of the data set; In one embodiment, the generation of the initial data set can also combine multi-source data fusion; Specifically, if there is insufficient historical data in the database, matching information can be supplemented from associated external engineering archives, but still limited to the field of engineering projects, such as only for infrastructure construction; It can be understood that this query and matching method can achieve efficient preliminary screening of resources, providing a reliable foundation before the actual engineering project starts and avoiding blind allocation. In addition, in the maintenance of engineering types, for example, querying the maintenance project of a wind power station, matching resource conditions such as special tools and safety equipment form an initial data set containing historical maintenance records. Finally, through the above steps, the initial data set supports the preliminary verification of project requirements, ensuring the smooth progress of subsequent engineering processes.

[0014] Step S102: Grouping resource conditions and engineering types according to the initial data set, and determining a feature set after grouping; The process of determining the feature set after grouping specifically includes: By querying a pre-set database, resource conditions and engineering type information matching the project demand are obtained from historical data to obtain an initial data set; According to the initial data set, a K-means clustering analysis method is used to group the resource conditions and engineering types, and a feature set after grouping is determined, wherein the K-means clustering analysis method is input as the resource conditions and engineering type information in the initial data set, and the output is the feature set after grouping; For the feature set after grouping, condition screening is used to obtain an extended resource distribution feature from the matching information, and an association condition is determined for the type association in the extended resource distribution feature; If the association condition exceeds a preset threshold, it is determined that an optimized resource distribution is obtained, and an optimized resource allocation mode is obtained; Distribution parameters are extracted from the optimized resource allocation mode, and a final grouping feature set is determined.

[0015] For example, in an embodiment, according to the initial data set, a clustering analysis method is used to group the resource conditions and engineering types, and a feature set after grouping is determined; Specifically, this process first represents the data points in the initial data set as multi-dimensional vectors, where the resource conditions such as the number of workers and the type of materials are quantitatively coded, and the engineering type is converted into a classification label. Through this representation, the diversity data of historical engineering projects can be processed; It should be noted that the clustering analysis method here uses the K-means algorithm as the basis for implementation; The K-means algorithm is an unsupervised learning method, and its principle is to calculate the distance of data points to the cluster center by iteration, and to assign similar data points to the same group; The specific process includes initializing the cluster center, for example, randomly selecting several data points in the initial data set as the starting point, then calculating the Euclidean distance of each data point to these centers, and grouping according to the minimum distance principle; then, update the center of each group to the mean of the data points in the group, and repeat this process until the center position is stable. In this way, in engineering projects, resource conditions such as equipment allocation and engineering types such as installation projects can be effectively grouped.

[0016] For example, in the bridge construction scenario, the initial data set contains the resource conditions of multiple historical projects, such as the number of workers being 40 to 60 and the amount of steel being 100 to 200 tons, and the engineering type is marked as civil engineering; applying clustering analysis, these data are grouped into three clusters: one cluster represents low-resource-intensive bridge projects, the features include less manpower and basic materials; another cluster corresponds to high-resource projects, emphasizing heavy equipment; the third cluster focuses on maintenance-related mixed types. Through this grouping, the feature set is determined as the representative vector of each cluster, capturing the commonality of resources and types.

[0017] Further, in one possible implementation, to handle the diversity of project types, a hierarchical clustering method can be introduced as an alternative; the principle of hierarchical clustering is to start from individual data points and gradually merge the nearest clusters until a preset number of clusters is reached; its process involves constructing a distance matrix, for example, using Manhattan distance to measure the difference in resource conditions, and then forming a tree structure through bottom-up merging; In the maintenance project such as a wind power station project, this method groups historical data into safety equipment-intensive and regular tool-type clusters, thereby extracting feature sets such as average equipment usage rate; Preferably, the grouping process takes into account the data preprocessing step; Specifically, the resource conditions are first normalized to unify the numerical range to 0-1 to avoid the scale difference affecting the clustering accuracy; For example, human resources are converted from actual number to proportion value, and engineering types are converted into binary vectors through one-hot encoding. The data prepared in this way is input into the clustering algorithm to ensure the robustness of grouping; It can be understood that determining the feature set after grouping involves extracting key attributes from each cluster; In one embodiment, for civil engineering grouping, the feature set includes the resource vector of the cluster center, such as average material consumption and typical equipment list, and the distribution statistics of the engineering type label. This extraction is achieved by calculating the mean and variance within the cluster, supporting the reference for subsequent engineering planning; Exemplarily, in the data set of installation engineering types, the clustering analysis can generate two main groups: one focusing on time-critical projects with high labor input as a feature; and the other emphasizing material efficiency with optimized equipment configuration as a feature, thereby determining the feature set covering the quantitative description of these groups; In one implementation, to enhance universality, a density-based clustering method such as DBSCAN algorithm can be combined, which principle is to identify high-density areas as cluster cores and expand to adjacent points, ignoring noise data; The specific process includes setting a radius parameter and a minimum point threshold, for example, in resource condition data, defining a similarity threshold of 0.5 radius, and then clustering from the core point, in the bridge maintenance scenario, this method groups out abnormal resource usage patterns such as clusters of sudden equipment demand, thereby forming a unique feature set.

[0018] Step S103: If the dynamic change in the grouped feature set exceeds a preset threshold, train a usage rule model to obtain a trained usage rule model; As Figure 2 shown, the process of obtaining the trained usage rule model includes: Step S1031: Obtain dynamic change data related to dynamic changes of features from the grouped feature set, and use a preset threshold to judge the dynamic change data to obtain a change result exceeding the threshold; Step S1032: According to the change result, train a usage rule model by a random forest algorithm using the change result as input, wherein the usage rule model is obtained by constructing a decision tree set and processing input data by voting. Step S1033: For the trained usage rule model, obtain extended resource distribution features from resource condition matching, wherein the resource condition matching is based on engineering type grouping to determine type association conditions. Step S1034: If the type association condition exceeds a preset threshold, determine the resource optimization distribution, obtain the optimized resource allocation mode, extract distribution parameters from the optimized resource allocation mode, and determine the final grouped feature set.

[0019] For example, in an embodiment, when the grouped feature set changes dynamically, first evaluate the degree of change; Specifically, dynamic change refers to the fluctuation of resource conditions or engineering type related attributes in the feature set, such as real-time adjustment of manpower allocation or material consumption. It should be noted that the preset threshold can be set as a percentage value, such as 10%, by comparing the difference between the current feature vector and the historical mean value to determine if the change exceeds the threshold, and if the change exceeds the threshold, the subsequent training process is triggered. This evaluation ensures the sensitivity of the model to resource fluctuations in engineering projects, supporting timely updating of usage rules; Further, the random forest algorithm, as an integrated learning method, is used to train the usage rule model. Its principle is to construct multiple decision trees, each tree is trained based on a randomly sampled data subset and a feature subset, and then the results are integrated by voting or averaging to improve the generalization ability and robustness of the model. In the engineering project scenario, this algorithm is particularly suitable for processing high-dimensional feature data, such as integrating manpower, materials, and engineering type data in bridge construction. For example, in the grouped feature set of civil engineering, if the dynamic change exceeds the threshold, collect the updated data set, including the resource usage records of recent projects. The training process of the random forest algorithm starts with the data preparation stage, which converts the feature set into an input matrix, where rows represent project samples and columns represent quantitative features such as equipment utilization. Then, the algorithm randomly selects a sub-sample to generate a decision tree, for example, using the bootstrapping method to sample each tree to ensure diversity.

[0020] In one possible implementation, the training process involves hyperparameter tuning, such as setting the number of trees to 100 to balance computational efficiency and accuracy. Each decision tree starts from a root node and splits nodes based on information gain or Gini index, for example, in a maintenance engineering scenario, the split can be based on whether material consumption exceeds the average value. In this way, multiple trees form a forest together, and the output is a usage pattern model for predicting resource allocation patterns.

[0021] It can be understood that the usage pattern model is obtained by integrating the prediction results of all decision trees.

[0022] Specifically, in an installation-type engineering project, the model input is a feature set after grouping, such as a vector of high-labor-input clusters, and the output is a pattern description vector that captures the association between resources and engineering types. This model supports optimization of engineering planning, such as predicting equipment demand fluctuations in wind power station maintenance.

[0023] In one embodiment, to handle noisy data, the random forest algorithm introduces a random feature selection mechanism, which randomly selects a subset from the total features to build each tree, for example, in a bridge maintenance grouping, only manpower and steel consumption are selected as sub-features to avoid overfitting, and through iterative training, the final model is obtained until the forest is stable.

[0024] For example, in another civil engineering scenario, if the dynamic change of the feature set is caused by a sudden material shortage exceeding a threshold, the trained model analyzes historical patterns and provides adjustment suggestions, such as increasing the reserve of spare resources. This implementation demonstrates the practicality of the algorithm in resource management; Further, the trained usage pattern model can be stored as a serializable object to support subsequent calls; in a mixed-type grouping of maintenance engineering, the model output includes a probability distribution describing the occurrence patterns of different resource conditions, thereby providing a reference for project decision-making.

[0025] It should be noted that the entire process emphasizes data privacy by anonymizing input features to ensure compliance in engineering projects, and this method remains consistent across various implementation scenarios, such as the transition from low-resource-intensive to high-resource projects.

[0026] Step S104: For the trained usage pattern model, input the current market supply and demand data to determine the influence degree of market supply and demand on price prediction, and obtain the influence weight; The process of obtaining the influence weight specifically includes: Obtain current supply and demand data from the market environment, input the supply and demand data into the usage pattern model using a preset input mechanism, and obtain the preliminary prediction output; According to the preliminary prediction output, in combination with supply and demand dynamic monitoring, the influence degree of market supply and demand on price prediction is judged, and an influence factor set is determined; For the influence factor set, market trend analysis is fused through a weight calculation process to obtain a weight distribution result; Key indicators are extracted from the weight distribution result, and a prediction verification link is used to obtain an influence weight.

[0027] The input mechanism includes data standardization and format conversion; The supply and demand dynamic monitoring tracks changes by real-time collection of market indicators; The weight calculation process uses a weighted summation method to obtain a quantitative value based on the contribution degree of each element in the influence factor set; The verification link analyzes the deviation by comparing the weight distribution result with the preset benchmark to determine the overall influence degree.

[0028] For example, in one embodiment, for the trained usage pattern model, first prepare the current market supply and demand data as input; These data include the real-time supply and demand of resources required by the engineering project, for example, in civil engineering, collect market supply indicators and project demand prediction values of steel and cement, and the usage pattern model receives these data for preliminary processing based on the resource allocation pattern previously trained by the random forest algorithm.

[0029] Specifically, the model converts the supply and demand data into a feature vector, where the supply is represented as a positive indicator and the demand is represented as a negative indicator, ensuring that the input format is consistent with the training. This preparation process supports the model's response to market fluctuations, and in the context of bridge construction, local market data can be integrated to reflect material availability. Further, after inputting the current market supply and demand data, the model performs the process of judging the influence degree of market supply and demand on price prediction. The usage pattern model contains multiple decision tree sets, and each tree evaluates the potential influence of supply and demand differences on resource prices.

[0030] For example, in a maintenance engineering project, if the market supply is lower than the demand, the model calculates the difference value to quantify the influence by comparing historical regularity data.

[0031] It should be noted that the judgment mechanism relies on the integrated output of the model, and the influence score is determined by averaging the results of the decision trees, where a higher score indicates that the imbalance between supply and demand will drive a larger price increase. This mechanism ensures accuracy in engineering resource management and avoids the bias of a single indicator.

[0032] Preferably, in the judgment process, the supply and demand ratio is introduced as a key parameter; Exemplarily, in the installation type engineering, the supply-demand ratio is calculated as the demand quantity divided by the supply quantity, and if the ratio exceeds 1, it indicates that the shortage may lead to price increase. The model further analyzes the correlation between this ratio and historical price data, generates a quantitative description of the impact degree through built-in rule vector matching; It can be understood that this analysis process does not involve complex numerical iteration, but is based on fast matching of pre-trained tree structure, supporting real-time decision-making.

[0033] For example, in the maintenance grouping of wind power stations, the model inputs the supply-demand data of wind power equipment components, and judges that the impact of supply shortage on spare parts prices reaches a medium level.

[0034] Specifically, the impact weight is obtained by weighted summation, and the trained usage rule model converts the impact degree into a weight value, for example, the weight range is set to 0 to 1, where close to 1 indicates strong impact; Further, in the civil engineering scenario, the model considers multiple supply-demand dimensions, such as independent weights of manpower and materials, and then calculates the overall weight through an integrated formula, the specific process includes first extracting the impact score of each dimension, and then applying the model's internal voting mechanism to aggregate, for example, when the majority decision tree supports a high weight, the final output is 0.8, such weight reflects the contribution of market supply and demand to overall price prediction, which can be used to adjust budget allocation in bridge maintenance projects.

[0035] In one possible implementation, to enhance applicability, the model allows adjustment of the granularity of input data; For example, in the mixed type engineering grouping, using monthly supply-demand data input, it is judged that the impact weight on long-term price trend is low, while using weekly data may get a higher weight.

[0036] It should be noted that this flexibility is achieved through the model's random feature selection mechanism, ensuring robustness under different engineering scales; further, in another embodiment, for high resource-intensive engineering, such as large-scale civil engineering projects, global market supply-demand data is input to judge cross-border impact; The impact weight obtained by the model analysis can be used to predict the price fluctuations of imported materials, for example, when there is a global shortage of steel supply, the weight increases to 0.9, supporting the project team to purchase in advance; It can be understood that the whole process emphasizes seamless integration with engineering project resources.

[0037] Exemplarily, in the installation engineering, the obtained weight is applied to the correction link of the price prediction model, improving the prediction accuracy; In one embodiment, the impact weight is stored as a subsequent reference, supporting continuous monitoring of market changes.

[0038] Step S105: introducing a multi-factor price driving analysis based on an attention mechanism, combining the influence weight and time series decomposition to extract the potential periodic characteristics and external impact effects of price fluctuations, and generating a dynamic price adjustment factor; Obtaining multi-factor indicators from market data, and using an attention allocation mechanism for the multi-factor indicators, wherein the attention allocation mechanism obtains a fused driving vector by calculating a correlation score of each factor and weighted sum; For the fused driving vector, a weight calculation process is combined, wherein the weight calculation process is updated based on historical influence weights and incorporates multi-factor fusion attributes to determine sequence decomposition input; Through the sequence decomposition input, a time series decomposition method is used, wherein the time series decomposition method splits the price sequence into trend component, seasonal component and residual component and combines cycle identification attributes to obtain potential periodic characteristics; For the potential periodic characteristics, an impact evaluation link is incorporated, wherein the impact evaluation link determines external impact effects by comparing the influence of exogenous variables on residual components and fusing impact evaluation attributes; According to the external impact effects, the effect quantization results are fused, wherein the effect quantization results use a weighted integration method and incorporate adjustment generation attributes to generate a dynamic price adjustment factor.

[0039] For example, in one embodiment, the process of introducing a multi-factor price driving analysis based on an attention mechanism first prepares price data for engineering project resources. This analysis aims to handle multiple influencing factors such as market supply and demand, historical price trends and external events, and dynamically allocates attention through an attention mechanism.

[0040] Specifically, the attention mechanism is a neural network component that generates weights by calculating the similarity between query, key and value vectors, highlighting the most relevant factors driving prices; For example, in civil engineering, input data includes steel supply and demand ratio and historical price sequence, and the attention mechanism evaluates the relative importance of these factors to ensure that the analysis focuses on elements that dominate price fluctuations. The principle of this mechanism is to simulate human attention allocation, avoiding equal weight processing of all factors, thereby improving the accuracy of the analysis; It should be noted that in the preparation stage, data is organized in sequence form to support layer-by-layer calculation of the mechanism; in the bridge construction scenario, this analysis can integrate local market data to identify supply and demand imbalance as the primary driving factor. In this way, the attention mechanism provides a foundation for subsequent combination steps, ensuring that the interaction between multi-factors is effectively captured.

[0041] It can be understood that the whole process does not rely on complex iteration, but realizes fast response based on pre-defined attention layers, supporting real-time engineering resource management; Further, the analysis depth is expanded by combining the impact weight with the time series decomposition step. The impact weight is derived from the output of a previous regular model, representing the contribution of market supply and demand to price prediction; The time series decomposition then splits the price data into trend components, seasonal components, and residual components, where the trend reflects long-term changes, the season captures periodic patterns, and the residual represents random fluctuations. The specific process includes first applying additive or multiplicative models to decompose the price sequence, for example, in the maintenance of engineering projects, the historical data of cement prices are decomposed to extract annual trends and monthly seasonal fluctuations; Then, the impact weight is combined with these decomposition components to adjust the strength of each component through weighted fusion.

[0042] For example, if the impact weight is 0.7, the amplitude of the trend component is amplified accordingly, highlighting the impact of supply and demand on long-term prices. This combination ensures the comprehensiveness of the analysis, which can be used to evaluate the seasonal adjustment of equipment component prices in installation engineering.

[0043] Preferably, the decomposition process uses standard statistical methods, such as variants of the STL algorithm, to ensure the stability of the decomposition results. Through this integration, the analysis can reveal the internal structure of price fluctuations, providing reliable input for feature extraction.

[0044] In one possible implementation, the extraction of potential periodic features of price fluctuations focuses on identifying repetitive patterns, where potential periodic features refer to the hidden cyclical regularity in price sequences, such as the price rise cycle caused by seasonal supply and demand peaks.

[0045] Specifically, based on the seasonal components after time series decomposition, Fourier transform or autocorrelation analysis is further applied to quantify the length and strength of the cycle.

[0046] For example, in the maintenance group of wind power stations, the price of wind power equipment spare parts is analyzed to extract the price cycle feature corresponding to the winter wind demand peak every year. The principle of this extraction process is to convert the decomposition components into frequency domain representation, identify the dominant frequency, and thus quantify the influence amplitude of the cycle.

[0047] It should be noted that if the periodic feature shows a 12-month cycle, it is marked as a strong cycle, supporting the prediction of future fluctuations. In the context of civil engineering, this feature can be combined with the impact weight to evaluate the driving effect of the cycle on the overall price.

[0048] It can be understood that the extraction does not involve numerical iteration, but filters significant cycles through pre-set thresholds, ensuring applicability in different engineering scales.

[0049] Specifically, the process of extracting external shock effects targets the immediate impact of an incident on prices. External shock effects refer to non-periodic disturbances, such as sudden supply and demand mutations caused by natural disasters or policy changes, and the implementation includes detecting outliers from the residual component of time series decomposition, identifying shock points through threshold methods or anomaly detection algorithms.

[0050] For example, in a bridge maintenance project, if a sudden flood affects steel supply, the analysis extracts the shock amplitude from the residual, calculates its instantaneous boosting effect on prices, and the principle of this effect lies in the residual capturing the variation not explained by trend and seasonality, the shock is quantified as the product of the deviation value and the duration; Further, combined with attention mechanisms, shock effects are given higher weights if they are associated with high-impact-weight supply and demand factors. In installation engineering, this extraction can reveal the impact of global events on imported material prices, supporting the emergency response of project teams.

[0051] Preferably, the process allows real-time event data to be input, enhancing the dynamics of extraction. Further, the step of generating dynamic price adjustment factors integrates the aforementioned elements to form the final output, a dynamic price adjustment factor, which is a numerical indicator used to correct the base price forecast, reflecting the combined impact of periodic characteristics and shock effects; The specific process includes first converting the extracted periodic characteristics and shock effects into vector form, then weighted summation through attention mechanisms, and fusion with impact weights to generate factor values.

[0052] For example, in civil engineering, if the periodic characteristics indicate an upward trend and the shock effect is positive, the factor is calculated as the product of the periodic intensity and the shock amplitude, multiplied by the impact weight, and the result is set to a range of 0.5 to 1.5, used to amplify or reduce the predicted price. The principle of this generation lies in providing adjustable parameters to support real-time correction of the price model. In the context of bridge construction, the factor can be applied to budget planning to ensure the adaptability of resource allocation.

[0053] It should be noted that the generation does not require complex calculations, but is based on linear combination to achieve efficient output; In one embodiment, to enhance flexibility, the analysis process allows adjusting the number of layers of attention mechanisms; in maintenance engineering projects, single-layer attention is used to process simple supply and demand data, generating adjustment factors for short-term price prediction, while multi-layer attention is suitable for complex multi-factor scenarios, such as integrating global market shocks. This adjustment is achieved through model parameter configuration, supporting customization for different engineering needs.

[0054] Exemplarily, in the resource management of a wind power station, the entire analysis starts from the input of spare parts supply and demand data, combines the decomposition of time series of impact weights, extracts the winter cycle characteristics and sudden policy impact effects, and finally generates a factor value of 1.2 for adjusting the maintenance budget. This application demonstrates the versatility of the technology. Further, in installation-type engineering, attention is paid to price-driven analysis of human cost, and the attention mechanism highlights the weight of demand peaks to generate dynamic factors to support contract negotiations. In another embodiment, for large-scale civil engineering projects, external data sources are introduced to enhance the extraction of impact effects, and factors are generated to predict material price fluctuations, supporting early procurement decisions.

[0055] It can be understood that these embodiments emphasize seamless integration with engineering resource management, ensuring the robustness of analysis in various scenarios.

[0056] Step S106: Use the adaptive prediction engine to model price prediction as a sequence decision problem. The model learns the optimal prediction strategy based on market state and dynamic adjustment signals, and outputs a dynamic price interval with risk perception ability. The process of outputting a dynamic price interval with risk perception ability specifically includes: Obtain time series data from market data, and use a time series convolution network to extract features from the time series data to obtain a fusion feature vector; Integrate the fusion feature vector into a reinforcement learning framework, wherein the reinforcement learning framework determines adaptive prediction parameters based on a state-action-reward mechanism and optimizes the decision path through iterative updates; Construct a sequence decision model based on the adaptive prediction parameters to determine the trend of market state changes; Integrate dynamic adjustment signals through the trend of market state changes to obtain a risk perception index; Generate a dynamic price interval for the risk perception index to determine an adaptive price fluctuation range; The time series convolution network captures sequence dependency and integrates multi-time point correlation through multi-layer convolution operations. The sequence decision model converts price prediction into a continuous decision process and integrates state transition probabilities. The dynamic adjustment signal is derived from external factor quantization and processed by weighted summation. The dynamic price interval integrates the prediction strategy output and is integrated into the fluctuation boundary calculation.

[0057] Exemplarily, in an embodiment, an adaptive prediction engine employing a fusion of temporal convolutional network and reinforcement learning is adopted to firstly initialize the price data of the engineering project resources, and the adaptive prediction engine is an integrated model combining the sequence processing capability of the temporal convolutional network and the decision optimization mechanism of the reinforcement learning, and is used for processing the dynamic characteristics of the price fluctuation.

[0058] Specifically, the temporal convolutional network is a neural network structure which captures long-term dependencies in time series through multi-layer convolutional operations, and the principle thereof is to use causal convolution and dilated convolution to expand the receptive field, avoid information leakage, and ensure sequential analysis of historical price data. In civil engineering, for example, the price sequence of steel is input into the engine, and the network layer extracts the time sequence pattern of supply and demand changes to provide basic features for subsequent decision-making. The construction of such an engine does not rely on a single model, but realizes adaptive adjustment through hierarchical fusion to support real-time prediction of project resources. Further, modeling the price prediction as a sequence decision problem expands the function of the engine. The sequence decision problem refers to treating the prediction task as a series of continuous decisions, each of which selects an action based on the current state to maximize long-term returns. The specific implementation includes defining the state space as the market supply-demand ratio and the historical price vector, the action space as the prediction adjustment parameter, and the reward function as the return value calculated according to the prediction accuracy and risk bias.

[0059] For example, in the context of bridge construction, the engine models the steel price prediction as a decision sequence. If the market state shows that the supply and demand are tight, the model selects to enlarge the prediction interval to perceive the risk. The principle of this modeling is to simulate the learning path of the agent in an uncertain environment, organize the data flow through the Markov decision process framework, and ensure that the prediction is not just a static output, but a dynamic response to market changes.

[0060] It should be noted that this modeling allows the engine to make iterative decisions at different time steps, supporting the optimization of long-term project budgets. In the maintenance of engineering projects, this method can be applied to the sequence decision of cement prices, and seasonal supply and demand are identified as key state variables.

[0061] Preferably, the step of the model learning the optimal prediction strategy based on the market state and dynamic adjustment signal focuses on the training of the reinforcement learning component. Self-learning refers to the optimization of the strategy by the model through a trial-and-error mechanism without pre-set rules. The principle thereof is to update the parameters of the agent using a policy gradient or Q-learning algorithm, and the dynamic adjustment signal is used as an external input, such as a sudden event indicator, to correct the state transition.

[0062] Specifically, in the installation type engineering, the model randomly selects a prediction strategy in the initial state, calculates the reward by simulating market fluctuations, and strengthens the strategy if the prediction bias is small, gradually converging to the optimal path.

[0063] Exemplarily, if the dynamic adjustment signal indicates a policy change, the model autonomously increases the risk weight, ensuring policy adaptability, and the business process of this learning includes data batch input, reward feedback cycle and policy update stage, which can be used for spare parts price prediction in wind power station maintenance, and the optimal adjustment strategy for the winter demand peak is learned.

[0064] It can be understood that the whole process realizes efficient training through batch simulation, supporting decision support of engineering teams.

[0065] In a possible implementation, the mechanism of fusing the time sequence convolution network and the reinforcement learning ensures the adaptability of the engine. The specific process first processes the input sequence by the time sequence convolution network to generate a feature representation, and then inputs the feature representation into the reinforcement learning agent as a state input. The agent outputs an action to adjust the prediction. The principle of fusion is that the output vector of the network is used as the observation value of the reinforcement learning, which promotes end-to-end optimization.

[0066] For example, in a large project of civil engineering, the network captures the trend component of the historical price, and the agent learns a decision-making strategy based on the trend. If the market state fluctuates greatly, the agent selects a conservative action to generate a wide price interval. This fusion does not involve independent training, but jointly optimizes the parameters to improve the overall robustness. It should be noted that in the bridge maintenance scenario, this mechanism can integrate local supply and demand data, and the agent autonomously learns a response strategy to external shocks to ensure the dynamics of the prediction strategy. Further, the step of outputting a dynamic price interval with risk perception capability integrates the aforementioned learning results to form a final prediction. The risk perception capability refers to the interval not only giving a point prediction, but also integrating uncertainty quantification, such as indicating potential fluctuations through a confidence interval. The specific implementation includes the agent generating upper and lower bound values at the end of decision-making, and evaluating the risk level based on cumulative rewards.

[0067] For example, in a maintenance engineering project, the output of the cement price is an interval, such as a 10% fluctuation around the base value. If the learning strategy detects a high risk signal, the interval is expanded to cover possible shocks. The principle of this output is to combine the reward function of reinforcement learning with Bayesian uncertainty to provide a more reliable basis for decision-making. In installation engineering, this interval can be used for equipment procurement planning to help teams assess budget risks.

[0068] Preferably, the interval generation allows parameterized adjustment, such as setting a risk threshold, to support applications of different engineering scales.

[0069] In an embodiment, to enhance the flexibility of the engine, the number of layers of the timing convolution network and the exploration rate of reinforcement learning are allowed to be adjusted, in the resource management of a wind power plant, a shallow network is used to process simple sequences, and a conservative strategy is learned by combining a low exploration rate, and a spare parts price interval is output for a short-term maintenance budget, and such adjustment is achieved by configuring parameters, supporting customized prediction.

[0070] Illustratively, the entire engine starts from the input market state, fuses the processing of sequence data, autonomously learns the strategy, and finally outputs a risk perception interval value of 0.8 to 1.2 times the base price, which is used to adjust the purchase decision. This application shows the universality of the engine in engineering resource management, which can realize sensitive response to price fluctuations.

[0071] Specifically, in another embodiment, for large-scale civil engineering projects, real-time market signals are introduced to enhance dynamic adjustment, and the engine autonomously optimizes the strategy according to these signals, and outputs an interval reflecting the impact of global events, supporting advance material reserves. In bridge construction, the model learns the optimal prediction path of supply and demand imbalance to ensure that the interval covers potential rising risks.

[0072] It can be understood that these embodiments emphasize integration with engineering management, ensuring the applicability of the engine in various scenarios.

[0073] Further, the artificial intelligence-based engineering cost material price list calculation method described in the embodiment further comprises: For the predicted price value, the final consumption and price combination of the material list are generated by combining the use rule model output, and it is judged whether the combination meets the cost control requirements to obtain the optimized list data.

[0074] The process of obtaining the optimized list data specifically includes: From the demand fluctuation analysis, initial material list data is obtained, and the initial material list data is matched by comparing the current inventory level with the historical consumption record to obtain a preliminary consumption combination; For the preliminary consumption combination, the supplier selection basis is integrated, and the corresponding price value is calculated by integrating market quotation data and contract terms to determine the price combination result; For the price combination result, if it exceeds the cost evaluation standard, the proportion of high-priced materials is reduced by combining the budget allocation strategy to obtain a revised combination; For the revised combination, the final material list data is generated by continuously comparing different iteration versions of the integrated risk identification factor.

[0075] Illustratively, in an embodiment, for the predicted price value, the concept and role of the use rule model need to be understood first in combination with the use rule model output; The usage pattern model is an analytical framework based on historical data and statistical patterns to capture the consumption behavior of materials in engineering projects. The principle is to analyze the periodicity of material usage through time series or statistical distribution, such as identifying seasonal demand peaks or project phase consumption rates.

[0076] Specifically, in civil engineering, this model inputs historical procurement records and project progress data, and outputs the expected pattern vector of material usage, such as the average consumption coefficient of cement in the foundation pouring stage. In this way, it ensures that the predicted price value is combined with the actual usage pattern to form reliable basic data. Further, the process of generating the final usage and price combination of the bill of materials integrates the aforementioned output. The specific implementation includes inputting the predicted price value as the unit price, using the output of the usage pattern model as the usage estimate, and then generating the combined value through multiplication.

[0077] For example, in the context of bridge construction, the predicted price value of steel is 5000 yuan per ton, and the usage pattern model outputs the steel usage of the main structure of the bridge as 200 tons, then the combination is the total price of 1 million yuan. This generation is not limited to a single material, but is extended to multiple items in the list, processed in batches through matrix operations, supporting the preliminary formation of the overall budget of the project.

[0078] It should be noted that this combination takes into account multiple variables such as material type and project size, ensuring that the output list data reflects actual engineering needs.

[0079] Preferably, the step of judging whether the combination meets the cost control requirements focuses on setting thresholds and comparison logic, and the cost control requirements refer to pre-set budget limits or efficiency standards. The principle is to define a cost function, such as the total price not exceeding 90% of the project budget, and then compare the generated combination value with it. If the combination exceeds the threshold, it is marked as not meeting the requirements.

[0080] Specifically, in maintenance engineering projects, if the combined total price of the cement list exceeds the control limit, record the deviation value for subsequent optimization. This judgment process is implemented through conditional branching, supporting real-time feedback to help the engineering team identify potential overbudget risks.

[0081] In one possible implementation, the operation of obtaining the optimized list data is based on the iterative adjustment of the judgment result. The specific process includes applying optimization algorithms to modify the usage or replace material options if the combination does not meet the requirements, such as reducing the usage of non-essential materials or selecting low-cost substitutes. The principle is to use greedy search or linear programming methods to minimize the cost function while maintaining project quality standards.

[0082] For example, in installation engineering, if the quantity of a component exceeds the initial 100, the quantity is adjusted to 80 after optimization, and a new combination is generated by combining the updated price until the control requirement is met. This optimization ensures the practicability of the list data. In wind power station maintenance, the method can be applied to spare parts list, and the final data reflecting cost savings is output after adjustment.

[0083] Illustratively, the whole process starts from the input of predicted prices, integrates the use of regular models, generates and judges combinations, and finally optimizes the list to support the continuity of engineering management. In the scenario of bridge maintenance, this method integrates local material data, and the optimized list reduces the total cost by 10%, providing a more efficient basis for resource allocation.

[0084] It can be understood that the logical order of these steps ensures a complete chain from data input to optimization, which is suitable for various engineering implementations.

[0085] The above embodiment is only one of the preferred embodiments of the present application and should not be used to limit the protection scope of the present application. Any modification or embellishment made within the main design idea and spirit of the present application without substantial meaning, which still solves the technical problems consistent with the present application, should be included in the protection scope of the present application.

Claims

1. An artificial intelligence-based engineering cost material price list estimation method, characterized by, The method comprises the following steps: Step S101: obtaining resource conditions and engineering type information matched with project requirements from historical data by querying a preset database to obtain an initial data set; Step S102: grouping resource conditions and engineering types according to the initial data set to determine a feature set after grouping; Step S103: if dynamic changes in the feature set after grouping exceed a preset threshold, training a usage rule model to obtain a trained usage rule model; Step S104: inputting current market supply and demand data into the trained usage rule model to determine the influence degree of market supply and demand on price prediction to obtain an influence weight; Step S105: introducing a multi-factor price driving analysis based on an attention mechanism, combining the influence weight and time series decomposition to extract potential periodic characteristics and external impact effects of price fluctuations to generate a dynamic price adjustment factor; Step S106: using an adaptive prediction engine to model price prediction as a sequence decision problem, and the model autonomously learns an optimal prediction strategy according to market status and dynamic adjustment signals to output a dynamic price interval with risk perception ability.

2. The method according to claim 1, wherein, The process of obtaining the trained usage rule model comprises: Step S1031: obtaining dynamic change data related to feature dynamic changes from the feature set after grouping, and determining a change result exceeding a threshold by using a preset threshold for the dynamic change data; Step S1032: training a usage rule model by using a random forest algorithm with the change result as input according to the change result to obtain a trained usage rule model; Step S1033: obtaining an extended resource distribution feature from resource condition matching for the trained usage rule model to determine a type association condition; Step S1034: if the type association condition exceeds a preset threshold, determining an optimized resource distribution to obtain an optimized resource allocation mode, extracting distribution parameters from the optimized resource allocation mode, and determining a final grouping feature set.

3. The method according to claim 1, wherein, The process of obtaining the influence weight specifically comprises: obtaining current supply and demand data from a market environment, inputting the supply and demand data into the usage rule model using a preset input mechanism to obtain a preliminary prediction output; determining an influence factor set by judging the influence degree of market supply and demand on price prediction according to the preliminary prediction output and combining supply and demand dynamic monitoring; obtaining a weight distribution result by fusing market trend analysis through a weight calculation process for the influence factor set; extracting key indicators from the weight distribution result to obtain an influence weight by using a prediction verification link.

4. The method according to claim 1, wherein, The process of generating the dynamic price adjustment factor specifically comprises: obtaining multi-factor indicators from market data, and obtaining a fused driving vector by using an attention distribution mechanism for the multi-factor indicators; determining a sequence decomposition input by combining a weight calculation process for the fused driving vector; obtaining potential periodic characteristics by using a time series decomposition method through the sequence decomposition input; judging external impact effects by integrating a shock evaluation link for the potential periodic characteristics; According to the external impact effect, the fusion effect quantitative result is generated to generate a dynamic price adjustment factor.

5. The method according to claim 1, wherein, The process of outputting the dynamic price interval with risk perception capability specifically includes: Obtain time series data from market data, and use a time series convolution network to extract features from the time series data to obtain a fusion feature vector; Integrate the fusion feature vector into a reinforcement learning framework to determine adaptive prediction parameters; Construct a sequence decision model according to the adaptive prediction parameters to determine the market state change trend; Integrate the dynamic adjustment signal through the market state change trend to obtain a risk perception index; Generate a dynamic price interval for the risk perception index to determine an adaptive price fluctuation range.

6. The artificial intelligence-based engineering cost material list measurement method according to claim 5, characterized in that, The time series convolution network captures sequence dependence and fuses multi-time point correlations through multi-layer convolution operations; The sequence decision model converts price prediction into a continuous decision process and integrates state transition probabilities; The dynamic adjustment signal is derived from external factor quantization and processed by weighted summation; The dynamic price interval integrates the prediction strategy output and is integrated into the fluctuation boundary calculation.

7. The artificial intelligence-based engineering cost material list measurement method according to claim 1, characterized in that, The process of obtaining the initial data set specifically includes: Obtain resource conditions and engineering type information matching the project demand from historical data by querying a preset database to obtain an initial data set; According to the initial data set, use conditional screening to obtain extended resource distribution characteristics from the matching information, determine the association condition for the type association in the extended resource distribution characteristics, and if the association condition exceeds a preset threshold, determine the optimized resource distribution.

8. The artificial intelligence-based engineering cost material list measurement method according to claim 1, characterized in that, The process of determining the grouped feature set specifically includes: Obtain resource conditions and engineering type information matching the project demand from historical data by querying a preset database to obtain an initial data set; According to the initial data set, use a K-means clustering analysis method to group resource conditions and engineering types to determine a grouped feature set, wherein the K-means clustering analysis method inputs resource conditions and engineering type information in the initial data set and outputs the grouped feature set; For the grouped feature set, use conditional screening to obtain extended resource distribution characteristics from the matching information, and determine the association condition for the type association in the extended resource distribution characteristics; If the association condition exceeds a preset threshold, determine the optimized resource distribution to obtain an optimized resource allocation mode; Extract distribution parameters from the optimized resource allocation mode to determine the final grouped feature set.

9. The artificial intelligence-based engineering cost material list measurement method according to claim 1, wherein, Further includes: For the predicted price value, generate a final usage and price combination of the material list by combining the usage rule model output, determine whether the combination meets the cost control requirements, and obtain optimized list data.

10. The artificial intelligence-based engineering cost material list measurement method according to claim 9, characterized in that, The process of obtaining the optimized list data specifically includes: Obtain initial material list data from demand fluctuation analysis, and match the current inventory level with historical usage records to obtain a preliminary usage combination; Integrate the supplier selection basis into the preliminary usage combination, calculate the corresponding price value by integrating market quotation data and contract terms, and determine the price combination result; In combination with the budget allocation strategy, the price combination result is judged to reduce the proportion of high-priced materials if the cost evaluation standard is exceeded, to obtain a modified combination; Through continuous comparison of different iteration versions of the risk identification factor, the final material list data is generated.

Images