Machine learning-based device material price regulation system

Abstract

The present application relates to pumped storage engineering technical field, specifically for a kind of equipment material price regulation system based on machine learning, system includes warehouse position feature acquisition module mapping environment wave generation correlation table, chain type association construction module tracks transmission path and generates association chain, price energy regulation module constructs energy spectrum and extracts scale set, node priority allocation module divides urgency grade and generates calibration table, comprehensive prediction optimization module checks quotation rate and repairs trend anomaly, generates equipment material price optimization result table.The present application, by multidimensional acquisition constructs chain structure to map price activity, utilizes dynamic calibration and time sequence matching to repair abnormal section trend, combined with supply-demand rhythm matching to realize real-time optimization arrangement, converts cost data into active regulation mode, by the linkage balance mechanism under the coupling of multiple factors, establish the price balance trend with adaptive correction ability, improve the coherence and reliability of cost research.

Description

Technical Field

[0001] This invention relates to the field of pumped storage engineering technology, and in particular to a machine learning-based system for predicting and optimizing equipment and material prices. Background Technology

[0002] The field of pumped storage engineering technology encompasses the related technologies for energy storage and regulation in power systems using reversible hydro-turbine generator units. The core of this field lies in achieving bidirectional conversion of electrical energy to potential energy through the water potential difference between high and low reservoirs, thereby realizing peak-valley regulation and energy redistribution during power system load changes. Against the backdrop of accelerated construction of new power systems and adjustments to power plant capacity pricing policies, pumped storage engineering is facing a new situation of continuously expanding construction scale and accelerated pace. Its engineering cost management is gradually shifting from traditional investment control to comprehensive management covering the entire process from decision-making, design, procurement, construction to operation and maintenance. Due to the large size, variety, and high investment proportion of the engineering equipment, there is an urgent need for... A systematic, unified, and timely equipment and material price reference system is needed. However, current equipment and material price information suffers from problems such as fragmented sources, inconsistent definitions, insufficient utilization of historical data, and a lack of systematic forecasting capabilities. These issues hinder refined cost control and investment decisions at the group level. To meet the needs of comprehensive cost management, it is necessary to systematically review the differentiated requirements for equipment and material price information at each construction stage. This requires laying a technical foundation in areas such as historical price data collection and organization, equipment and material attribute analysis, key category identification, classification and attribute combination methods, and reference price statistical model construction. Furthermore, it necessitates the development of price trend prediction models and price index models to achieve price level monitoring and trend analysis. Simultaneously, a comprehensive application system encompassing institutional frameworks, business processes, and information platforms needs to be established, forming a closed-loop mechanism for compilation, publication, application, feedback, and correction. The equipment and material reference price publication platform should be improved to achieve centralized management, real-time querying, statistical analysis, and visualization of price information, providing support for cost control and procurement management of pumped storage projects.

[0003] The Machine Learning-Based Equipment and Material Price Prediction and Optimization System refers to a system that uses machine learning methods to predict and optimize the price information of major equipment and materials throughout the construction, operation, and maintenance of pumped storage projects. Based on multi-source data such as historical procurement records, market price fluctuation data, equipment and material attribute parameters, supply cycle information, and regional transportation costs, the system constructs a price prediction mechanism through sample feature extraction, variable weight calculation, and model comparison. This enables the modeling and calculation of price change trends for different categories of equipment and materials. The system further integrates a price index model to calculate and dynamically adjust reference prices for typical equipment and materials, forming a price database and update rules applicable to each construction stage. Using data collection, attribute combination, classification statistics, and trend analysis as technical means, a unified price data structure and prediction process are constructed, providing data sources and algorithmic support for subsequent price information management, publication, and optimization.

[0004] Existing technologies often employ static statistics and experience comparisons in equipment and material price analysis. The warehousing environment and material flow status are not incorporated into the price calculation logic, leading to a disconnect in price data over time. This fails to reflect the potential impact of environmental changes on procurement costs, and price fluctuations remain at the level of fitting historical data. The linkage effect between different warehousing segments or material categories is ignored, resulting in significant deviations when comparing prices across multiple time periods or locations. The data update cycle is long and lacks a self-correction mechanism, making it impossible to identify and handle abnormal fluctuations in a timely manner. Consequently, the price forecast results are not responsive enough to real-time changes, causing a lag in price trend judgments and making it difficult to meet the management needs of cost coordination and construction schedule planning. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a machine learning-based system for predicting and optimizing equipment and material prices.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a machine learning-based equipment and material price prediction and optimization system, the system comprising: a warehouse feature acquisition module, which acquires data on transportation time, environmental temperature and humidity and storage period of the factory and upper and lower reservoirs, compares the differences between transportation and environment, determines the environmental trend based on the storage period, records fluctuations and establishes a number mapping, filters synchronous segments, and generates a warehouse feature association table. The chain association construction module, based on the synchronous segment and warehouse number in the warehouse feature association table, calls the transportation distance and equipment specifications to determine the association, combines the material supply cycle to match the location, identifies the response order of warehouse nodes and tracks the transmission path, and generates a warehouse progressive association chain. The price and energy control module obtains the purchase price sequence and entry batch of each material according to the material category in the progressive association chain of the warehouse, constructs an energy distribution map by weighted mapping, compares the activity differences of different materials, performs smoothing and enhancement, and obtains the price and energy balance set. The node priority allocation module calls upon the price energy balance concentration energy characteristics and price trends to obtain equipment and material supply intervals, usage rhythms, and importance level codes. It compares and matches differences to divide the hierarchical node sets, calculates the distribution density, sets an urgency threshold, defines the levels, and generates a warehouse node priority calibration table.

[0007] As a further embodiment of the present invention, the position feature association table includes environmental fluctuation feature parameters, position number index and material category mapping information; the position progressive association chain includes transmission path sequence, node response relationship and time sequence association elements; the price energy balance set includes energy activity distribution, price fluctuation direction and material energy balance coefficient; and the position node priority calibration table includes node hierarchy division, urgency level and distribution density characteristics.

[0008] As a further embodiment of the present invention, the warehouse feature acquisition module includes an environmental monitoring submodule, a state difference determination submodule, and a feature mapping construction submodule; The environmental monitoring submodule acquires the transportation time and environmental temperature and humidity values ​​of the underground plant, upper reservoir and lower reservoir, detects the integrity of the sampled data and performs linear filling of adjacent time periods for missing points, performs time alignment matching of transportation time and temperature and humidity values ​​according to the sampling time interval, calculates the mean sequence and change gradient of each warehouse in a continuous time period, organizes the time series of each warehouse into a structured array, and generates a warehouse time series matrix. The state difference determination submodule, based on the warehouse time series matrix, calls the transportation time value and temperature and humidity value of each warehouse, calculates the rate of change between adjacent time nodes, judges the degree of state deviation at the same time node according to the transportation time change threshold and temperature and humidity fluctuation threshold, takes the difference between the two deviation values ​​to calculate the comprehensive deviation, calculates the time average of the comprehensive deviation according to the warehouse number, calculates the corresponding periodic change interval, and obtains the warehouse environment difference amplitude value. The feature mapping construction submodule calls the difference amplitude value of the warehouse environment and the material storage cycle value, aggregates the difference amplitude value within the time window, calculates the synchronous volatility of the difference value between adjacent warehouses, logically maps the volatility to the warehouse number and material category identifier according to the time node, filters out the warehouse combination segment with the synchronous rate higher than the set threshold, establishes the corresponding mapping relationship, and generates a warehouse feature association table.

[0009] As a further aspect of the present invention, the chain association construction module includes a segment matching submodule, a node response identification submodule, and a dynamic path reorganization submodule; The segment matching submodule obtains the synchronously changing segments and corresponding storage location number combinations from the storage location feature association table, collects transportation distance data and equipment specification data between each storage location, performs numerical correspondence matching on the transportation distance and equipment specification of the same number combination, calculates the ratio of the transportation distance difference to the equipment specification difference, compares the ratio with the preset synchronous judgment benchmark value, filters storage location combination segments that meet the conditions based on the comparison results, and generates a storage location matching association coefficient set. The node response identification submodule calls the warehouse matching correlation coefficient set, collects the material supply cycle time series, performs time series position matching on the warehouse combination of each segment, calculates the delay interval between the material supply point and the warehouse response point under continuous time nodes, performs sorting judgment on the change direction of the delay interval and the warehouse number combination, generates a continuous interval response sequence based on the sorting relationship, and obtains the warehouse node response order value. The dynamic path reorganization submodule, based on the response sequence value of the warehouse node, calls the combination of warehouse number in the corresponding segment, calculates the time interval difference between the response sequences of the nodes, performs a correlation judgment on adjacent interval values, filters node pairs with a time difference less than the judgment threshold, arranges the continuous node chain in ascending order of time, records the sequence structure of the continuous node chain as the associated path, and generates a progressive association chain of warehouses.

[0010] As a further embodiment of the present invention, the price energy control module includes a price data mapping submodule, an energy distribution calculation submodule, and a price trend extraction submodule; The price data mapping submodule obtains the purchase price sequence and the number of batches of materials corresponding to each warehouse node in the warehouse progressive association chain. It synchronously detects the purchase price sequences of silicon steel sheets, steel plates and concrete lining materials according to time nodes, calculates the fluctuation range value of the price sequence of each material, performs a weighted product operation on the price fluctuation range value based on the number of batches of materials entering the warehouse as the weight, and takes the time average after summing as the price fluctuation level of each material to generate a price fluctuation level value. The energy distribution calculation submodule calculates the data difference between adjacent regions of each material based on the price fluctuation level value, maps the difference results into a two-dimensional coordinate grid in time order, calculates the regional activity based on the root mean square value of fluctuation of adjacent points in the grid, performs difference smoothing on regions with an activity level higher than the activity level threshold, performs amplitude amplification on regions with an activity level lower than the activity level threshold, and integrates the processed activity matrix into energy spatial distribution information to obtain a set of energy activity coefficients. The price trend extraction submodule calls the energy activity coefficient set, performs direction judgment operation on the activity values ​​arranged along the time axis in the matrix, calculates the activity change ratio between consecutive time nodes, confirms the time period with consecutive change ratio signs, distinguishes between positive and negative sequences and counts the frequency proportion, determines the dominant trend of price energy change based on the direction with the largest proportion, and generates a price energy balance set.

[0011] As a further embodiment of the present invention, the node priority allocation module includes a matching degree determination submodule, a hierarchy division calculation submodule, and an urgency ranking submodule; The matching degree determination submodule acquires the energy activity characteristics and price change trend information of each position node in the price energy balance set, collects equipment and material construction supply interval and usage rhythm data, performs difference calculation on the cycle length of the supply interval and usage rhythm, compares the supply cycle difference ratio with the energy activity value, determines the matching degree deviation interval, generates deviation degree index for each node, establishes a matching deviation result sequence, and obtains the node matching deviation mapping. The hierarchical division calculation submodule, based on the node matching deviation mapping, calls the importance level code of each node, performs cross-judgment by the importance level code and activity feature, filters out nodes with deviation values ​​less than the set judgment threshold and classifies them into the high matching layer, and nodes with deviation values ​​greater than the threshold are classified into the low matching layer, calculates the occurrence frequency of each layer of nodes in the position sequence and calculates the average distribution density, establishes the hierarchical node classification result, and obtains the hierarchical node distribution map; The urgency ranking submodule, based on the hierarchical node distribution map, calls price change trend information to divide the distribution density of each level of nodes into intervals, calculates the activity weighted difference between adjacent intervals, compares the weighted difference with the urgency judgment threshold, filters intervals exceeding the threshold as high urgency level, arranges all urgency level intervals in time axis order to form the response sequence of position nodes, and generates a position node priority calibration table.

[0012] As a further aspect of the present invention, the process of calculating the difference between the supply interval and the cycle length of the usage rhythm is specifically as follows: Based on the actual time length of adjacent nodes in the equipment and material construction supply interval time series and the usage rhythm time series, calculate the time values ​​of the supply interval cycle and the usage rhythm cycle respectively, take the difference between the two and calculate the supply cycle difference ratio with the supply interval cycle as the denominator. The specific process of calculating based on the supply cycle difference ratio and the energy activity value is as follows: The energy activity value corresponding to each warehouse node is compared with the difference between its supply cycle. If the difference exceeds the set supply matching benchmark threshold, it is determined that there is a matching deviation at the node, and the degree of deviation of the node is recorded.

[0013] As a further aspect of the present invention, the process of classifying nodes whose deviation value is less than the set judgment threshold into the high matching layer is specifically as follows: Among all the calculated node deviation indices, select the set of nodes that are less than the matching degree judgment threshold, and arrange them in ascending order according to the value of the deviation index to form a high matching layer node sequence list. The process of calculating the frequency of occurrence of each layer of nodes in the warehouse sequence and determining the average distribution density is as follows: The ratio of the number of nodes in the high-matching layer and low-matching layer appearing in the position sequence to the total number of nodes is calculated. The distribution ratio of each layer is obtained, and the average value of the product of the distance between adjacent nodes and the node density is calculated. The process of calculating the activity-weighted difference between adjacent intervals is as follows: Based on the average activity value of adjacent intervals in the hierarchical node distribution map, the difference is calculated and multiplied by the weighting coefficient of the time span of each interval to generate the activity weighted difference result. This result is then compared with the urgency determination threshold to identify high urgency level intervals for forming the response sequence of position nodes.

[0014] As a further aspect of the present invention, the system further includes: The comprehensive forecasting and optimization module, based on the priority sequence and time axis distribution information of the warehouse node priority calibration table, calls the purchase quotation and update time series, groups and arranges them according to priority and calculates the rate of change, verifies the consistency of the trend, filters out abnormal segments for correction and repair, integrates them to form an overall price balance trend, and generates an equipment and material price optimization result table. The equipment and material price optimization result table includes balance trend data, correction segment information, and price optimization sequence.

[0015] As a further embodiment of the present invention, the comprehensive prediction optimization module includes a sequence grouping calculation submodule, a trend deviation correction submodule, and a price balance integration submodule; The sequence grouping calculation submodule obtains the node priority sequence and time axis distribution in the warehouse node priority calibration table, calls the quotation sequence and update time sequence of the pumped storage project procurement plan, groups and arranges the quotations according to the node priority, calculates the change amplitude of the quotations in the same group between adjacent time nodes, calculates the change rate in combination with the update time interval, and organizes the rate trend in time order to generate a quotation grouping time series graph. The trend deviation correction submodule, based on the price group time series graph, calls the time series logic in the position node priority calibration table, performs a consistency comparison of the change rate direction between groups, calculates the deviation angle between the rate direction and the logic direction, filters the node segments whose deviation angle exceeds the set threshold, adjusts the direction of abnormal segments and continuously smooths the adjacent rate difference, organizes and repairs the node sequence, and obtains the price trend repair structure. The price balance integration submodule, based on the price trend repair structure, calls the node priority information, rearranges the repaired sequence in chronological order, calculates the balance ratio of price changes between adjacent nodes, extracts and aggregates the time segments of continuous balance distribution, organizes the aggregation results to form the overall price evolution pattern, and generates an optimized table of equipment and material prices.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by parallel collection and dynamic comparison of transportation timeliness, environmental parameters, and material cycle data from multiple storage areas, a multi-dimensional chain structure reflecting the coupling relationship between material flow logic and environmental response is constructed. This forms a continuous mapping mechanism between price status and material activity, enabling price change patterns to no longer rely on static time nodes for judgment, but to form an extensible equilibrium trend across multiple stages. Dynamic calibration and fluctuation correction of price energy distribution make data trends more stable, and abnormal segments are adaptively repaired through time-series matching. Price change direction has the ability to actively correct under multi-factor coupling conditions. The matching process of material usage rhythm and supply interval achieves real-time priority arrangement, forming an overall linkage equilibrium among price elements. This transforms cost data from passive statistics to active control, improving the consistency and reliability of price trend analysis. Attached Figure Description

[0017] Figure 1 To illustrate a flowchart of the equipment and material price prediction and optimization system according to an embodiment of the present invention; Figure 2 To illustrate the acquisition flowchart of the warehouse feature acquisition module of the equipment and material price prediction and optimization system according to an embodiment of the present invention; Figure 3 To illustrate the flowchart of obtaining the chain-linked construction module of the equipment and material price prediction and optimization system according to one embodiment of the present invention; Figure 4 To illustrate the flowchart of the price and energy control module of the equipment and material price prediction and optimization system according to one embodiment of the present invention; Figure 5 To illustrate the flowchart of the node priority allocation module of the equipment and material price prediction and optimization system according to an embodiment of the present invention; Figure 6This is a flowchart illustrating the acquisition process of the comprehensive prediction and optimization module of the equipment and material price prediction and optimization system according to an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art will understand that this description is exemplary, and the present invention is not limited to these specific embodiments.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] Figure 1 This invention illustrates a machine learning-based equipment and material price prediction and optimization system according to a specific embodiment, such as... Figure 1As shown, the system includes: a warehouse feature acquisition module, which acquires transportation time, environmental temperature and humidity, and material storage cycle values ​​for underground plants, upper reservoirs, and lower reservoirs; compares the differences in transportation time and environmental temperature and humidity values; determines the environmental change trend within a time window based on the material storage cycle value; records environmental fluctuations at different time points; establishes a logical mapping between environmental fluctuations and warehouse numbers and material category identifiers; filters segments where the environment and transportation status change synchronously among multiple warehouses; verifies feature consistency through continuous time point comparisons; and generates a warehouse feature association table; and a chain association construction module, which is based on warehouse features... The correlation table synchronizes changes in segments and corresponding warehouse number combinations, retrieves transportation distance data and equipment specification data between warehouses, determines the correlation between transportation distance data and equipment specification data, and performs location matching within the corresponding segments in conjunction with the material supply cycle time series. It identifies the response sequence of warehouse nodes within continuous time periods, tracks the dynamic transmission path between warehouses based on the response sequence, verifies the correlation between changes in warehouse nodes along the path, and reassembles the verified warehouse nodes into a dynamic response logic chain in chronological order, generating a progressive warehouse correlation chain. The price and energy control module, based on the warehouse nodes corresponding to each warehouse node in the progressive warehouse correlation chain... For each material category, the procurement price sequence and quantity of incoming batches of silicon steel sheets for the stator core of pumped storage units, steel plates for the inner lining of pressure steel pipes, and concrete lining materials are obtained. The fluctuation status of the procurement price sequence for each material category is weighted and mapped to the incoming batches to construct an energy distribution map. The activity differences between adjacent areas of different materials in the energy distribution map are compared. Smoothing is performed on areas exceeding a set fluctuation threshold, while feature enhancement is performed on areas below the set threshold. Based on the processed map, the concentrated trend direction of price changes is extracted to obtain a price energy balance set. The node priority allocation module calls the nodes of each position in the price energy balance set. The system uses the energy activity characteristics and price trend information of points to obtain data on the supply intervals, usage rhythms, and importance level codes of equipment and materials. It compares the matching degree differences between the supply intervals and usage rhythms, divides the matching results into hierarchical node sets based on the importance level codes and energy activity characteristics, and counts the distribution density of each hierarchical node set in the position node sequence. It uses price trend information to guide the setting of urgency judgment thresholds, classifies the distribution areas that exceed the set thresholds into high urgency levels, and serializes each urgency level along the time axis to establish the order of price response and generate a position node priority calibration table.The comprehensive forecasting and optimization module, based on the priority sequence and timeline distribution information of the storage node priority calibration table, calls the quotation sequence and update time series from the pumped storage project procurement plan. It groups the quotation sequences according to storage node priority, calculates the rate of change of each group between adjacent time nodes based on the update time series, verifies the consistency of the rate of change trend with the temporal logic of the storage node priority calibration table, filters out abnormal segments with trend deviations, performs direction correction and smoothing repair, integrates the repaired data sequences to form an overall price balance trend, and generates an optimized equipment and material price result table.

[0021] In this embodiment, the aforementioned equipment and material price optimization result table and its historical training data samples are used to construct a neural network model with the warehouse node feature vector as the input layer, the price change feature weight as the hidden layer, and the price optimization result as the output layer. The model parameterizes the correlation paths and energy activity differences between warehouse nodes, and iteratively adjusts the network weights and thresholds through the backpropagation algorithm to minimize the error between the predicted price and the actual purchase price. At the same time, the model is continuously trained based on historical time series data, enabling it to achieve adaptive convergence and generalization updates under multi-period samples. This forms a warehouse node price prediction model with dynamic learning capabilities, providing intelligent support for subsequent procurement trend prediction and resource regulation for pumped storage projects.

[0022] The position feature association table includes environmental fluctuation characteristic parameters, position number index and material category mapping information; the position progressive association chain includes transmission path sequence, node response relationship and time sequence association elements; the price energy balance set includes energy activity distribution, price fluctuation direction and material energy balance coefficient; the position node priority calibration table includes node hierarchy division, urgency level and distribution density characteristics; and the equipment and material price optimization result table includes balance trend data, correction segment information and price optimization sequence.

[0023] like Figure 2 As shown, the warehouse feature acquisition module includes an environmental monitoring submodule, a state difference judgment submodule, and a feature mapping construction submodule. The environmental monitoring submodule acquires the transportation time and environmental temperature and humidity values ​​for the underground plant, upper reservoir, and lower reservoir. It checks the integrity of the sampled data and performs linear imputation of missing points in adjacent time periods. Based on the sampling time interval, it performs time alignment matching on the transportation time and temperature / humidity values, calculates the mean sequence and gradient of change for each warehouse in continuous time periods, and organizes the time series of each warehouse into a structured array to generate a warehouse time series matrix. The module also acquires the transportation time and environmental temperature and humidity values ​​for the underground plant, upper reservoir, and lower reservoir. Using GPS positioning units and onboard timers installed on transport vehicles, it records the start and end timestamps of vehicles in different operating areas in real time, and obtains the transportation time data for the underground plant through difference calculation. Data on transportation time to the upper reservoir And the transportation time data of the reservoir Simultaneously, industrial-grade temperature and humidity sensors (such as PT100 temperature sensors and capacitive humidity transmitters) deployed at each warehouse site synchronously read the ambient temperature values ​​at each corresponding location. With humidity value The raw data stream is collected and mapped into a one-dimensional vector sequence via a wireless transmission protocol. The index positions in the vector sequence are traversed one by one to check for missing points with empty values ​​or values ​​marked as NaN. When an index is detected... Data at the location When a data point is missing, retrieve the valid data points immediately before and after that location. and Calculate the time difference weights between the two and perform linear interpolation. This formula, based on the assumption of time continuity, is used to reconstruct the physical state at missing moments. For example, in the transportation time sequence of an underground factory, the value at 10:00 is 45 minutes, and the value at 10:10 is 48 minutes. If the data at 10:05 is missing, the imputation value is calculated to be 46.5 minutes. After completing the imputation of the entire sequence, a uniform time alignment step size is set. For a period of 15 minutes, all asynchronously sampled heterogeneous data points are mapped onto a standard time axis grid. For multiple samples falling within the same time grid, an arithmetic mean is performed to eliminate sensor random noise, resulting in an aligned standard sequence with a defined length of [value missing]. A sliding window is used to calculate the mean series of transportation time, temperature, and humidity within the window. Simultaneously, the gradient of change between adjacent time steps is calculated using the first-order difference operator. The physical meaning of this gradient lies in characterizing the instantaneous rate of change of the environment or operational efficiency, such as in calculating the temperature of a reservoir. The gradient of change at time, if The temperature at that moment was 22.5 degrees Celsius. With a time of 22.0 degrees Celsius and an interval of 15 minutes, the gradient value is 0.033. The calculated mean sequence and gradient sequence are concatenated according to the channel dimension to construct a high-dimensional feature tensor and generate a position time series matrix.

[0024] For the state difference determination submodule, based on the warehouse time series matrix, the transportation time value and temperature and humidity value of each warehouse are called to calculate the rate of change between adjacent time nodes. Based on the transportation time change threshold and temperature and humidity fluctuation threshold, the degree of state deviation at the same time node is judged. The difference between the two deviation values ​​is taken to calculate the comprehensive deviation. The time average of the comprehensive deviation is calculated according to the warehouse number, and the corresponding periodic change interval is calculated to obtain the warehouse environment difference amplitude value. For the warehouse location time series matrix, the transportation time and temperature / humidity values ​​for each warehouse location are retrieved, and sorted by time step. Extract the eigenvectors from the matrix row by row, and extract the transportation time components respectively. With temperature and humidity components Calculate the instantaneous rate of change of the current time point relative to the previous time point. This dimensionless ratio is used to eliminate the order-of-magnitude differences between different physical units (minutes and degrees Celsius) and to set a benchmark value for judging changes in transportation time. Based on the benchmark value for judging environmental temperature and humidity fluctuations The benchmark setting process involves selecting a dataset from a historical normal operating cycle and calculating its standard deviation. ,Pick As a baseline value, for example, if the standard deviation of historical transportation time is 2 minutes, then set The rate of change corresponding to 6 minutes will be used to calculate the instantaneous rate of change. The numerical values ​​are compared with the corresponding baseline values ​​to calculate the degree of state deviation. If the calculated transportation time deviation The value is 1.5, indicating the degree of temperature and humidity deviation. A value of 0.8 physically means that the deterioration in transportation efficiency exceeds the impact of environmental degradation, suggesting the existence of logistical congestion due to non-environmental factors. A subtraction operation is performed to calculate the scalar difference between the two. This yields the overall offset. For example, the overall offset for the above value is 0.7, and vice versa. In this case, the difference reflects the passive delay driven by the environment. Using the position number as the index key, it represents the overall offset of the position throughout the entire time period. Collect and calculate their arithmetic mean over the time dimension. This characterizes the asynchronicity of the long-term state of the position, and the average value is used to define the periodic variation range, for example, mapping the average value of 0.7 to the range. The difference in the position environment was obtained.

[0025] The feature mapping submodule calls the difference amplitude value of the warehouse environment and the material storage cycle value, aggregates the difference amplitude value within the time window, calculates the synchronous volatility of the difference value between adjacent warehouses, logically maps the volatility to warehouse number and material category identifier according to time nodes, filters out warehouse combination segments with a synchronous rate higher than a set threshold, establishes corresponding mapping relationships, and generates a warehouse feature association table; it also calls the difference amplitude value of the warehouse environment and the material storage cycle value to retrieve the material storage cycle value from the warehouse management system (WMS). Based on this value, the length of the time aggregation window is set. For example, for stator core silicon steel sheets, whose storage period is marked as 72 hours, the aggregation window length is set to 288 15-minute time steps. Within this window, the sequence of warehouse environment difference amplitude values ​​is extracted, segment aggregation operation is performed, and the cumulative sum or weighted average of the amplitude values ​​within the window is calculated to form an aggregation feature vector. This vector represents the overall environmental stress over a complete storage cycle, selecting two adjacent storage location numbers. and Retrieve their corresponding aggregated feature vectors respectively and The synchronous volatility calculation between vectors is performed using the cosine similarity formula. Measuring the consistency of the changing trends of the two, such as vectors for ,vector for The similarity score was calculated to be 0.99, and a simultaneous screening threshold was set. The threshold is 0.90. This threshold is obtained by training and fitting historical data of known strongly correlated position pairs. The calculated synchronous volatility is compared with this threshold. When the calculated result (0.99) is greater than the threshold (0.90), the position combination is determined to have a strong coupling relationship within the current time window. Combinations below the threshold are determined to be weakly correlated or uncorrelated and are not recorded. The start and end times and position number combinations of this segment are extracted. The relevant material category identifiers are encapsulated into associated tuples. All tuples that meet the conditions are written into the structured database in chronological order to generate a warehouse feature association table.

[0026] like Figure 3 As shown, the chain-linked construction module includes a segment matching submodule, a node response identification submodule, and a dynamic path reorganization submodule. The segment matching submodule retrieves synchronously changing segments and corresponding warehouse number combinations from the warehouse feature association table, collects transportation distance data and equipment specification data between warehouses, performs numerical matching on transportation distances and equipment specifications for combinations with the same number, calculates the ratio of the transportation distance difference to the equipment specification difference, and compares the ratio with a preset synchronization judgment benchmark. Based on the comparison results, it filters warehouse combination segments that meet the conditions and generates a warehouse matching association coefficient set. The module also retrieves synchronously changing segments and corresponding warehouse number combinations from the warehouse feature association table. For identified warehouse pairs with synchronized environmental characteristics, such as the combination "Shangshui Reservoir Yard A Area - Underground Plant B2 Warehouse," it extracts the actual road network transportation distance between the two points by calling the API interface of the GIS geographic information system. For example, for a distance of 5.2 kilometers, the specifications of the respective transportation equipment can be retrieved from the equipment asset management ledger database. This parameter is defined as the product of the equipment's maximum load capacity and average travel speed, and has the physical meaning of "momentum flow rate." For example, the specification value of equipment in area A is 20 tons × 40 km / h = 800, and the specification value of equipment in warehouse B2 is 15 tons × 30 km / h = 450. A numerical matching operation is performed, pairing the distance value with the specification value into a pair of tuples, and then using the difference ratio formula... Calculate the ratio, where and These refer to the access distances from the two warehouses to the nearest main road. Assuming the access distance for warehouse A is 0.5km and for warehouse B2 it is 0.1km, the numerator is 0.4, and the denominator is |800-450|=350, thus the ratio is calculated. ≈0.00114, its physical dimension is "km / (ton·km / hour) = hour / ton", characterizing the distance impedance under unit transport capacity difference, and setting a synchronization judgment benchmark value. This benchmark value was derived by analyzing the distance-to-specification ratio distribution in historical high-efficiency logistics cases, taking the median of the distribution, 0.0012, as the benchmark, and then calculating the... and Perform a numerical comparison and calculate the relative deviation between the two. Set the screening tolerance range to ±15%. If the ratio is less than 0.15, it is considered to be within a reasonable range, indicating that the warehouse combination has physical compatibility in terms of logistics logic. Therefore, the warehouse combination and its calculated ratio are retained. The ratio is then used as a weighting coefficient to assign to the combination, generating a set of position matching correlation coefficients.

[0027] For the node response identification submodule, the warehouse location matching correlation coefficient set is invoked to collect the material supply cycle time series. Time series position matching is performed on each warehouse location combination within each segment. The delay interval between the material supply point and the warehouse response point at continuous time nodes is calculated. The direction of change of the delay interval is combined with the warehouse number for sorting judgment. Based on the sorting relationship, a continuous interval response sequence is generated to obtain the warehouse node response order value. The warehouse location matching correlation coefficient set is also invoked to collect the material supply cycle time series. The planned supply table is exported through the ERP supply chain module. For each selected warehouse location combination segment, the key time node sequence in the material supply cycle is extracted. For example, the expected supply node for a certain batch of steel plates is (Unit: hours), and simultaneously, the status response time sequence of the corresponding warehouse within this segment is extracted by scanning and recording using on-site PDA handheld terminals. Perform time series position matching for each ,exist Find the closest one Calculate the time difference between the two. Construct a set of delay intervals, for example, for supply points. Matching response point The delay will be +2 hours. correspond The delay is +1 hour. Calculate the direction of change within the delay interval, i.e., calculate the difference between adjacent delay values. ,like This indicates that response latency is decreasing and efficiency is improving, and vice versa. The sign and absolute value of the position numbers are used to sort the position number combinations, prioritizing those with shorter delays and smaller absolute values. For example, if the delay sequence of combination A is... Combination B is If A is determined to be superior to B, the sorted warehouse nodes are connected in order to form an index list representing the response priority, thus obtaining the warehouse node response order value.

[0028] For the dynamic path reorganization submodule, based on the response order value of the warehouse node, it calls the combination of warehouse numbers within the corresponding segment, calculates the time interval difference between the node response orders, performs a correlation judgment on adjacent interval values, filters node pairs with a time difference less than the judgment threshold, arranges continuous node chains in ascending order of time, records the sequence structure of continuous node chains as associated paths, and generates a progressive association chain of warehouses; based on the response order value of the warehouse node, it calls the combination of warehouse numbers within the corresponding segment, and extracts two adjacent nodes from the sorted node sequence. and Response time and Perform subtraction to calculate the time interval difference For example, if node A's response time is 12.5 and node B's response time is 13.2, then... Hours, set a threshold for determining time correlation. This threshold is set based on the minimum physical flow time of materials between processes, for example, a range of 0.5 hours to 2.0 hours, to determine the calculated... Does it fall within the range? inside, if If the correlation coefficient is less than 0.5, it is considered data overlap or anomaly; if it is greater than 2.0, it is considered a broken link. Only all node pairs that meet the criteria are selected. Following the logic of increasing time, these scattered nodes are connected end-to-end. For example, if there exists a pair... , and If all conditions are met, the chain is reorganized. Construct a directed graph structure, traverse the graph to find the longest path, lock the sequence of nodes on the longest path as the determined dynamic transmission path, and then establish the relationships between the nodes. The values ​​are stored as weight attributes of the path edges to generate a progressive association chain of positions.

[0029] like Figure 4 As shown, the price and energy control module includes a price data mapping submodule, an energy distribution calculation submodule, and a price trend extraction submodule. The price data mapping submodule obtains the purchase price sequence and batch quantity of each material category corresponding to each warehouse node in the warehouse progression chain. It synchronously detects the purchase price sequences of silicon steel sheets, steel plates, and concrete lining materials according to time nodes, calculates the fluctuation range of each material's price sequence, and performs a weighted product operation on the price fluctuation range based on the batch quantity as a weight. The summation and time average are then used as the price fluctuation level for each material category, generating a price fluctuation level value. The module also obtains the purchase price sequence and batch quantity of each material category corresponding to each warehouse node in the warehouse progression chain. For the three key materials—stator core silicon steel sheets, pressure steel pipe inner lining steel plates, and concrete lining materials—it establishes independent data index tables from the electronic procurement and trading platform's database, recording the unit price at the time of each procurement activity. and the corresponding quantity of goods entering the site for this batch. The procurement price series of the three types of materials undergoes synchronous time node detection. A baseline time axis is selected, for example, with 12:00 PM daily as the scale. The system checks whether price data for each material exists on the same date. If a material has no procurement record on a specific date, the price data from the previous valid date is used to fill the gap, ensuring the continuity and alignment of the three price curves over time. The fluctuation range of the price series for each material is calculated, and the formula is applied. For example, if the price of silicon steel sheets rises from 12,000 yuan / ton to 12,600 yuan / ton on a certain day, the fluctuation range is 0.05, and the weighting coefficient is calculated based on the quantity of batches entering the market. Assuming 500 tons of silicon steel sheets arrive on that day, and the total arrivals for the month are 5000 tons, then this batch has a weight of 0.1. A weighted product operation is then performed on the price fluctuation range value, i.e., the calculation... For example, 0.05 0.1 = 0.005. The calculated single-point weighted volatility values ​​are summed over the selected assessment period (e.g., one month); the formula is... Finally, take the average time value, that is... If the cumulative value within a month is 0.15, covering 30 time points, the average value is 0.005. This value quantifies the comprehensive price risk of this type of material during this period and generates a price fluctuation level value.

[0030] For the energy distribution calculation submodule, based on price fluctuation level values, the data difference between adjacent regions of each material is calculated. The difference results are mapped into a two-dimensional coordinate grid in chronological order. The regional activity is calculated based on the root mean square value of fluctuation of adjacent points in the grid. Difference smoothing is performed on regions with an activity level above the activity level threshold, and amplitude amplification is performed on regions with an activity level below the activity level threshold. The processed activity matrix is ​​integrated into energy spatial distribution information to obtain a set of energy activity coefficients. Based on price fluctuation level values, a two-dimensional coordinate grid system is constructed, where the horizontal axis represents discretized time nodes. The vertical axis represents different material categories. (like Represents silicon steel sheets, (Representing steel plates), the price fluctuation levels of each material at each point in time calculated above. Fill in the corresponding grid coordinates and calculate the data difference between adjacent areas of each material, that is, calculate the adjacent points in the grid. and and and Difference between This is used to construct a difference matrix, define a local window (such as a 3x3 grid), and calculate the regional activity within the window based on the root mean square value of fluctuations of adjacent points in the grid, applying the formula... ,in The formula represents the number of adjacent pairs within a window. Physically analogous to energy calculations, this formula reflects the degree of price volatility across time and product categories, for example, calculating the activity level of a specific local area. Set the activity threshold to 0.08. The threshold is set at 0.05, based on the average activity level during historical periods of stability. and If a comparison is made, If a region is identified as a high-activity region, a difference smoothing operation is performed on the differences within this region, and the node values ​​are updated using the Laplace smoothing algorithm. To suppress excessive short-term fluctuation noise; if the calculated activity level is 0.02, which is lower than the threshold of 0.05, then an amplitude amplification operation is performed; a nonlinear gain function is applied. ,in The gain coefficient (e.g., set to 0.5) is used to enhance the saliency of weak signals. After completing the full grid traversal process, the corrected activity matrix is ​​obtained, and the energy activity coefficient set is generated.

[0031] For the price trend extraction submodule, the energy activity coefficient set is invoked. A direction determination operation is performed on the activity values ​​arranged along the time axis in the matrix, calculating the ratio of activity changes between consecutive time nodes. The time period with consecutive signs of the ratio is identified, and positive and negative sequences are distinguished and their frequency proportions are calculated. The dominant trend of price energy change is determined based on the direction with the largest proportion, generating a price energy balance set. The energy activity coefficient set, which is represented as a state matrix evolving over time, is then invoked to extract the activity value sequence arranged along the time axis for each row (representing a type of material). Perform direction determination operations on the sequence and calculate the ratio of activity changes between consecutive time nodes. For example, time Its activity level is 0.04, at time... If the value is 0.044, then the change ratio is +0.1 at time [time value missing]. The value is 0.042, and the change ratio is -0.045. After confirming the time period with consecutive positive (increasing activity) or consecutive negative (decreasing activity) subsequence segments in the scanned sequence, for example, identifying the segment... The changes during this period were all positive, and this region was marked as an "energy accumulation zone". All values ​​are negative, marked as the "energy dissipation region". The frequency ratios of the forward and reverse sequences within the total observation period are calculated. and If , If the current price energy is in a dominant cycle of accumulation and rise, the dominant trend of price energy change is determined based on the direction with the largest proportion (positive in this case). If the difference between the two proportions is less than the set balance threshold (e.g., 0.05), it is determined to be a period of oscillation, and no dominant trend is specified. If the trend is determined, the trend direction (+1 or -1) is combined with the dominant proportion value (0.6) to construct a trend vector and generate a price energy balance set.

[0032] like Figure 5 As shown, the node priority allocation module includes a matching degree determination submodule, a hierarchical division calculation submodule, and a urgency ranking submodule. The matching degree determination submodule acquires the energy activity characteristics and price change trend information of each position node in the price energy balance set, collects data on equipment and material construction supply intervals and usage rhythms, calculates the difference between the supply interval and usage rhythm cycle lengths, compares the supply cycle difference ratio with the energy activity value, determines the matching degree deviation range, generates a deviation degree index for each node, establishes a matching deviation result sequence, and obtains the node matching deviation mapping. The specific process of calculating the difference between the supply interval and usage rhythm cycle lengths is as follows: Based on the actual time lengths of adjacent nodes in the equipment and material construction supply interval time series and the usage rhythm time series, the time values ​​of the supply interval period and the usage rhythm period are calculated respectively. The difference between the two is taken, and the supply cycle difference ratio is calculated with the supply interval period as the denominator. The specific calculation process based on comparing the supply cycle difference ratio with the energy activity value is as follows: The energy activity value corresponding to each position node is compared with the difference between its supply cycle and the value of the supply cycle. If the difference exceeds the set supply matching benchmark threshold, the node is determined to have a matching deviation, and the degree of deviation of the node is recorded. The energy activity characteristics of each position node in the price energy balance set are obtained. Information on price change trends Collect the time series of equipment and material supply intervals from the construction planning management system. and the use of rhythm time series For each corresponding node pair in the sequence, such as node The delivery interval record is as follows (e.g., 5 days / time), using rhythm recording as (e.g., 3 days / batch) Calculate the time values ​​for the supply interval cycle and the usage cycle respectively, and perform subtraction. To obtain the difference (e.g.) (days), with the delivery interval period as the denominator to calculate the delivery cycle difference ratio: Calculated This ratio reflects the lag between supply and demand, and represents the energy activity value corresponding to each warehouse node. (Assuming a value of 0.75) and the difference in delivery cycle calculated at this node Perform a numerical comparison and calculate the absolute value of the difference between the two. (For example Set supply matching benchmark thresholds. This threshold is set based on the supply chain resilience coefficient, for example, a value of 0.25. The system checks whether the calculated difference of 0.35 exceeds the threshold of 0.25. If it does, the node is considered to have a mismatch; and the degree of deviation for that node is recorded. If it does not exceed Recorded as 0, all node deviation indicators are associated and stored according to position index ID, establishing a system containing... The matching deviation sequence of tuples is used to obtain the node matching deviation mapping.

[0033] For the hierarchical division calculation submodule, based on the node matching deviation mapping, the importance level code of each node is called. Cross-judgment is performed by the importance level code and activity features. Nodes with deviation values ​​less than a set judgment threshold are classified into a high-matching layer, and nodes with deviation values ​​greater than the threshold are classified into a low-matching layer. The frequency of occurrence of nodes in each layer in the position sequence is calculated and the average distribution density is obtained to establish the hierarchical node classification results, resulting in a hierarchical node distribution map. The specific process of classifying nodes with deviation values ​​less than the set judgment threshold into the high-matching layer is as follows: Among all the calculated node deviation indices, the set of nodes with deviations less than the matching threshold is selected and arranged in ascending order according to the deviation index values ​​to form a high-matching layer node sequence list; the specific process of calculating the frequency of occurrence of each layer of nodes in the position sequence and calculating the average distribution density is as follows: The ratio of the number of nodes in the high-matching and low-matching layers appearing in the position sequence to the total number of nodes is calculated to obtain the distribution ratio of each layer. The average value of the product of the distance between adjacent nodes and the node density is calculated. Based on the node matching deviation mapping, the importance level code of each node is called. (e.g., first-level node) Level 2 Set a deviation value judgment threshold. (For example, set to 0.3), traverse all nodes, and perform cross-validation based on importance level code and activity feature. The nodes are assigned to the high-matching layer set. and based on The values ​​are sorted in ascending order from smallest to largest to form a sequence list of high-matching layer nodes. The nodes are assigned to the low-matching layer set. Count the number of nodes in the high matching layer. With low matching layer node count and the total number of nodes Calculate the distribution ratio of each layer. and ,For example ,but Calculate the average of the product of the distance between adjacent nodes and the node density, and calculate the number of adjacent nodes in the high-matching layer. and Spacing on the time axis Calculate the local density of this layer. Calculate the arithmetic mean of all local densities. This characterizes the degree of clustering of high-matching layer nodes over time, and the average density value is compared with the distribution ratio. By combining these elements, structured data containing hierarchical identifiers, distribution ratios, and average densities is constructed to obtain a hierarchical node distribution map.

[0034] For the urgency ranking submodule, based on the hierarchical node distribution map, price change trend information is invoked to divide the distribution density of nodes at each level into intervals. The activity weighted difference between adjacent intervals is calculated, and the weighted difference is compared with the urgency judgment threshold. Intervals exceeding the threshold are identified as high urgency levels. All urgency level intervals are arranged in chronological order to form the response sequence of position nodes, generating a position node priority labeling table. The specific process of calculating the activity weighted difference between adjacent intervals is as follows: Based on the average activity values ​​of adjacent intervals in the hierarchical node distribution map, the difference is calculated and multiplied by a weighting coefficient for the time span of each interval to generate an activity-weighted difference result. This result is then compared with the urgency judgment threshold to identify high-urgency intervals for forming the response sequence of position nodes. Based on the hierarchical node distribution map, price change trend information is retrieved. (e.g., if the trend is upward +1), the distribution density of nodes at each level is divided into intervals, dividing the time axis into several consecutive evaluation intervals. Calculate the average activity value of high-matching layer nodes within each interval. Select adjacent intervals and Calculate the difference in their average activity levels. Obtain the time span of the two intervals. and Calculate the weighting coefficients This weighting factor is used to balance the impact of different interval lengths on the total activity level, and a multiplication operation is performed to generate the weighted difference of activity levels. , introduced here This is to adjust the positive or negative attribute of the weighted difference based on the direction of the price trend. For example, if the trend is upward (+1) and the activity increases (... If the weighted difference is significantly positive, then a threshold for determining urgency is set. (For example, set to 0.5), and If a comparison is made, Then the interval is selected. Marked as high urgency, this indicates a surge in activity in the area due to an upward price trend, requiring immediate attention. The nodes are marked as medium-level and the rest as low-level. Finally, all intervals marked with urgency levels are arranged in chronological order. Nodes in high-urgency intervals are assigned the highest priority, and medium and low-level nodes are assigned sequentially to form the response sequence of position nodes, thus generating a position node priority calibration table.

[0035] like Figure 6As shown, the comprehensive forecasting and optimization module includes a sequence grouping calculation submodule, a trend deviation correction submodule, and a price balance integration submodule. The sequence grouping calculation submodule obtains the node priority sequence and time axis distribution from the position node priority calibration table, calls the quotation sequence and update time sequence of the pumped storage project procurement plan, groups and arranges the quotations according to node priority, calculates the change amplitude of quotations in the same group between adjacent time nodes, calculates the change rate in combination with the update time interval, and organizes the rate trend in chronological order to generate a quotation grouping time series graph. It also obtains the node priority sequence from the position node priority calibration table. and time axis distribution Calling the quotation sequence of the pumped storage project procurement plan With updated time series For each priority group (e.g., high urgency group) The entire sequence of quotes belonging to this group is sorted in ascending order by timestamp to form a subsequence. Calculate the bids in the same group at adjacent time points. and The range of variation ,For example , ,but Calculate the rate of change by combining the update time interval. If the time interval is 2 days, then Yuan / day, traversing all groups and all adjacent nodes within a group, a series of rate values ​​are calculated, and the rate trends are arranged in chronological order. The rate calculated at each moment is then... The values ​​are mapped to a two-dimensional coordinate system (X-axis for time, Y-axis for rate), and the points are connected to form a continuous polyline. For cases where there are multiple rate groups at the same time point, different colors or line types are used to distinguish and mark them, thereby constructing a multi-dimensional view that reflects the dynamic changes in the prices of materials with different priorities and generating a time series graph of price groups.

[0036] For the trend deviation correction submodule, based on the price group time series graph, it calls the time series logic in the position node priority calibration table to compare the consistency of the change rate direction between groups, calculate the deviation angle between the rate direction and the logic direction, filter node segments with deviation angles exceeding a set threshold, adjust the direction of abnormal segments, and continuously smooth the difference between adjacent rates. The corrected node sequence is then processed to obtain the price trend correction structure. Based on the price group time series graph, it calls the time series logic vector in the position node priority calibration table. (For example, it indicates that the price should rise gradually as the project progresses, with the direction vector being...) The consistency of the rate of change direction among groups is compared, the deviation angle between the rate direction and the logical direction is calculated, and the rate trend of a certain segment in the time series spectrum is fitted into a vector. Apply the formula for the angle between vectors: Calculate the angle, for example, calculate the angle. Set the deviation angle threshold (For example ), filter node segments whose deviation angle exceeds the set threshold, if The section is determined to be an abnormal section, and its direction is adjusted using a rotation matrix. Will Forced calibration to Within the boundary, or directly using the logical direction vector. The projection replaces the outlier vector, and the difference between adjacent rates is continuously and smoothly adjusted. Cubic spline interpolation or moving average (window size 3) is used to process the data of the correction point and its immediate vicinity, eliminating abrupt breakpoints caused by forced adjustment. For example, the corrected rate points... and Perform weighted average The repaired node sequence is organized, and the corrected rate value is back-introduced into the price sequence to obtain the price trend repair structure.

[0037] For the price balance integration submodule, based on the price trend repair structure, node priority information is invoked to rearrange the repaired sequence in chronological order, calculate the balance ratio of price changes between adjacent nodes, extract and aggregate time segments of continuous balance distribution, organize the aggregation results to form the overall price evolution pattern, and generate an optimized table of equipment and material prices. Based on the price trend repair structure, node priority information is invoked to rearrange the repaired sequence in chronological order to ensure that all data points return to a unified time axis, and calculate the balance ratio of price changes between adjacent nodes. ,in For example, the price value after repair The price for a single time slot is 5200. If the time is 5252, then Extract and aggregate time intervals of continuous equilibrium distributions, and set the equilibrium tolerance range. Scan the entire sequence and identify consecutive sequences that satisfy... Subsequences falling within this range (e.g., a ratio of 1.01 for 5 consecutive days) are aggregated into a stable price evolution range. The average growth rate or mean within this range is calculated as a representative feature. For non-equilibrium segments, the original fluctuation characteristics are retained. The aggregation results are sorted to form an overall price evolution pattern. The start and end times, representative prices, and trend characteristics of each segment are filled into a structured table to generate an optimized table of equipment and material prices.

[0038] By calling the equipment and material price optimization result table and its previous training data samples, a three-layer backpropagation neural network model is constructed, and the feature vectors of the corresponding warehouse nodes are defined for the input layer nodes. This includes the correlation path weights between positions and the energy activity difference values ​​calculated in the preceding sequence. Hidden layer nodes are configured to handle the nonlinear mapping of price change characteristics, and each connection is assigned an initial weight. and threshold The output layer nodes are set to directly correspond to the final optimized price prediction value. During the forward propagation phase, the input signal is passed layer by layer through activation functions (such as Sigmoid or ReLU) to calculate the output result. Then calculate the predicted output. Compared with the actual purchase price Mean square error between Then, in the backpropagation phase, the partial derivatives of the error with respect to the weights and thresholds of each layer are calculated using the gradient descent method, for example, to update the weights. (in (where the learning rate is used) iteratively adjusts the network parameters in the opposite direction of the gradient until the error... Converging to a preset minimum precision range (e.g.) Meanwhile, historical time-series data is introduced as a validation set to continuously train the model. The hyperparameters are optimized using cross-validation, enabling the model to achieve adaptive convergence and generalization updates under multi-period samples. This forms a position node price prediction model with dynamic learning capabilities, providing intelligent support for subsequent procurement trend prediction and resource regulation of pumped storage projects.

[0039] In summary, the present invention has been described in detail with reference to specific embodiments. However, those skilled in the art will understand that various changes or substitutions can be made, as long as they do not depart from the spirit and purpose of the present invention. All such changes or substitutions should be covered within the scope of protection of the present invention, which is defined by the appended claims.

Claims

1. A machine learning-based equipment and material price control system, characterized in that, The system includes: The warehouse feature acquisition module obtains data on transportation time, environmental temperature and humidity, and storage period of the factory and upper and lower reservoirs. It compares the differences between transportation and environment, determines the environmental trend based on the storage period, records fluctuations and establishes a number mapping, filters synchronous segments, and generates a warehouse feature association table. The chain association construction module, based on the synchronous segment and warehouse number in the warehouse feature association table, calls the transportation distance and equipment specifications to determine the association, combines the material supply cycle to match the location, identifies the response order of warehouse nodes and tracks the transmission path, and generates a warehouse progressive association chain. The price and energy control module obtains the purchase price sequence and entry batch of each material according to the material category in the progressive association chain of the warehouse, constructs an energy distribution map by weighted mapping, compares the activity differences of different materials, performs smoothing and enhancement, and obtains the price and energy balance set. The node priority allocation module calls upon the price energy balance concentration energy characteristics and price trends to obtain equipment and material supply intervals, usage rhythms, and importance level codes. It compares and matches differences to divide the hierarchical node sets, calculates the distribution density, sets an urgency threshold, defines the levels, and generates a warehouse node priority calibration table.

2. The equipment and material price prediction and optimization system based on machine learning according to claim 1, characterized in that: The position feature association table includes environmental fluctuation characteristic parameters, position number index and material category mapping information; the position progressive association chain includes transmission path sequence, node response relationship and time sequence association elements; the price energy balance set includes energy activity distribution, price fluctuation direction and material energy balance coefficient; and the position node priority calibration table includes node hierarchy division, urgency level and distribution density characteristics.

3. The equipment and material price prediction and optimization system based on machine learning according to claim 1, characterized in that: The warehouse feature acquisition module includes an environmental monitoring submodule, a state difference determination submodule, and a feature mapping construction submodule; The environmental monitoring submodule acquires the transportation time and environmental temperature and humidity values ​​of the underground plant, upper reservoir and lower reservoir, detects the integrity of the sampled data and performs linear filling of adjacent time periods for missing points, performs time alignment matching of transportation time and temperature and humidity values ​​according to the sampling time interval, calculates the mean sequence and change gradient of each warehouse in a continuous time period, organizes the time series of each warehouse into a structured array, and generates a warehouse time series matrix. The state difference determination submodule, based on the warehouse time series matrix, calls the transportation time value and temperature and humidity value of each warehouse, calculates the rate of change between adjacent time nodes, judges the degree of state deviation at the same time node according to the transportation time change threshold and temperature and humidity fluctuation threshold, takes the difference between the two deviation values ​​to calculate the comprehensive deviation, calculates the time average of the comprehensive deviation according to the warehouse number, calculates the corresponding periodic change interval, and obtains the warehouse environment difference amplitude value. The feature mapping construction submodule calls the difference amplitude value of the warehouse environment and the material storage cycle value, aggregates the difference amplitude value within the time window, calculates the synchronous volatility of the difference value between adjacent warehouses, logically maps the volatility to the warehouse number and material category identifier according to the time node, filters out the warehouse combination segment with the synchronous rate higher than the set threshold, establishes the corresponding mapping relationship, and generates a warehouse feature association table.

4. The equipment and material price prediction and optimization system based on machine learning according to claim 1, characterized in that: The chain-linked construction module includes a segment matching submodule, a node response recognition submodule, and a dynamic path reorganization submodule; The segment matching submodule obtains the synchronously changing segments and corresponding storage location number combinations from the storage location feature association table, collects transportation distance data and equipment specification data between each storage location, performs numerical correspondence matching on the transportation distance and equipment specification of the same number combination, calculates the ratio of the transportation distance difference to the equipment specification difference, compares the ratio with the preset synchronous judgment benchmark value, filters storage location combination segments that meet the conditions based on the comparison results, and generates a storage location matching association coefficient set. The node response identification submodule calls the warehouse matching correlation coefficient set, collects the material supply cycle time series, performs time series position matching on the warehouse combination of each segment, calculates the delay interval between the material supply point and the warehouse response point under continuous time nodes, performs sorting judgment on the change direction of the delay interval and the warehouse number combination, generates a continuous interval response sequence based on the sorting relationship, and obtains the warehouse node response order value. The dynamic path reorganization submodule, based on the response sequence value of the warehouse node, calls the combination of warehouse number in the corresponding segment, calculates the time interval difference between the response sequences of the nodes, performs a correlation judgment on adjacent interval values, filters node pairs with a time difference less than the judgment threshold, arranges the continuous node chain in ascending order of time, records the sequence structure of the continuous node chain as the associated path, and generates a progressive association chain of warehouses.

5. The equipment and material price prediction and optimization system based on machine learning according to claim 1, characterized in that: The price energy control module includes a price data mapping submodule, an energy distribution calculation submodule, and a price trend extraction submodule; The price data mapping submodule obtains the purchase price sequence and the number of batches of materials corresponding to each warehouse node in the warehouse progressive association chain. It synchronously detects the purchase price sequences of silicon steel sheets, steel plates and concrete lining materials according to time nodes, calculates the fluctuation range value of the price sequence of each material, performs a weighted product operation on the price fluctuation range value based on the number of batches of materials entering the warehouse as the weight, and takes the time average after summing as the price fluctuation level of each material to generate a price fluctuation level value. The energy distribution calculation submodule calculates the data difference between adjacent regions of each material based on the price fluctuation level value, maps the difference results into a two-dimensional coordinate grid in time order, calculates the regional activity based on the root mean square value of fluctuation of adjacent points in the grid, performs difference smoothing on regions with an activity level higher than the activity level threshold, performs amplitude amplification on regions with an activity level lower than the activity level threshold, and integrates the processed activity matrix into energy spatial distribution information to obtain a set of energy activity coefficients. The price trend extraction submodule calls the energy activity coefficient set, performs direction judgment operation on the activity values ​​arranged along the time axis in the matrix, calculates the activity change ratio between consecutive time nodes, confirms the time period with consecutive change ratio signs, distinguishes between positive and negative sequences and counts the frequency proportion, determines the dominant trend of price energy change based on the direction with the largest proportion, and generates a price energy balance set.

6. The equipment and material price prediction and optimization system based on machine learning according to claim 1, characterized in that: The node priority allocation module includes a matching degree determination submodule, a hierarchy division calculation submodule, and an urgency ranking submodule; The matching degree determination submodule acquires the energy activity characteristics and price change trend information of each position node in the price energy balance set, collects equipment and material construction supply interval and usage rhythm data, performs difference calculation on the cycle length of the supply interval and usage rhythm, compares the supply cycle difference ratio with the energy activity value, determines the matching degree deviation interval, generates deviation degree index for each node, establishes a matching deviation result sequence, and obtains the node matching deviation mapping. The hierarchical division calculation submodule, based on the node matching deviation mapping, calls the importance level code of each node, performs cross-judgment by the importance level code and activity feature, filters out nodes with deviation values ​​less than the set judgment threshold and classifies them into the high matching layer, and nodes with deviation values ​​greater than the threshold are classified into the low matching layer, calculates the occurrence frequency of each layer of nodes in the position sequence and calculates the average distribution density, establishes the hierarchical node classification result, and obtains the hierarchical node distribution map; The urgency ranking submodule, based on the hierarchical node distribution map, calls price change trend information to divide the distribution density of each level of nodes into intervals, calculates the activity weighted difference between adjacent intervals, compares the weighted difference with the urgency judgment threshold, filters intervals exceeding the threshold as high urgency level, arranges all urgency level intervals in time axis order to form the response sequence of position nodes, and generates a position node priority calibration table.

7. The equipment and material price prediction and optimization system based on machine learning according to claim 6, characterized in that: The process of calculating the difference between the supply interval and the cycle length of the usage rhythm is as follows: Based on the actual time length of adjacent nodes in the equipment and material construction supply interval time series and the usage rhythm time series, calculate the time values ​​of the supply interval cycle and the usage rhythm cycle respectively, take the difference between the two and calculate the supply cycle difference ratio with the supply interval cycle as the denominator. The specific process of calculating based on the supply cycle difference ratio and the energy activity value is as follows: The energy activity value corresponding to each warehouse node is compared with the difference between its supply cycle. If the difference exceeds the set supply matching benchmark threshold, it is determined that there is a matching deviation at the node, and the degree of deviation of the node is recorded.

8. The equipment and material price prediction and optimization system based on machine learning according to claim 6, characterized in that: The process of classifying nodes whose deviation value is less than the set judgment threshold into the high matching layer is as follows: Among all the calculated node deviation indices, select the set of nodes that are less than the matching degree judgment threshold, and arrange them in ascending order according to the value of the deviation index to form a high matching layer node sequence list. The process of calculating the frequency of occurrence of each layer of nodes in the warehouse sequence and determining the average distribution density is as follows: The ratio of the number of nodes in the high-matching layer and low-matching layer appearing in the position sequence to the total number of nodes is calculated. The distribution ratio of each layer is obtained, and the average value of the product of the distance between adjacent nodes and the node density is calculated. The process of calculating the activity-weighted difference between adjacent intervals is as follows: Based on the average activity value of adjacent intervals in the hierarchical node distribution map, the difference is calculated and multiplied by the weighting coefficient of the time span of each interval to generate the activity weighted difference result. This result is then compared with the urgency determination threshold to identify high urgency level intervals for forming the response sequence of position nodes.

9. The equipment and material price prediction and optimization system based on machine learning according to claim 1, characterized in that: The system also includes: The comprehensive forecasting and optimization module, based on the priority sequence and time axis distribution information of the warehouse node priority calibration table, calls the purchase quotation and update time series, groups and arranges them according to priority and calculates the rate of change, verifies the consistency of the trend, filters out abnormal segments for correction and repair, integrates them to form an overall price balance trend, and generates an equipment and material price optimization result table. The equipment and material price optimization result table includes balance trend data, correction segment information, and price optimization sequence.

10. The equipment and material price prediction and optimization system based on machine learning according to claim 9, characterized in that: The comprehensive forecasting and optimization module includes a sequence grouping calculation submodule, a trend deviation correction submodule, and a price balance integration submodule; The sequence grouping calculation submodule obtains the node priority sequence and time axis distribution in the warehouse node priority calibration table, calls the quotation sequence and update time sequence of the pumped storage project procurement plan, groups and arranges the quotations according to the node priority, calculates the change amplitude of the quotations in the same group between adjacent time nodes, calculates the change rate in combination with the update time interval, and organizes the rate trend in time order to generate a quotation grouping time series graph. The trend deviation correction submodule, based on the price group time series graph, calls the time series logic in the position node priority calibration table, performs a consistency comparison of the change rate direction between groups, calculates the deviation angle between the rate direction and the logic direction, filters the node segments whose deviation angle exceeds the set threshold, adjusts the direction of abnormal segments and continuously smooths the adjacent rate difference, organizes and repairs the node sequence, and obtains the price trend repair structure. The price balance integration submodule, based on the price trend repair structure, calls the node priority information, rearranges the repaired sequence in chronological order, calculates the balance ratio of price changes between adjacent nodes, extracts and aggregates the time segments of continuous balance distribution, organizes the aggregation results to form the overall price evolution pattern, and generates an optimized table of equipment and material prices.

Images