Engineering cycle material supply and demand prediction method and system based on deep learning

Abstract

The application relates to the technical field of resource management, and discloses an engineering cycle material supply and demand prediction method and system based on deep learning, which comprises the following steps: performing multi-dimensional feature analysis on historical material consumption data to obtain a historical consumption feature spectrum; performing mode recognition on multi-source market information flow to obtain periodic price trend features and accidental price event features; dynamically coupling and correcting a material demand prediction value to obtain a material demand comprehensive prediction value; performing performance attenuation deduction on resource relationships of cycle material data in a target project to obtain an adjustable hypergraph resource network; performing supply and demand bidirectional optimization on the material demand comprehensive prediction value and the adjustable hypergraph resource network to obtain a gap filling scheme; and performing disturbance simulation on decision resilience in the adjustable hypergraph resource network to obtain a resilience decision report; and the application can improve the efficiency of engineering cycle material supply and demand prediction.

Description

Technical Field

[0001] This invention relates to the field of resource management technology, and in particular to a method and system for predicting the supply and demand of engineering cycle materials based on deep learning. Background Technology

[0002] Existing technologies for forecasting the supply and demand of engineering materials only process historical material consumption data at a superficial statistical level. They fail to combine engineering construction blueprints and milestone nodes to conduct multi-dimensional feature analysis, making it impossible to accurately anchor consumption data at different stages and classify modes. It is difficult to effectively distinguish between baseline consumption modes and offset consumption modes, and even more difficult to construct a complete historical consumption feature spectrum that includes steady-state spectrum and transient disturbance set. As a result, subsequent demand forecasting lacks accurate basic data support.

[0003] Existing technologies, when processing multi-source market information flows, fail to effectively separate trend components from cyclical fluctuation components, making it impossible to accurately extract cyclical price trend characteristics. Furthermore, they lack systematic methods for attribution analysis of abnormal market price events, making it difficult to identify the characteristics of sporadic price events. In the processing of cyclical material resource relationships, they do not conduct time-series extrapolation of efficiency decay, cannot construct a dynamic and adjustable hypergraph resource network, and struggle to form scientific gap-filling solutions during the two-way optimization process of supply and demand. Moreover, they cannot verify the decision-making resilience of the solutions through disturbances, ultimately leading to a significant deviation between supply and demand forecasts and actual engineering needs. Summary of the Invention

[0004] This invention provides a method and system for predicting the supply and demand of engineering cyclic materials based on deep learning, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a deep learning-based method for predicting the supply and demand of engineering recyclable materials, comprising: S1. Perform multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project; S2. Perform pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend characteristics and occasional price event characteristics of the target project; S3. Based on the historical consumption characteristic spectrum, the price trend characteristics, and the price event characteristics, the predicted material demand of the target project is dynamically coupled and corrected to obtain the comprehensive predicted material demand of the target project. S4. Perform efficiency decay deduction on the resource relationships of the cyclic material data in the target project to obtain the adjustable hypergraph resource network of the target project; S5. Perform two-way optimization of supply and demand on the comprehensive forecast value of material demand and the adjustable hypermap resource network to obtain the gap filling solution for the target project; S6. Based on the gap-filling scheme, perform perturbation simulation on the decision resilience in the configurable hypergraph resource network to obtain the resilience decision report of the target project.

[0006] In a preferred embodiment, the step of performing multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project includes: By stripping the historical material consumption data of the target project from time sequence, the basic consumption data stream of the target project is obtained. Based on the construction blueprint and milestone nodes of the target project, the basic consumption data stream is stage-anchored to obtain the consumption data fragments of the target project. Multimodal data fusion is performed on the consumed data segments to obtain the baseline consumption mode and offset consumption mode of the target project; Trend fitting is performed on the baseline consumption mode to construct the steady-state spectrum of the target project; By attributing the offset consumption mode to its source, the transient disturbance set of the target project is obtained; Based on the transient disturbance set, the steady-state spectrum is annotated and corrected to obtain the historical consumption characteristic spectrum of the target project.

[0007] In a preferred embodiment, the step of performing pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend characteristics and occasional price event characteristics of the target project includes: The multi-source market information flow of the target project is subjected to component separation to obtain the trend component and periodic fluctuation component of the target project; Based on the trend component and the periodic fluctuation component, the multi-source market information flow is reconstructed to obtain the periodic price trend characteristics of the target project. By comparing the deviations between the multi-source market information flow and the cyclical price trend characteristics, candidate price anomaly events for the target project are obtained; External correlation attribution is performed on the candidate price anomaly events to obtain the characteristics of the sporadic price events of the target project.

[0008] In a preferred embodiment, the step of dynamically coupling and correcting the predicted material demand of the target project based on the historical consumption characteristic spectrum, the price trend characteristics, and the price event characteristics to obtain the comprehensive predicted material demand of the target project includes: The transient components in the historical consumption feature spectrum and the price event features are interpreted with time lag to obtain the transient causal relationship of the target project; Based on the transient causal relationship, the predicted value of material demand for the target project is reshaped to obtain the event-corrected prediction sequence of the target project. By performing long-term evolution analysis on the steady-state components in the historical consumption characteristic spectrum and the price trend characteristics, the synergistic elasticity parameters of the target project are obtained. Based on the aforementioned collaborative elasticity parameter, the overall trend baseline of the predicted material demand is progressively adjusted to obtain the trend-corrected prediction baseline for the target project. The event-corrected prediction sequence and the trend-corrected prediction baseline are fused in a time-varying manner to obtain the comprehensive prediction value of the material demand for the target project.

[0009] In a preferred embodiment, the step of time-varying fusion of the event-corrected prediction sequence and the trend-corrected prediction baseline to obtain the comprehensive predicted value of the material demand for the target project includes: The activity of the local fluctuation pattern and occurrence frequency of the event correction prediction sequence is evaluated to obtain the fluctuation activity data of the target project; The stability of the long-term smoothness and directional consistency of the trend correction prediction baseline is assessed to obtain the trend stability data of the target project. Based on the fluctuating active data and the trend stable data, the event impact and trend impact of the target project are weighted to obtain the dynamic fusion weight of the target project; Based on the dynamic fusion weights, the event-corrected prediction sequence and the trend-corrected prediction baseline are integrated in the time domain to obtain the comprehensive prediction value of the material demand of the target project.

[0010] In a preferred embodiment, the dynamic fusion weight is calculated using the following formula: ; In the formula, The dynamic fusion weights, The volatility activity parameter of the volatility active data. The trend stability parameter of the trend-stable data. The preset periodic adjustment range, The preset market cycle length, The preset periodic phase adjustment constant, It is a time variable.

[0011] In a preferred embodiment, the step of performing performance attenuation deduction on the resource relationships of cyclic material data in the target project to obtain the adjustable hypergraph resource network of the target project includes: Attribute features are extracted from the cyclic material data of the target project to obtain multidimensional material feature data of the target project. Based on the multidimensional feature data of the materials, the cyclic material data is correlated and identified to obtain the resource topology relationship of the target project; The performance degradation trajectory of the multidimensional feature data of the materials is extrapolated over time to obtain the material efficiency decay sequence of the target project; Using the multidimensional feature data of the materials as nodes and the resource topology relationships as hyperedges, an initial hypergraph framework for the target project is constructed. Based on the material effectiveness decay sequence, the callable state attributes in the initial hypergraph framework are dynamically updated, and the connection strength of the hyperedges in the initial hypergraph structure is attenuated and adjusted to obtain the configurable hypergraph resource network of the target project.

[0012] In a preferred embodiment, the step of performing two-way optimization of the comprehensive forecast of material demand and the adjustable hypergraph resource network to obtain a gap-filling solution for the target project includes: The comprehensive forecast of material demand is deconstructed into demand elements to obtain the demand constraints of the target project. Based on the aforementioned demand constraints, the scalable hypergraph resource network is traversed and filtered to obtain the basic allocation plan for the target project. Based on the hyperedge relationships in the scalable hypergraph resource network, the basic allocation plan is combined and alternatively deduced to obtain the extended allocation plan for the target project; The demand constraints are compared item by item with the extended allocation plan to obtain the material gap list of the target project, and a gap filling plan for the target project is constructed based on the material gap list.

[0013] In a preferred embodiment, the perturbation simulation of decision resilience in the scalable hypergraph resource network based on the gap-filling scheme to obtain a resilience decision report for the target project includes: Based on preset typical risk scenarios, perturbations are applied to key nodes or hyperedges in the configurable hypergraph resource network to obtain the perturbed network state of the target project. Based on the state of the disturbed network, the gap filling scheme is re-executed for verification, and the performance degradation path of the gap filling scheme is observed to obtain the performance degradation trajectory of the target project. Based on the performance degradation trajectory of the above scheme, the source location of the configurable hypergraph resource network is traced to obtain the key vulnerable nodes and their relationships in the target project. The gap-filling scheme, the performance degradation trajectory of the scheme, and the key vulnerable nodes and their relationships are annotated with strategies to obtain the resilience decision report of the target project.

[0014] To address the above problems, this invention also provides a deep learning-based engineering recurring material supply and demand forecasting system, the system comprising: The consumption feature analysis module is used to perform multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project. The price feature recognition module is used to perform pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend features and occasional price event features of the target project. The demand coupling correction module is used to dynamically couple and correct the predicted value of the material demand of the target project based on the historical consumption characteristic spectrum, the price trend characteristics and the price event characteristics, so as to obtain the comprehensive predicted value of the material demand of the target project. The hypergraph network inference module is used to perform efficiency decay inference on the resource relationships of cyclic material data in the target project, and obtain the adjustable hypergraph resource network of the target project. The supply and demand two-way optimization module is used to perform two-way optimization of the comprehensive forecast value of the material demand and the adjustable hypergraph resource network to obtain the gap filling solution for the target project. The resilience decision simulation module is used to perform perturbation simulation on the decision resilience in the scalable hypergraph resource network based on the gap filling scheme, and obtain the resilience decision report of the target project.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention analyzes historical material consumption data from multiple dimensions and combines the cyclical trend characteristics of market prices with the characteristics of occasional events to carry out dynamic coupling correction, thereby generating a comprehensive forecast value of material demand that fits the actual needs of the project and improving the accuracy and adaptability of the demand forecast results.

[0016] 2. This invention constructs a deployable hypergraph resource network by performing efficiency decay simulation on cyclical materials, forms a gap-filling solution by combining supply and demand optimization, and generates a resilience decision report through decision resilience disturbance simulation. This can enhance the stability of supply and demand matching of engineering materials and improve the risk resistance of material allocation schemes. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a deep learning-based method for predicting the supply and demand of engineering recyclable materials, as provided in an embodiment of the present invention. Figure 2A functional block diagram of a deep learning-based engineering cyclic material supply and demand forecasting system provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a deep learning-based method for predicting the supply and demand of engineering cyclic materials. The execution entity of this deep learning-based method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the deep learning-based method for predicting the supply and demand of engineering cyclic materials can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating a deep learning-based method for predicting the supply and demand of engineering cyclic materials according to an embodiment of the present invention. In this embodiment, the deep learning-based method for predicting the supply and demand of engineering cyclic materials includes: S1. Perform multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project; In this embodiment of the invention, the step of performing multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project includes: By stripping the historical material consumption data of the target project from time sequence, the basic consumption data stream of the target project is obtained. Based on the construction blueprint and milestone nodes of the target project, the basic consumption data stream is stage-anchored to obtain the consumption data fragments of the target project. Multimodal data fusion is performed on the consumed data segments to obtain the baseline consumption mode and offset consumption mode of the target project; Trend fitting is performed on the baseline consumption mode to construct the steady-state spectrum of the target project; By attributing the offset consumption mode to its source, the transient disturbance set of the target project is obtained; Based on the transient disturbance set, the steady-state spectrum is annotated and corrected to obtain the historical consumption characteristic spectrum of the target project.

[0021] The historical material consumption data of the target project is divided into continuous and non-overlapping time intervals such as daily, weekly, and monthly, according to the time granularity of the construction log records. The material consumption records in each time interval are separated and distinguished. The content of each record is checked one by one. Consumption data that is repeatedly entered in the same batch at the same time and abnormal values ​​that are unrelated to actual construction activities are removed. The data is then organized into a basic consumption data stream of the target project that covers the entire construction cycle, with clear data entries and no redundancy.

[0022] Based on the clearly defined construction stages, including foundation construction, main structure construction, and decoration and finishing construction, as outlined in the target project's construction blueprint, and combined with the specific time nodes corresponding to each milestone node set during the construction process, such as foundation completion, main structure capping, and electromechanical installation completion, each piece of consumption data in the foundation consumption data stream is mapped to its corresponding construction stage and milestone node according to the time of consumption, thus forming consumption data segments of the target project that correspond one-to-one with each construction stage and have clear data boundaries.

[0023] By integrating information from different dimensions such as material type, consumption quantity, consumption unit, and consumption time contained in the consumption data fragments, for each material type, all consumption data under the same construction stage are summarized, and the maximum, minimum, and average values ​​of the material consumption data within the construction stage are calculated. The numerical range of 10% above and below the average value is taken as the normal consumption range of this type of material under the construction stage. This normal consumption range is defined as the baseline consumption mode of the target project, and the consumption data exceeding this normal consumption range is classified as the offset consumption mode of the target project.

[0024] This study analyzes the specific numerical changes of baseline consumption modes at different construction stages of the target project. Following the construction progress sequence of foundation construction, main structure construction, and decoration and finishing construction, the numerical characteristics of baseline consumption modes of various materials in each construction stage are analyzed sequentially. The numerical change patterns of baseline consumption modes as construction stages progress are summarized. These stage-specific numerical characteristics and cross-stage change patterns are systematically sorted out and integrated to construct a steady-state spectrum of the target project that can comprehensively reflect the conventional material consumption status of the target project.

[0025] For each set of offset consumption modes, the material consumption data is analyzed by reviewing construction logs, material procurement records, on-site supervision reports, and other materials to trace the detailed construction process during the period when the data was generated. This includes the specific content of construction process adjustments, changes in the batches and quantities of material supplies, and unexpected situations such as equipment failures or personnel adjustments at the construction site. The specific reasons that caused the material consumption data to deviate are then linked and bound to the corresponding offset data, and integrated to form a transient disturbance set for the target project that includes details of the offset data and cause analysis.

[0026] Each record in the transient disturbance set is mapped to the corresponding construction stage in the steady-state spectrum according to the consumption occurrence time. Next to the corresponding stage entry in the steady-state spectrum, the specific value of the material consumption deviation data that occurred in that stage, the occurrence time, and the specific reason for the deviation are marked. Based on the marked content, the stage consumption characteristics of the steady-state spectrum are supplemented and corrected. Finally, the corrected steady-state spectrum is fully integrated with the transient disturbance set to obtain the historical consumption characteristic spectrum of the target project that takes into account both normal consumption patterns and abnormal consumption situations.

[0027] The beneficial effects are that by performing a series of operations on the historical material consumption data of the target project, such as time series stripping, stage anchoring, multimodal data fusion, trend fitting, attribution and source tracing, and steady-state spectrum annotation correction, it is possible to achieve refined processing of historical consumption data, accurately divide the basic consumption data stream and the consumption data segments of the corresponding construction stage, clearly distinguish the baseline consumption mode and the offset consumption mode, and completely construct a historical consumption feature spectrum that includes steady-state consumption patterns and transient disturbance causes. This effectively provides accurate and comprehensive basic data support for subsequent prediction of the supply and demand of cyclical materials in the project, and enhances the data reliability and adaptability of subsequent prediction work.

[0028] S2. Perform pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend characteristics and occasional price event characteristics of the target project; In this embodiment of the invention, the step of performing pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend characteristics and occasional price event characteristics of the target project includes: The multi-source market information flow of the target project is subjected to component separation to obtain the trend component and periodic fluctuation component of the target project; Based on the trend component and the periodic fluctuation component, the multi-source market information flow is reconstructed to obtain the periodic price trend characteristics of the target project. By comparing the deviations between the multi-source market information flow and the cyclical price trend characteristics, candidate price anomaly events for the target project are obtained; External correlation attribution is performed on the candidate price anomalies to obtain the characteristics of occasional price events in the target project.

[0029] The data includes various types of information from multiple sources in the target project's market, such as supplier quotations, industry price indices, government price control announcements, raw material supply reports from production areas, and logistics company transportation cost lists. All data entries are organized into a daily time series, and invalid content such as duplicate quotations and price data irrelevant to the target project's material types is removed. The data is then differentiated into those showing a continuous upward or downward trend for more than six months and those fluctuating regularly in a fixed quarterly or annual cycle. The continuous upward or downward trend for more than six months is defined as the trend component of the target project, and the regular fluctuation in a fixed quarterly or annual cycle is defined as the cyclical fluctuation component of the target project.

[0030] The daily base price value of the trend component is used as the benchmark for market price changes. The daily fluctuation amplitude of the cyclical fluctuation component is superimposed on the same time node of the benchmark. The correlation between the two at each time node is integrated in the order of time progression. The superimposed price value and corresponding cyclical attribute of each time node are retained to form the cyclical price trend characteristics of the target project that can reflect the cyclical change law of the target project's market price.

[0031] Align the actual price data and the corresponding time points of the cyclical price trend characteristics in the multi-source market information flow one by one, extract the actual price value and the theoretical price value of the cyclical price trend characteristics for each time point, calculate the absolute value of the difference between the two, divide the absolute value by the theoretical price value of the cyclical price trend characteristics to obtain the deviation range, set the deviation range threshold to 15%, filter out all price data of time points with deviation range calculation results exceeding 15%, and record information such as material type, price value, and deviation range of the corresponding node to obtain candidate price anomaly events for the target project.

[0032] For each candidate price anomaly event record, the specific time interval of the event is determined. Market policy documents within that time interval are retrieved to check for the introduction of new price control policies. Raw material supply records are retrieved to check for production reductions or shortages in production areas. Logistics and transportation data are retrieved to check for issues such as transportation route interruptions or fuel price increases. The specific external factors that triggered each candidate price anomaly event are identified. The specific external factors are then linked to information such as the type of material, price deviation data, and time of occurrence of the corresponding candidate price anomaly event to obtain the characteristics of the occasional price events of the target project.

[0033] The beneficial effects are that by carrying out a series of operations such as component separation, feature reconstruction, deviation comparison, and external correlation attribution on the multi-source market information flow of the target project, it is possible to accurately distinguish the trend component and cyclical fluctuation component of market price changes, effectively construct cyclical price trend characteristics that reflect the cyclical change law of market prices, accurately screen out candidate events related to price anomalies and clarify their triggering factors, and form a price-related feature system that combines regularity and anomaly. This provides comprehensive and reliable market data support for subsequent supply and demand forecasting of cyclical materials for the project, and strengthens the adaptability of subsequent demand forecasting work to the actual market environment.

[0034] S3. Based on the historical consumption characteristic spectrum, the price trend characteristics, and the price event characteristics, the predicted material demand of the target project is dynamically coupled and corrected to obtain the comprehensive predicted material demand of the target project. In this embodiment of the invention, the step of dynamically coupling and correcting the predicted material demand of the target project based on the historical consumption characteristic spectrum, the price trend characteristics, and the price event characteristics to obtain the comprehensive predicted material demand of the target project includes: The transient components in the historical consumption feature spectrum and the price event features are interpreted with time lag to obtain the transient causal relationship of the target project; Based on the transient causal relationship, the predicted value of material demand for the target project is reshaped to obtain the event-corrected prediction sequence of the target project. By performing long-term evolution analysis on the steady-state components in the historical consumption characteristic spectrum and the price trend characteristics, the synergistic elasticity parameters of the target project are obtained. Based on the aforementioned collaborative elasticity parameter, the overall trend baseline of the predicted material demand is progressively adjusted to obtain the trend-corrected prediction baseline for the target project. The event-corrected prediction sequence and the trend-corrected prediction baseline are fused in a time-varying manner to obtain the comprehensive prediction value of the material demand for the target project.

[0035] The process of time-varyingly fusing the event-corrected prediction sequence with the trend-corrected prediction baseline to obtain the comprehensive predicted value of material demand for the target project includes: The activity of the local fluctuation pattern and occurrence frequency of the event correction prediction sequence is evaluated to obtain the fluctuation activity data of the target project; The stability of the long-term smoothness and directional consistency of the trend correction prediction baseline is assessed to obtain the trend stability data of the target project. Based on the fluctuating active data and the trend stable data, the event impact and trend impact of the target project are weighted to obtain the dynamic fusion weight of the target project; Based on the dynamic fusion weights, the event-corrected prediction sequence and the trend-corrected prediction baseline are integrated in the time domain to obtain the comprehensive prediction value of the material demand of the target project.

[0036] The formula for calculating the dynamic fusion weight is as follows: ; In the formula, The dynamic fusion weights, The volatility activity parameter of the volatility active data. The trend stability parameter of the trend-stable data. The preset periodic adjustment range, The preset market cycle length, The preset periodic phase adjustment constant, It is a time variable.

[0037] By aligning the occurrence time intervals of transient components in the historical consumption characteristic spectrum with the occurrence time intervals of occasional price event characteristics, and performing daily matching at the daily time granularity, the overlap between the two time intervals is checked. Only transient components and price event combinations with an overlap time ratio of 100% are retained. The material consumption change magnitude of the transient component within the combination and the specific impact of the price event are analyzed. The response time and change magnitude of each type of transient component change after the occurrence of the price event are clarified, and a transient causal relationship of the target project that can reflect the temporal correlation and impact logic between the two is established.

[0038] Based on the direction and scope of the impact of price events on transient consumption as clearly defined in the transient causal relationship, the direction and magnitude of the numerical adjustment for the same time interval in the target project's material demand forecast are determined accordingly. The specific proportion of this adjustment is determined by referring to the proportion of material consumption changes caused by similar price events in the past, so that the adjusted forecast can accurately match the consumption changes caused by price events, forming an event correction forecast sequence for the target project.

[0039] Extract all change data of steady-state components in the historical consumption feature spectrum during the entire construction cycle of the target project, match them with all trend data of periodic price trends during the same construction cycle, and analyze the correlation and adaptation rules of steady-state components with long-term price trend changes in different construction stages such as foundation construction, main structure construction, and decoration construction. Extract the synergistic elasticity parameters of the target project that can characterize the linkage between the two at different construction stages.

[0040] Using the synergistic elasticity parameter as the core adjustment basis, the overall trend baseline of the material demand forecast is adjusted in stages according to the construction cycle. In each construction stage, the adjustment range and direction of the baseline are determined based on the synergistic elasticity parameter, so that the adjusted trend baseline can conform to the long-term linkage law between the steady-state component and the price trend, thus obtaining the trend correction forecast baseline of the target project.

[0041] The event-corrected prediction sequence is evaluated for its local fluctuation patterns and frequency to obtain active fluctuation data. The trend-corrected prediction baseline is analyzed for its long-term smoothness and directional consistency to obtain trend-stable data. Based on the specific values ​​of the active fluctuation data and trend-stable data in different time intervals, the event-corrected prediction sequence and the trend-corrected prediction baseline are weighted and integrated at the time domain level. The weight of the event-corrected prediction sequence is increased during periods of active event fluctuation, and the weight of the trend-corrected prediction baseline is increased during periods of trend stability to obtain the comprehensive forecast value of material demand for the target project.

[0042] Extract the full construction cycle data of the target project covered by the event correction prediction sequence, break it down into standard construction stages such as foundation construction, main structure construction, and decoration and finishing construction, and count the fluctuation range of the predicted value deviating from the average value of the stage within each construction stage. Set the fluctuation range threshold as 10% of the average value of the stage, and record the frequency of fluctuation exceeding the threshold within each construction stage. Distinguish the fluctuation pattern into sudden rise and fall type and continuous fluctuation type. Systematically integrate the combination information of fluctuation range, frequency of occurrence and fluctuation pattern corresponding to each construction stage to obtain the fluctuation activity data of the target project.

[0043] Extract all values ​​of the trend correction prediction baseline throughout the entire construction period, calculate the change range of baseline values ​​between adjacent construction stages, set the change range threshold to 5% of the value of the previous construction stage, and determine the construction stage with a change range below this threshold as a smooth state. Calculate the proportion of smooth state construction stages throughout the entire construction period, and at the same time, calculate the proportion of cases where the baseline trend direction remains consistent within three or more consecutive construction stages. Integrate the combined information of the smooth state proportion and the trend direction consistency proportion to obtain the trend stability data of the target project.

[0044] By retrieving information on fluctuation amplitude, frequency, and pattern from the active fluctuation data, and information on the proportion of smooth state and the proportion of trend direction consistency from the trend stable data, the weighting of the event impact corresponding to the event correction prediction sequence is increased during the construction phase when the active fluctuation data shows large fluctuation amplitude, high frequency, and sudden rise and fall. During the construction phase when the trend stable data shows a high proportion of smooth state and a high proportion of trend direction consistency, the weighting of the trend impact corresponding to the trend correction prediction baseline is increased. This forms a weighting scheme covering each stage of the entire construction cycle, resulting in the dynamic fusion weights of the target project.

[0045] According to the time sequence of the target project construction cycle, the values ​​of the event-corrected prediction sequence and the trend-corrected prediction baseline are mapped one by one to each construction stage. Based on the dynamic fusion weight corresponding to each construction stage, the values ​​of the two sequences are weighted and calculated. The result of the weighted calculation is the predicted value of material demand for that construction stage. The predicted values ​​of all construction stages are connected and integrated in chronological order to obtain the comprehensive predicted value of material demand for the target project.

[0046] The volatility activity parameter is derived from volatility activity data obtained by evaluating the local volatility patterns and frequency of occurrence in the event-corrected prediction sequence.

[0047] The trend stability parameter is derived from trend stability data obtained by assessing the long-term smoothness and directional consistency of the trend correction prediction baseline.

[0048] The preset periodic adjustment range is a value pre-set based on historical data of the cyclical fluctuations in the market prices of engineering materials in the industry in which the target project is located.

[0049] The preset market cycle length is a value pre-set after the time span of a complete market price fluctuation cycle of similar cyclical materials in the target project is statistically analyzed.

[0050] The preset cycle phase adjustment constant is a value pre-set based on the difference between the start time of the target project construction cycle and the start time of the market price cycle.

[0051] The time variable refers to the specific time nodes corresponding to each construction stage of the target project, from basic construction to decoration and renovation.

[0052] This content aims to dynamically allocate the event impact weights for the event-corrected prediction sequence and the trend impact weights for the trend-corrected prediction baseline based on the volatility and trend stability at different time points. By dynamically adjusting the weight proportions, the event impact weight is increased at time points where the event-corrected prediction sequence experiences large fluctuations and high frequency. Similarly, the trend impact weight is increased at time points where the trend-corrected prediction baseline shows a smooth and consistent trend. By integrating the event-corrected prediction sequence and the trend-corrected prediction baseline through time-domain weighted aggregation, a comprehensive forecast of material demand for the target project is obtained, which takes into account both short-term event fluctuations and long-term trend directions.

[0053] The beneficial effects are as follows: by interpreting the transient components and price event characteristics in the historical consumption feature spectrum with time lag, the transient causal relationship is clarified. Based on this relationship, the magnitude of the predicted value of material demand is reconstructed to form an event-corrected prediction sequence. At the same time, long-term evolution analysis of the steady-state components and price trend characteristics is carried out to extract the synergistic elasticity parameters and adjust them to obtain the trend-corrected prediction baseline. Then, by combining the volatile active data and the trend stable data and configuring dynamic fusion weights to complete the time-varying fusion, the comprehensive prediction value of material demand can take into account the impact of short-term event fluctuations and long-term trend trends, accurately match the actual material consumption pattern of the target project, improve the adaptability and reliability of the demand prediction results, and provide accurate and comprehensive data support for the two-way optimization of supply and demand of cyclical materials in subsequent projects.

[0054] S4. Perform efficiency decay deduction on the resource relationships of the cyclic material data in the target project to obtain the adjustable hypergraph resource network of the target project; In this embodiment of the invention, the step of performing performance attenuation deduction on the resource relationships of cyclic material data in the target project to obtain the adjustable hypergraph resource network of the target project includes: Attribute features are extracted from the cyclic material data of the target project to obtain multidimensional material feature data of the target project. Based on the multidimensional feature data of the materials, the cyclic material data is correlated and identified to obtain the resource topology relationship of the target project; The performance degradation trajectory of the multidimensional feature data of the materials is extrapolated over time to obtain the material efficiency decay sequence of the target project; Using the multidimensional feature data of the materials as nodes and the resource topology relationships as hyperedges, an initial hypergraph framework for the target project is constructed. Based on the material effectiveness decay sequence, the callable state attributes in the initial hypergraph framework are dynamically updated, and the connection strength of the hyperedges in the initial hypergraph structure is attenuated and adjusted to obtain the configurable hypergraph resource network of the target project.

[0055] Extract all attribute information from the target project's cyclic material data, including material type, specifications, remaining service life, current performance parameters, storage location, number of cycles, and maintenance records. Classify and aggregate these attributes according to material type. Verify specifications using a unified coding standard. Calculate and calibrate remaining service life based on factory standards and usage time. Review current performance parameters against factory test reports. Verify the completeness and continuity of maintenance records. Records with duplicate entries of the same batch of materials, records missing more than three performance parameters, or blank maintenance records with more than ten cycles are deemed invalid and removed. Integrate all verified attribute information according to classification standards to obtain multidimensional characteristic data of the target project's materials.

[0056] Based on the various attribute information contained in the multidimensional feature data of materials, the matching use relationships, substitution use relationships, and storage association relationships among different materials are sorted out. Matching use relationships are divided into two categories: strong matching and weak matching. Strong matching relationships are material combinations that must be used simultaneously during construction, while weak matching relationships are material combinations that can be used in combination during construction. Substitution use relationships are divided into two categories: complete substitution and partial substitution. Complete substitution means that a single material can directly replace another material without affecting construction quality, while partial substitution means that construction process parameters need to be adjusted after material substitution. Storage association relationships are divided into two categories: same-database storage and partitioned storage. At the same time, the degree of association is divided according to the monthly frequency of material co-use. The monthly frequency of co-use is determined as high degree of association, the monthly frequency of co-use is determined as medium degree of association, and the monthly frequency of co-use is determined as low degree of association. The association type and degree of association between each material and other materials are clarified. The association relationships of all materials are systematically sorted out and summarized to obtain the resource topology relationship of the target project.

[0057] Historical performance degradation data of materials of the same type as those used in the target project and which have completed their full life cycle are retrieved. Complete data from the past three years are selected as a reference. Combined with information such as the current cumulative usage time, average daily usage intensity, and weekly maintenance frequency of the circulating materials, usage time is converted into a standard unit of eight hours per day. Usage intensity is divided into three levels: light load, medium load, and heavy load according to equipment load rate. Maintenance frequency is divided into three levels: high frequency maintenance, medium frequency maintenance, and low frequency maintenance according to the number of maintenance times per week. The changes in various performance parameters of the circulating materials at different time nodes are predicted according to the time progression of the construction cycle. The predicted changes in performance parameters are arranged in chronological order to obtain the material performance degradation sequence of the target project.

[0058] Each data item in the multidimensional feature data of materials is treated as an independent node. A unique identification code is assigned to each node according to the rule of material type + specification + storage location. All attribute information of the corresponding material is marked in the node, including the verified specification, remaining service life, current performance parameters, etc. The resource topology relationship is used as the hyperedge connecting different nodes. The association type, association tightness and association influence range between the corresponding nodes are marked on the hyperedge. A hierarchical network structure is built according to the correspondence between nodes and hyperedges to obtain the initial hypergraph framework of the target project.

[0059] Referring to the material performance parameters at different time points in the material effectiveness decay sequence, 60% of the material's factory standard performance parameters is used as the preset usable threshold. Nodes with performance parameters higher than this threshold are marked as callable, while nodes with performance parameters lower than this threshold are marked as uncallable. At the same time, the connection strength of the hyperedges in the initial hypergraph structure is adjusted according to the degree of material effectiveness decay. The degree of effectiveness decay is divided into three levels: mild decay, moderate decay, and severe decay, based on the percentage decrease in performance parameters. Mild decay corresponds to a performance parameter decrease of less than 20%, and the hyperedge connection strength remains unchanged from the initial value. Moderate decay corresponds to a performance parameter decrease of 20% to 50%, and the hyperedge connection strength is reduced by one gradient. Severe decay corresponds to a performance parameter decrease of more than 50%, and the hyperedge connection strength is reduced by two gradients. After completing the status update of all nodes and the adjustment of the hyperedge connection strength, the configurable hypergraph resource network of the target project is obtained.

[0060] The beneficial effects are as follows: by extracting attribute features, identifying resource topology relationships, extrapolating performance degradation trajectories over time, constructing an initial hypergraph framework, and dynamically adjusting the callable state and hyperedge connection strength of the target project's cyclic material data, it is possible to accurately obtain multi-dimensional attribute information of materials, clarify the types and degrees of association between different materials, grasp the decay law of material performance over time, and construct an adjustable hypergraph resource network that can dynamically reflect the callable state and association strength of materials. This provides accurate and comprehensive resource data support for the two-way optimization of supply and demand and gap filling of cyclic materials in subsequent projects, and improves the utilization efficiency of cyclic materials and the scientificity and adaptability of resource allocation schemes.

[0061] S5. Perform two-way optimization of supply and demand on the comprehensive forecast value of material demand and the adjustable hypermap resource network to obtain the gap filling solution for the target project; In this embodiment of the invention, the step of performing two-way optimization of the comprehensive forecast of material demand and the adjustable hypergraph resource network to obtain a gap-filling solution for the target project includes: The comprehensive forecast of material demand is deconstructed into demand elements to obtain the demand constraints of the target project. Based on the aforementioned demand constraints, the scalable hypergraph resource network is traversed and filtered to obtain the basic allocation plan for the target project. Based on the hyperedge relationships in the scalable hypergraph resource network, the basic allocation plan is combined and alternatively deduced to obtain the extended allocation plan for the target project; The demand constraints are compared item by item with the extended allocation plan to obtain the material gap list of the target project, and a gap filling plan for the target project is constructed based on the material gap list.

[0062] The comprehensive forecast of material demand is broken down into all its components. The forecast is then broken down according to the construction stages of the target project. Core demand elements for each construction stage are extracted, including material type, specifications, consumption quantity, usage time, performance parameter requirements, and storage location requirements. Specific standards for each element are defined. The usage time must perfectly match the start and end times of the corresponding construction stage of the target project. Performance parameter requirements must not be lower than the national standards for the construction and use of this type of material. Storage location requirements must be close to the designated storage point in the corresponding construction area. These defined elements and standards are then categorized and organized according to the construction stages to obtain the demand constraints for the target project.

[0063] Based on the standards specified in the demand constraints, each node in the available hypermap resource network is checked individually. The checks include whether the material type and specifications marked on the node are completely consistent with the constraints, whether the material performance parameters meet the preset national standards for construction and use, whether the material availability status is marked as available, and whether the hyperedge relationships corresponding to the node meet the construction support requirements. All qualified nodes and their corresponding hyperedge relationships are integrated according to the construction stage to form a basic allocation plan for the target project that can meet the basic requirements.

[0064] Information on the association type and degree of association of hyperedge annotations in the deployable hypermap resource network is retrieved. For each material in the basic deployment plan, a combination substitution deduction is carried out based on the substitution relationship of complete substitution, partial substitution, etc. of the hyperedge annotations. Materials that are completely substituted must meet the requirements of being completely consistent with the original materials in terms of specifications, performance parameters, and usage scenarios, and can directly replace the corresponding materials in the basic deployment plan. Materials that are partially substituted need to be replaced after adjusting the usage ratio or construction time according to the construction process requirements. All the deployment plans after substitution combination are classified and summarized according to the substitution type to obtain the extended deployment plan of the target project.

[0065] Each standard in the demand constraints is checked against each combination scheme in the extended allocation plan. The checks include whether the quantity of materials meets the demand, whether the performance parameters meet the national standards for construction use, whether the usage time nodes match the corresponding construction stages, and whether the storage location meets the specified requirements. Items in the extended allocation plan that do not meet the standards are marked, and information such as the type of materials that do not meet the standards, the quantity of shortages, performance differences, and time deviations are compiled. These marked items are systematically organized to obtain a material shortage list for the target project. Based on the type and scale of each shortage in the material shortage list, targeted filling measures are formulated. External procurement requires screening qualified suppliers and clarifying the delivery cycle. Cross-project material allocation requires internal approval processes and determination of allocation time. Performance upgrades and modifications require specific technical modification plans. These measures are bound one by one with the corresponding items in the shortage list to obtain the shortage filling plan for the target project.

[0066] The beneficial effects are as follows: by deconstructing demand elements from the comprehensive forecast of material demand to clarify demand constraints, a basic allocation plan is formed by traversing and screening the available hypermap resource network based on demand constraints. An extended allocation plan is obtained by combining and substituting the basic allocation plan based on the hyperedge relationships of the available hypermap resource network. The demand constraints and the extended allocation plan are compared item by item to clarify the material gap list and construct gap-filling solutions. This can accurately anchor the various standards of engineering material demand, fully explore the utilization potential of available resources, enrich the range of material allocation solutions, clearly define the specific content of material supply and demand gaps, and form scientific and targeted gap-filling solutions. This provides reliable support for the efficient allocation of engineering cyclical materials and improves the accuracy and rationality of engineering material supply and demand matching.

[0067] S6. Based on the gap-filling scheme, perform perturbation simulation on the decision resilience in the configurable hypergraph resource network to obtain the resilience decision report of the target project.

[0068] In this embodiment of the invention, the step of performing perturbation simulation on the decision resilience in the scalable hypergraph resource network based on the gap-filling scheme to obtain a resilience decision report for the target project includes: Based on preset typical risk scenarios, perturbations are applied to key nodes or hyperedges in the configurable hypergraph resource network to obtain the perturbed network state of the target project. Based on the state of the disturbed network, the gap filling scheme is re-executed for verification, and the performance degradation path of the gap filling scheme is observed to obtain the performance degradation trajectory of the target project. Based on the performance degradation trajectory of the above scheme, the source location of the configurable hypergraph resource network is traced to obtain the key vulnerable nodes and their relationships in the target project. The gap-filling scheme, the performance degradation trajectory of the scheme, and the key vulnerable nodes and their relationships are annotated with strategies to obtain the resilience decision report of the target project.

[0069] This study identifies and pre-sets typical risk scenarios commonly encountered in the engineering field, such as material supply disruptions, transportation disruptions, and sudden performance failures. Material supply disruptions specifically refer to scenarios where the duration of a core supplier's supply interruption exceeds the duration of a single construction phase of the target project. Transportation disruptions specifically refer to scenarios where the closure of major transportation channels leads to material transportation delays exceeding the pre-set duration. Sudden performance failures specifically refer to scenarios where the failure of core components of critical equipment causes a sharp drop in material performance parameters. The study clarifies that the affected objects for each risk scenario are key nodes or hyperedges marked with high density in the deployable hypergraph resource network that support the core processes of the target project. For key nodes, a disturbance is applied that reduces performance parameters to below 60% of the material's factory standard. For hyperedges, a disturbance is applied that directly reduces connection strength from high density to low density by two gradients. The study comprehensively records the changes in the callable status of all nodes, performance parameter values, connection strength levels of all hyperedges, and the scope of associated impact after the disturbances are applied, thus obtaining the disturbed network state of the target project.

[0070] The gap-filling plan was fully applied to the disrupted network state. Internal material allocation, cross-project material transfer, and external supplier procurement were implemented step by step according to the priority order set in the plan. The execution of each measure followed the time nodes, material quantity and quality standards specified in the plan. The specific numerical changes of performance indicators such as material arrival rate, supply and demand matching rate and cost control rate were tracked and recorded throughout the entire process of plan execution. The material arrival rate is the ratio of the actual material arrival quantity to the planned material arrival quantity. The supply and demand matching rate is the ratio of the batches of materials that successfully matched the demand to the total batches of demand. The cost control rate is the ratio of the actual input cost to the planned cost. The complete change path of each performance indicator from the initial state value to the final state value was sorted out according to the time sequence of plan execution to obtain the performance degradation trajectory of the plan for the target project.

[0071] By comparing the decay magnitude values ​​of each performance index with the decay occurrence node in the performance decay trajectory of the scheme, the nodes and hyperedges involved in the operation links corresponding to the decay node in the reconfigurable hypergraph resource network are traced backward. The node corresponding to the performance index decay magnitude reaching the maximum decay magnitude of all corresponding nodes is identified as a critical vulnerable node, and the hyperedge associated with the critical vulnerable node is identified as a critical vulnerable association. The system integrates information such as the material type, performance parameters, callable status, association type, and tightness of critical vulnerable associations of all critical vulnerable nodes to obtain the critical vulnerable nodes and associations of the target project.

[0072] The entire execution process of the gap filling solution, operation records of each stage, details of indicator changes in the solution's performance degradation trajectory, statistical data on degradation magnitude, a specific list of key vulnerable nodes and their relationships, and judgment criteria are comprehensively summarized. Specific optimization strategies are proposed to address the shortcomings of the gap filling solution exposed under disrupted conditions, such as excessively long external procurement cycles and cumbersome cross-project transfer approval processes. Enhanced protection measures are proposed for key vulnerable nodes and their relationships, such as conducting regular performance testing, establishing a backup material reserve, and strengthening the stability of relationships. All summarized information and strategic measures are systematically integrated and detailed with annotations to obtain the resilience decision report for the target project.

[0073] The beneficial effects are as follows: by perturbing key nodes or hyperedges of the scalable hypergraph resource network based on preset typical risk scenarios to obtain the state of the disturbed network, the gap-filling scheme is re-executed and verified based on this state, and the performance degradation path of the scheme is observed to obtain the performance degradation trajectory of the scheme. Combined with the degradation trajectory, the hypergraph resource network is traced and located to obtain key vulnerable nodes and their relationships. Then, strategy annotations are added to the gap-filling scheme, the degradation trajectory, and the vulnerable nodes and their relationships to obtain a resilience decision report. This can fully verify the execution effectiveness and anti-disturbance capability of the gap-filling scheme under risk scenarios, accurately identify the weak links in the scalable hypergraph resource network, supplement the gap-filling scheme with targeted optimization strategies and protective measures, improve the resilience and reliability of engineering material supply and demand allocation decisions, and provide scientific and solid decision support for the material security of the target project.

[0074] like Figure 2 The diagram shown is a functional block diagram of a deep learning-based engineering cyclic material supply and demand forecasting system provided in an embodiment of the present invention.

[0075] The deep learning-based engineering cyclic material supply and demand forecasting system 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the deep learning-based engineering cyclic material supply and demand forecasting system 100 may include a consumption feature analysis module 101, a price feature recognition module 102, a demand coupling correction module 103, a hypergraph network inference module 104, a supply and demand bidirectional optimization module 105, and a resilience decision simulation module 106. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0076] In this embodiment, the functions of each module / unit are as follows: The consumption feature analysis module 101 is used to perform multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project. The price feature recognition module 102 is used to perform pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend features and occasional price event features of the target project. The demand coupling correction module 103 is used to dynamically couple and correct the predicted value of the material demand of the target project based on the historical consumption characteristic spectrum, the price trend characteristics and the price event characteristics, so as to obtain the comprehensive predicted value of the material demand of the target project. The hypergraph network inference module 104 is used to perform efficiency decay inference on the resource relationship of cyclic material data in the target project, and obtain the adjustable hypergraph resource network of the target project. The supply and demand two-way optimization module 105 is used to perform two-way optimization of the comprehensive forecast value of material demand and the adjustable hypermap resource network to obtain the gap filling solution for the target project. The resilience decision simulation module 106 is used to perform perturbation simulation on the decision resilience in the configurable hypergraph resource network based on the gap filling scheme, and obtain the resilience decision report of the target project.

[0077] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0078] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0079] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0080] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0081] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A deep learning-based method for predicting the supply and demand of engineering recurring materials, characterized in that, The method includes: S1. Perform multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project; S2. Perform pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend characteristics and occasional price event characteristics of the target project; S3. Based on the historical consumption characteristic spectrum, the price trend characteristics, and the price event characteristics, the predicted material demand of the target project is dynamically coupled and corrected to obtain the comprehensive predicted material demand of the target project. S4. Perform efficiency decay deduction on the resource relationships of the cyclic material data in the target project to obtain the adjustable hypergraph resource network of the target project; S5. Perform two-way optimization of supply and demand on the comprehensive forecast value of material demand and the adjustable hypermap resource network to obtain the gap filling solution for the target project; S6. Based on the gap-filling scheme, perform perturbation simulation on the decision resilience in the configurable hypergraph resource network to obtain the resilience decision report of the target project.

2. The deep learning-based engineering cyclic material supply and demand forecasting method as described in claim 1, characterized in that, The process involves performing multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project, including: By stripping the historical material consumption data of the target project from time sequence, the basic consumption data stream of the target project is obtained. Based on the construction blueprint and milestone nodes of the target project, the basic consumption data stream is stage-anchored to obtain the consumption data fragments of the target project. Multimodal data fusion is performed on the consumed data segments to obtain the baseline consumption mode and offset consumption mode of the target project; Trend fitting is performed on the baseline consumption mode to construct the steady-state spectrum of the target project; By attributing the offset consumption mode to its source, the transient disturbance set of the target project is obtained; Based on the transient disturbance set, the steady-state spectrum is annotated and corrected to obtain the historical consumption characteristic spectrum of the target project.

3. The deep learning-based method for predicting the supply and demand of engineering recurring materials as described in claim 1, characterized in that, The process of pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend characteristics and occasional price event characteristics of the target project includes: The multi-source market information flow of the target project is subjected to component separation to obtain the trend component and periodic fluctuation component of the target project; Based on the trend component and the periodic fluctuation component, the multi-source market information flow is reconstructed to obtain the periodic price trend characteristics of the target project. By comparing the deviations between the multi-source market information flow and the cyclical price trend characteristics, candidate price anomaly events for the target project are obtained; External correlation attribution is performed on the candidate price anomaly events to obtain the characteristics of the sporadic price events of the target project.

4. The deep learning-based method for predicting the supply and demand of engineering cyclic materials as described in claim 1, characterized in that, The dynamic coupling and correction of the material demand forecast for the target project based on the historical consumption characteristic spectrum, the price trend characteristics, and the price event characteristics to obtain the comprehensive material demand forecast for the target project includes: The transient components in the historical consumption feature spectrum and the price event features are interpreted with time lag to obtain the transient causal relationship of the target project; Based on the transient causal relationship, the predicted value of material demand for the target project is reshaped to obtain the event-corrected prediction sequence of the target project. By performing long-term evolution analysis on the steady-state components in the historical consumption characteristic spectrum and the price trend characteristics, the synergistic elasticity parameters of the target project are obtained. Based on the aforementioned collaborative elasticity parameter, the overall trend baseline of the predicted material demand is progressively adjusted to obtain the trend-corrected prediction baseline for the target project. The event-corrected prediction sequence and the trend-corrected prediction baseline are fused in a time-varying manner to obtain the comprehensive prediction value of the material demand for the target project.

5. The deep learning-based method for predicting the supply and demand of engineering cyclic materials as described in claim 4, characterized in that, The process of time-varyingly fusing the event-corrected prediction sequence with the trend-corrected prediction baseline to obtain the comprehensive predicted value of material demand for the target project includes: The activity of the local fluctuation pattern and occurrence frequency of the event correction prediction sequence is evaluated to obtain the fluctuation activity data of the target project; The stability of the long-term smoothness and directional consistency of the trend correction prediction baseline is assessed to obtain the trend stability data of the target project. Based on the fluctuating active data and the trend stable data, the event impact and trend impact of the target project are weighted to obtain the dynamic fusion weight of the target project; Based on the dynamic fusion weights, the event-corrected prediction sequence and the trend-corrected prediction baseline are integrated in the time domain to obtain the comprehensive prediction value of the material demand of the target project.

6. The deep learning-based method for predicting the supply and demand of engineering recurring materials as described in claim 5, characterized in that, The formula for calculating the dynamic fusion weight is as follows: ； In the formula, The dynamic fusion weights, The volatility activity parameter of the volatility active data. The trend stability parameter of the trend-stable data. The preset periodic adjustment range, The preset market cycle length, The preset periodic phase adjustment constant, It is a time variable.

7. The deep learning-based method for predicting the supply and demand of engineering recurring materials as described in claim 1, characterized in that, The process of performing efficiency decay deduction on the resource relationships of cyclic material data in the target project to obtain the adjustable hypergraph resource network of the target project includes: Attribute features are extracted from the cyclic material data of the target project to obtain multidimensional material feature data of the target project. Based on the multidimensional feature data of the materials, the cyclic material data is correlated and identified to obtain the resource topology relationship of the target project; The performance degradation trajectory of the multidimensional feature data of the materials is extrapolated over time to obtain the material efficiency decay sequence of the target project; Using the multidimensional feature data of the materials as nodes and the resource topology relationships as hyperedges, an initial hypergraph framework for the target project is constructed. Based on the material effectiveness decay sequence, the callable state attributes in the initial hypergraph framework are dynamically updated, and the connection strength of the hyperedges in the initial hypergraph structure is attenuated and adjusted to obtain the configurable hypergraph resource network of the target project.

8. The deep learning-based method for predicting the supply and demand of engineering recurring materials as described in claim 1, characterized in that, The process of optimizing the comprehensive forecast of material demand and the available hypergraph resource network to obtain a gap-filling solution for the target project includes: The comprehensive forecast of material demand is deconstructed into demand elements to obtain the demand constraints of the target project. Based on the aforementioned demand constraints, the scalable hypergraph resource network is traversed and filtered to obtain the basic allocation plan for the target project. Based on the hyperedge relationships in the scalable hypergraph resource network, the basic allocation plan is combined and alternatively deduced to obtain the extended allocation plan for the target project; The demand constraints are compared item by item with the extended allocation plan to obtain the material gap list of the target project, and a gap filling plan for the target project is constructed based on the material gap list.

9. The deep learning-based method for predicting the supply and demand of engineering cyclic materials as described in claim 1, characterized in that, Based on the gap-filling scheme, the decision resilience in the scalable hypergraph resource network is perturbed and simulated to obtain a resilience decision report for the target project, including: Based on preset typical risk scenarios, perturbations are applied to key nodes or hyperedges in the configurable hypergraph resource network to obtain the perturbed network state of the target project. Based on the state of the disturbed network, the gap filling scheme is re-executed for verification, and the performance degradation path of the gap filling scheme is observed to obtain the performance degradation trajectory of the target project. Based on the performance degradation trajectory of the above scheme, the source location of the configurable hypergraph resource network is traced to obtain the key vulnerable nodes and their relationships in the target project. The gap-filling scheme, the performance degradation trajectory of the scheme, and the key vulnerable nodes and their relationships are annotated with strategies to obtain the resilience decision report of the target project.

10. A deep learning-based engineering cyclic material supply and demand forecasting system, characterized in that, The system for implementing the deep learning-based engineering recurring material supply and demand forecasting method of claim 1, the system comprising: The consumption feature analysis module is used to perform multi-dimensional feature analysis on the historical material consumption data of the target project to obtain the historical consumption feature spectrum of the target project. The price feature recognition module is used to perform pattern recognition on the multi-source market information flow of the target project to obtain the periodic price trend features and occasional price event features of the target project. The demand coupling correction module is used to dynamically couple and correct the predicted value of the material demand of the target project based on the historical consumption characteristic spectrum, the price trend characteristics and the price event characteristics, so as to obtain the comprehensive predicted value of the material demand of the target project. The hypergraph network inference module is used to perform efficiency decay inference on the resource relationships of cyclic material data in the target project, and obtain the adjustable hypergraph resource network of the target project. The supply and demand two-way optimization module is used to perform two-way optimization of the comprehensive forecast value of the material demand and the adjustable hypergraph resource network to obtain the gap filling solution for the target project. The resilience decision simulation module is used to perform perturbation simulation on the decision resilience in the scalable hypergraph resource network based on the gap filling scheme, and obtain the resilience decision report of the target project.

Images