Power grid material purchasing demand generation system based on multi-source data fusion and intelligent prediction

Abstract

The invention belongs to the technical field of intelligent analysis, and discloses a power grid material purchasing demand generation system based on multi-source data fusion and intelligent prediction. The data sensing module is used for collecting equipment state data, environment market data and planning data; analyzing to obtain annotation data and a data quality score; the intelligent decision-making module is used for constructing a material association map, performing analysis in combination with the digital demand cell body set, and obtaining a credible demand pool formed by a credible demand cell body set; the intelligent decision-making module is used for constructing a material association map, performing analysis in combination with the digital demand cell body set, and obtaining a credible demand pool formed by a credible demand cell body set; the demand generation module is used for analyzing and obtaining an optimal power grid material purchasing scheme according to the credible demand pool and the annotation data; according to the method, the purchasing accuracy, the supply chain stability and the system self-evolution capability are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent analysis, more particularly, the present application relates to a power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction. BACKGROUND

[0002] In the current field of power grid material procurement demand prediction, methods based on multi-source data fusion and edge intelligence are gradually becoming mainstream. When the system adopts a distributed edge computing architecture, each edge node independently generates a procurement demand proposal based on its local perspective and data, which often produces irreconcilable contradictions due to the limitations of data perspective, model bias, or local event interference, i.e., "edge computing node consensus split". For example, a node responsible for power supply in a new urban area may strongly request the procurement of capacity expansion transformers based on load growth prediction, while a node covering an old urban area proposes to reduce the procurement of the same material due to the observation of some user migration. This split, which is not due to errors but due to different local optimal judgments, makes it difficult for the central system to simply form a globally consistent and logically self-consistent procurement plan through weighted averaging or voting. Traditional solutions either rely on the authority of the central node for forced arbitration, sacrificing the agility and accuracy advantages of edge computing, or fall into endless iterative coordination, greatly slowing down the decision-making speed and leading to procurement response delays, which cannot meet the actual needs of dynamic changes in the power grid. This dilemma reveals that the existing system lacks an effective mechanism to understand the internal logic of demand and achieve autonomous negotiation and consensus formation in the process of upgrading from "local intelligence" to "collective wisdom".

[0003] In view of this, the present application proposes a power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction to solve the above problems. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction, comprising: a data perception module: collecting equipment state data, environmental market data, and planning data; analyzing to obtain labeled data and data quality scores; an intelligent decision-making module: constructing a material correlation graph, analyzing in combination with a digital demand cell set, and obtaining a trusted demand pool composed of a trusted demand cell set; an intelligent decision-making module: constructing a material correlation graph, analyzing in combination with a digital demand cell set, and obtaining a trusted demand pool composed of a trusted demand cell set; a demand generation module: analyzing to obtain an optimal procurement scheme for power grid materials according to the trusted demand pool and the labeled data.

[0005] Further, the method for obtaining the labeled data comprises: predefining a set of core semantic concepts of the power grid, and attributes and relationships of the core semantic concepts; analyzing and standardizing the collected device state data, environmental market data and planning data based on NLP technology, and labeling according to preset semantic labeling rules to obtain labeled data about the core semantic concepts.

[0006] Further, the method for obtaining the data quality score comprises: calculating a ratio of the actual number of labeled data obtained to the number of data that should be obtained to obtain a data integrity indicator; calculating a ratio of the number of labeled data that meets the semantic labeling rules to the total number of labeled data to obtain a data consistency indicator; calculating a ratio of the number of labeled data whose difference between the generation time of the original data corresponding to the labeled data and the system receiving time is not higher than a difference threshold to the total number of labeled data to obtain a data timeliness indicator; weighting and calculating the data quality score according to the data integrity indicator, the data consistency indicator and the data timeliness indicator.

[0007] Further, the method for generating the set of standardized digital demand cells comprises: generating a cell head: generating a unique cell ID for each power grid material procurement demand, a unique edge node identifier, a generation timestamp and a version number; generating a demand ontology: generating a standardized code, a predicted value of the demand quantity of the power grid material procurement, a demand time window, and a spatial coordinate; calculating an optimal demand time that minimizes the total cost including procurement price cost, inventory cost and shortage risk cost according to the total cost of procurement at different times; generating an evidence chain: obtaining a set of hash values of the original data, a unique identifier of the prediction model used, decision logic and confidence score; generating a dynamic evolution identifier: marking the initial state as newborn.

[0008] Further, the method for obtaining the predicted value of the demand quantity comprises: calculating a predicted value based on the device state according to the current quantity of the same type of equipment, the average health index and the standard service life of the equipment in the labeled data; wherein, the standardized operation data of each device is calculated, and the standardized operation data is weighted and summed to obtain a single device health index, and the arithmetic mean of the health indexes of all the same type of equipment is taken to obtain an average health index; calculating the predicted value of the demand quantity based on the predicted value based on the device state and the predicted value based on the power grid planning.

[0009] Further, the method for obtaining the demand time window comprises: calculating the earliest demand time according to the current time and the shortest delivery cycle of the supplier; calculating the latest demand time according to the predicted failure time of the equipment and the emergency buffer time; wherein, the running data of the equipment whose service life exceeds the corresponding life threshold or the equipment whose running data exceeds the corresponding running threshold is collected in real time, the running data is taken as the input of the failure prediction model, and the predicted failure time of the equipment is obtained.

[0010] Further, the method for obtaining the trusted demand pool composed of the trusted demand cell set comprises: generating an edge node identified as a newborn standardized digital demand cell according to a dynamic evolution, and broadcasting complete data of the standardized digital demand cell to all other edge nodes in the consensus network; checking the integrity of the standardized digital demand cell evidence chain by the receiving node, and directly marking the standardized digital demand cell with an incomplete evidence chain as invalid; inquiring the association relationship between the material node and other nodes in the standardized digital demand cell, and if there is a logical conflict, marking it as a conflict cell and recording the corresponding conflict data; if the local data of the receiving node supports the demand of the standardized digital demand cell, then seconding the standardized digital demand cell and generating seconding information; calculating the final credibility according to the weighted calculation of the historical prediction accuracy of the node and the current data quality score of the node; calculating the total voting score according to the final credibility and the voting result of all edge nodes; if the total voting score is not lower than the consensus threshold, marking it as a confirmed state, generating a trusted demand cell, and recording the cell ID, confidence score, voting score, and state, otherwise marking it as to be optimized and returning the generating node to recalibrate; storing all trusted demand cells according to the material type and demand time window to obtain a trusted demand pool.

[0011] Further, the method for checking the integrity of the standardized digital demand cell evidence chain comprises: extracting a hash value set of the original data used when the standardized digital demand cell is generated, checking whether the standardized digital demand cell evidence chain contains at least one hash value of the original data, and comparing the hash values with the corresponding hash values in the preset reference library one by one to determine whether they are consistent; checking whether the decision logic description field in the standardized digital demand cell evidence chain exists, whether it is a structured rule statement, and whether it conforms to the checking rule in the checking rule reference library by combining NLP technology; Check whether the confidence score exists in the standardized digital demand cell evidence chain and the value is in the interval [0, 1]. Recalculate the check confidence score according to the data quality score of the standardized digital demand cell generation node and the model local test accuracy. Compare the difference between the check confidence score and the confidence score with the allowed error. If the evidence chain meets all the above determination results and all data are complete, the standardized digital demand cell is determined to be complete, otherwise, it is determined to be incomplete.

[0012] Further, the method for obtaining the optimal procurement scheme of the power grid materials includes: Set the optimization objective function, including total procurement cost minimization, supply chain risk minimization, inventory turnover rate optimization, and emergency response capability maximization. The total procurement cost is obtained by accumulating the total cost of different material categories in the trusted demand pool and based on the optimal demand time. The supply chain risk is obtained by weighted calculation of the supplier risks of different materials. The inventory turnover rate is obtained by calculating the predicted demand quantity of the material, the unit procurement price of the corresponding candidate procurement time, and the existing inventory quantity of the material. The emergency response capability is obtained by calculating the predicted delivery time of the material, the earliest demand time and the latest demand time in the demand time window of the material. Set the constraint conditions: Budget constraint: the procurement budget of all materials is not higher than the total procurement budget; the procurement budget of the material is calculated according to the demand quantity of the material and the unit procurement price of the material. Capacity constraint: the demand quantity of the material is not higher than the maximum capacity of the material supplier; when there are no less than two suppliers of the same material, the total supply quantity meets the demand quantity of the material. Time constraint: the predicted delivery time of the material is between the earliest demand time and the latest demand time in the demand time window of the material. Combine the optimization objective function and the constraint conditions, and perform multi-objective optimization based on the NSGA-II algorithm to obtain the optimal procurement scheme.

[0013] Further, the method for obtaining the supplier risk includes: The supplier risk is obtained by weighted calculation according to the standardized data of four indexes of delivery delay rate, quality unqualified rate, performance completion rate deviation and price abnormal fluctuation rate of each supplier; wherein, the delivery delay rate is obtained by calculating the ratio of the number of past delayed deliveries to the total number of past purchases; the quality unqualified rate is obtained by calculating the ratio of the total number of past unqualified materials in acceptance to the total delivery quantity in the past; the performance completion rate deviation is obtained by calculating according to the actual delivery quantity of the jth purchase and the contract quantity of the jth purchase; the price abnormal fluctuation rate is obtained by calculating the ratio of the number of past purchases with abnormal price fluctuation to the total number of past purchases; wherein, when the difference between the actual settlement unit price and the contract unit price is greater than the fluctuation threshold value, it is judged as abnormal.

[0014] The technical effects and advantages of the power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction are as follows: The present application firstly defines and standardizes the unified multi-source data caliber of the corresponding candidate procurement time NLP corresponding candidate procurement time with the semantic ontology, and combines the integrity, consistency and timeliness weighted score to guarantee the data quality, so as to eliminate the data understanding deviation of the edge node from the root; then the demand is materialized and traceable through the structured packaging of the digital demand cell body, the global correlation and evidence chain verification of the material correlation graph are relied on to filter the credible demand, and the node credibility weighted consensus mechanism is matched to aggregate the global cognition; for the node conflict, the local contradiction is resolved by global logic and resource constraint; finally, the multi-objective optimization of total procurement cost, supply chain risk, inventory turnover rate and emergency response capability is realized based on the corresponding candidate procurement time NSGA-Ⅱ corresponding candidate procurement time algorithm, and the global optimal procurement scheme is output, and the model and graph are continuously optimized through data feedback, which not only completely solves the decision-making contradiction problem of the edge node caused by the local perspective in the prior art, but also realizes the paradigm upgrade from the local reasonable corresponding candidate procurement time to the global optimal corresponding candidate procurement time, and significantly improves the procurement accuracy, supply chain stability and system self-evolution ability. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 It is a structure schematic diagram of the power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction of the present application; Figure 2 It is a method flowchart for obtaining a credible demand pool composed of credible demand cell bodies of the present application; Figure 3 It is a method flowchart for obtaining an optimal procurement scheme of power grid materials of the present application. DETAILED DESCRIPTION

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 Please see Figure 1 As shown, this embodiment provides a power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction, including: Data sensing module: Collects equipment status data, environmental market data, and planning data; analyzes and obtains labeled data and data quality scores; Equipment status data includes SCADA real-time monitoring data, equipment operation and maintenance records, and historical fault data. This data provides a unified objective basis for the equipment status of each edge node, avoiding judgment biases caused by nodes relying solely on fragmented local equipment data. Environmental and market data includes meteorological data, electricity price data, supplier capacity, delivery cycle, and price data. This data supplements the nodes' understanding of the external environment, reducing decision-making contradictions caused by missing local information. Planning data includes power grid construction plans, equipment upgrade and renovation plans, and procurement budget constraints. This data provides global strategic constraints, ensuring that decisions made by each node align with the overall power grid development goals and avoiding conflicts between local and global optima.

[0018] Methods for obtaining labeled data include: The system predefines the core semantic concepts of the power grid, as well as the attributes and relationships of these core semantic concepts. For example, core semantic concepts include materials, equipment, status, events, and needs. Material attributes include technical specifications, models, compatible equipment, and substitute codes. The relationship between equipment and materials is defined as matching use, replacement needs, etc. By unifying the understanding of core concepts such as materials, equipment, and needs among all nodes, the system eliminates semantic layer cognitive splits at the root and avoids contradictory judgments caused by different concept definitions among nodes. Based on NLP technology, the collected equipment status data, environmental market data, and planning data are analyzed and standardized, and labeled according to preset semantic labeling rules to obtain labeled data about core semantic concepts. Heterogeneous data is transformed into a unified semantic format, so that the data analysis of each edge node is based on data from the same source and in the same format, reducing decision-making bias caused by data format differences.

[0019] Methods for obtaining data quality scores include: A ratio of an actual number of labeled data pieces obtained to a number of data pieces that should be obtained is calculated to obtain a data integrity index; the number of data pieces that should be obtained is calculated according to data types, the data types including real-time monitoring data, periodic inspection data and static archive data, the number of data pieces that should be obtained of the real-time monitoring data being obtained by calculating a product of a collection time length, a collection frequency and a collection index number; the number of data pieces that should be obtained of the periodic inspection data being obtained by calculating a product of a total number of inspection objects, a number of inspection cycles in a time range and a single inspection index number; and the number of data pieces that should be obtained of the static archive data being obtained by calculating a product of a total number of archive objects and a number of fields that must be filled in for each object; the data integrity index quantifies a coverage degree of data of each node, and avoids information asymmetry type decision-making conflicts caused by missing of partial node data; A ratio of a number of labeled data pieces that meet semantic labeling rules to a total number of labeled data pieces is calculated to obtain a data consistency index; the data consistency index can ensure that the labeled data of each node meets uniform rules, and avoids data interpretation deviation type conflicts caused by data format corresponding to a candidate procurement time / semantic violation of the candidate procurement time; A ratio of a number of labeled data pieces for which a difference between a generation time of original data corresponding to the labeled data and a system receiving time is not higher than a difference threshold value to a total number of labeled data pieces is calculated to obtain a data timeliness index; the data timeliness index can ensure that fresh data is used by each node, and reduces timeliness difference type decision-making conflicts caused by different node data updates; A data quality score is obtained by weighted calculation according to the data integrity index, the data consistency index and the data timeliness index; the data quality score can provide a quantitative basis for credibility of data of each node, avoid unreasonable decision-making caused by blindly accepting low-quality data, and lay a foundation for weight allocation in a subsequent consensus stage.

[0020] An intelligent cognitive module: analyzing the labeled data and the data quality score to generate a standardized digital demand cell set; A method for generating the standardized digital demand cell set includes: Generating a cell head: generating a unique cell ID for each power grid material procurement demand, a unique edge node identifier, a generation time stamp and a version number; the unique identifier is used to trace a demand source and a version, and provides a basis for verifying demand legality in a subsequent consensus stage, and avoids responsibility ambiguity and decision-making conflicts caused by anonymous demands; Generating a demand ontology: generating a standardized code, a power grid material procurement demand quantity prediction value, a demand time window and a spatial coordinate; the standardized code is composed of a power grid material unified classification code, a technical specification parameter, a compatible equipment code and a preset substitute code; the standardized code integrates all-dimensional attributes of the material, can eliminate definition deviations of procurement materials by each node, and avoids procurement demand conflicts caused by inconsistent cognition of the materials; A method for obtaining the material demand quantity prediction value includes: The prediction value based on the equipment state is obtained according to the current same type equipment retention in the labeled data, the average health index and the equipment standard service life; wherein, the standardized operation data of each equipment is calculated, and then the standardized operation data is weighted and summed to obtain a single equipment health index, and the arithmetic average value of the health indexes of all same type equipment is obtained to obtain the average health index; the average health index can quantify the overall health level of the same type equipment, provide an objective and unified judgment standard for the demand quantity, and avoid excessive or insufficient demand caused by one-sided judgment of node data; The material demand quantity prediction value is obtained according to the prediction value based on the equipment state and the prediction value based on the power grid planning; wherein, the prediction value based on the power grid planning is obtained by extracting the planning file; combining the equipment state and the power grid planning can avoid the quantity prediction deviation caused by the node relying on local fragmented information only, and reduce the decision conflict of supply and demand mismatch; The method for obtaining the demand time window comprises: The earliest demand time is calculated according to the current time and the shortest delivery cycle of the supplier; The latest demand time is calculated according to the equipment predicted failure time and the emergency buffer time; wherein, the operation data of the equipment whose service life exceeds the corresponding life threshold or the operation data of the equipment whose operation data exceeds the corresponding operation threshold is collected in real time, the operation data is taken as the input of the failure prediction model to obtain the equipment predicted failure time; the life threshold is set according to the rated service life of each equipment, such as 80%; The optimal demand time that minimizes the total cost is calculated according to the total cost of purchasing at different times, including the purchase price cost, the inventory cost and the shortage risk cost; wherein, the purchase price cost is the product of the material demand quantity prediction value and the unit purchase price of the corresponding candidate purchase time; the inventory cost is the product of the material demand quantity prediction value, the unit purchase price of the corresponding candidate purchase time, the inventory holding rate and the inventory holding time; the shortage risk cost is the product of the material demand quantity prediction value, the shortage probability of the corresponding candidate purchase time and the unit shortage loss amount; the unit shortage loss amount is the loss caused by the late arrival of the material, which is calibrated according to the type of power grid equipment and can be directly collected from the operation and maintenance cost database; the inventory holding time is the time corresponding to the sum of the candidate purchase time and the delivery cycle minus the deployment start time; the shortage probability of the corresponding candidate purchase time is calculated based on the historical delivery delay rate of the supplier; the time boundary of the demand is clear, which can constrain the purchase time range of each node and avoid the contradiction of early purchase or delayed purchase caused by different judgments of the nodes on the demand urgency; the optimal demand time determined based on the total cost balance can avoid the global cost optimal conflict caused by the local cost considered by the node only; Generate evidence chain: obtain a set of hash values of the original data, the unique identifier of the prediction model used, the decision logic and the confidence score; the unique identifier of the prediction model used includes the model type and the parameter version; the decision logic is a structured rule statement corresponding to the predefined material replacement requirement, and the structured rule is generated based on the specific trigger condition of the current demand; the confidence score is the product of the data quality score and the local test accuracy of the prediction model, and the local test accuracy of the prediction model is the ratio of the number of correct predictions to the total number of predictions in the history of the prediction model; the evidence chain can provide complete traceability basis for demand generation, allowing nodes to cross-verify demand rationality and reduce conflicts caused by demands without basis; Generate dynamic evolution identifier: mark the initial state as newborn. The dynamic evolution identifier can clearly indicate the initial flow state of the demand, providing a unified standard for state synchronization in the subsequent consensus process, and avoiding decision conflicts caused by inconsistent understanding of demand progress among nodes.

[0021] Intelligent decision-making module: build a material association graph, analyze it in combination with the digital demand cell set, and obtain a trusted demand pool composed of a trusted demand cell set; The method for building the material association graph comprises: Define the nodes of the material association graph according to the core semantic concepts, and the node attributes are the key information of various entities, such as device ID, model, and running state for device nodes, and specification parameters for material nodes; Define the association relationship between nodes; Based on the power grid design drawings, equipment operation and maintenance records, and material matching manuals, input node information and association relationship to form a material association graph. The material association graph can provide a global association knowledge base for power grid equipment, materials, and engineering, eliminating logical conflicts caused by lack of local association knowledge among nodes.

[0022] Refer to Figure 2 The method for obtaining a trusted demand pool composed of a trusted demand cell set comprises: Generate a dynamic evolution identifier for the edge node of a newborn standardized digital demand cell, and broadcast the complete data of the standardized digital demand cell to all other edge nodes in the consensus network; this can break down the information barriers of edge nodes, allow demands to be cross-verified by all network nodes, and avoid the partiality of decisions made by a single node based on a local perspective; The receiving node checks the integrity of the evidence chain of the standardized digital demand cell, and directly marks the standardized digital demand cell with an incomplete evidence chain as invalid; integrity checking can filter out demands with no basis, fake or abnormal data, ensure that demands entering the consensus have a trusted traceability basis, and reduce conflicts caused by demands without evidence; The method for checking the integrity of the evidence chain of the standardized digital demand cell comprises: extracting a set of hash values of the original data used in the generation of the standardized digital demand cell, checking whether at least one hash value of the original data is contained in the standardized digital demand cell evidence chain, and comparing the hash value with the corresponding hash value in the preset reference library to determine whether they are consistent; checking whether the decision logic description field exists in the standardized digital demand cell evidence chain, whether it is a structured rule statement, and whether it conforms to the verification rule in the verification rule reference library by combining NLP technology analysis; checking whether the confidence score exists in the standardized digital demand cell evidence chain and whether the numerical value is within the [0, 1] interval, recalculating the verification confidence score according to the data quality score of the standardized digital demand cell generation node and the model local test accuracy, and comparing whether the difference between the verification confidence score and the confidence score is within the allowed error; The evidence chain satisfies all the above determination results, and all the data of the standardized digital demand cell are complete, and the standardized digital demand cell is determined to be complete, otherwise it is determined to be incomplete.

[0023] Querying the association relationship between the material node and other nodes in the standardized digital demand cell, if there is a logical conflict, marking it as a conflict cell and recording the corresponding conflict data; conflict resolution; association relationship query and conflict detection can actively identify the logical contradiction between demands and resolve the decision conflict caused by information asymmetry in advance; The conflict resolution method comprises: According to the node relationship of the conflict data based on the material association graph, the conflict type is determined, including direct logical conflict, resource constraint conflict and matching association conflict; Extracting the core attributes of the conflict cell, including material gene code, demand time window and spatial coordinates, etc. Based on the material association graph, the conflict association node is located; Call the material association graph to trace the logical link and check the integrity of the evidence chain of the conflict cell; the method steps for checking the integrity of the evidence chain of the standardized digital demand cell are the same; When the conflict type is a direct logical conflict, the priority determination rule is used for determination, the priority determination rule is based on the priority definition of the power grid business, and if the priority of both parties is consistent, the multi-node weighted voting is used for decision; When the conflict type is a resource constraint conflict, based on the power grid planning constraint, the total amount of resources involved in the conflict is calculated, and the optimization is performed according to the power grid safety guarantee degree, demand urgency and cost benefit, and the resources are allocated according to the optimization result; When the conflict type is a matching association conflict, based on the matching use relationship of the material association graph, the missing matching material demand is automatically identified, and the matching demand cell is generated according to the matching material demand.

[0024] If the local data of the receiving node supports the demand of the standardized digital demand cell, the standardized digital demand cell is seconded and seconded information is generated, the seconded information including the node ID and the hash value of the local support data; the seconding mechanism can aggregate the local data support of multiple nodes, upgrade the demand from a single node to multiple nodes, and reduce unreasonable demand caused by local data deviation; The final credibility is obtained by weighted calculation according to the historical prediction accuracy of the node and the current data quality score of the node; the final credibility can give higher speech power to the node with high historical performance and high current data quality, avoid the interference of the wrong judgment of the low-quality node on the global consensus, and improve the consistency of decision-making; According to the final credibility of all edge nodes and the voting result, the total voting score is calculated; If the total voting score is not less than the consensus threshold, it is marked as confirmed state, a credible demand cell is generated, and the cell ID, confidence score, voting score and state are recorded, otherwise it is marked as to be optimized and returned to the generating node for recalibration; the global unified credible demand standard is achieved through weighted voting, the local optimal of each node is converted into global optimal, and the decision-making contradiction is completely solved; All credible demand cells are classified and stored according to the material type and demand time window to obtain a credible demand pool, and then the unified demand after consensus is integrated to provide consistent input basis for subsequent procurement plan generation, avoiding conflicts at the execution level caused by scattered demand sources.

[0025] Demand generation module: according to the credible demand pool and the labeled data, the optimal procurement scheme of power grid materials is obtained.

[0026] Reference Figure 3 The method for obtaining the optimal procurement scheme of power grid materials comprises: Set the optimization objective function, including total procurement cost minimization, supply chain risk minimization, inventory turnover rate optimization and emergency response capability maximization; wherein, the total procurement cost is obtained by accumulating the total cost of different materials in the credible demand pool and based on the optimal demand time; the optimization objective function integrates the four global targets of total procurement cost, supply chain risk, inventory turnover rate and emergency response capability, avoiding the global optimal conflict caused by the focus of edge nodes on local cost; the total procurement cost is based on the total cost accumulation of the optimal demand time of each material, ensuring the global cost balance of the procurement scheme in the time dimension, and resolving the cost optimal contradiction caused by the difference in procurement time of the nodes; Supply chain risk is calculated by weighting the supplier risks of different materials; the weighting weights can be obtained through natural heuristic optimization algorithms; supplier risk is calculated by weighting standardized data of four indicators for each supplier: delivery delay rate, quality non-conformity rate, performance deviation, and abnormal price volatility rate; specifically, the delivery delay rate is calculated as the ratio of the number of delayed deliveries in the past to the total number of purchases in the past; the quality non-conformity rate is calculated as the ratio of the total number of non-conforming materials in the past to the total number of deliveries in the past; and the performance deviation is calculated based on the actual delivery quantity of the j-th purchase and the contractually agreed quantity of the j-th purchase. ,in, This represents the actual quantity delivered in the j-th procurement. The quantity stipulated in the contract for the j-th purchase; The total number of purchases from the current supplier is used as the basis for calculating the price volatility rate by dividing the number of purchases during periods of abnormal price fluctuations by the total number of purchases in the past. An abnormality is defined as the ratio of the difference between the actual settlement price and the contract price to the contract price exceeding a volatility threshold, which can be adjusted according to industry standards. Supply chain risk is assessed through unified historical data indicators and weighted calculations, providing a consistent supplier risk assessment standard for each node and preventing inconsistencies in supplier risk assessments due to differences in local cooperation experiences. Inventory turnover rate is calculated based on the predicted quantity of materials needed, the unit purchase price at the corresponding candidate procurement time, and the existing inventory quantity of the materials; for example, inventory turnover rate... ,in, For the first Forecasted demand quantities for this type of material; For the first The unit purchase price for the corresponding candidate procurement time for this type of material; For the first The current inventory quantity of this type of material; The number of types of materials required; the inventory turnover rate is calculated based on the linkage between global inventory data and procurement demand, avoiding the contradiction of excessive procurement occupying inventory or insufficient procurement leading to idle inventory due to neglecting the overall inventory level at each node; Emergency response capability is calculated based on the predicted delivery time of materials, the earliest demand time, and the latest demand time within the material demand time window; for example, emergency response capability... ,in, For the first The predicted delivery time for this type of material is the sum of the average delivery time of the same type of material in the past and the supplier's expected delivery cycle. For the first The latest time of need for this type of material; For the first The earliest demand time of the material; the emergency response capability is matched with the accurate prediction delivery time through the demand time window, so as to ensure that the procurement scheme meets the global emergency demand of the power grid and resolves the decision conflict caused by the difference in local emergency priority of the node; The constraint conditions are set, including the budget constraint: the procurement budget of all materials is not higher than the total procurement budget; the procurement budget of the material is obtained according to the demand quantity of the material and the unit procurement price of the material corresponding to the candidate procurement time; The capacity constraint: the demand quantity of the material is not higher than the maximum capacity of the material supplier; when there are no less than two suppliers of the same material, the total supply quantity meets the demand quantity of the material; The time constraint: the prediction delivery time of the material is between the earliest demand time and the latest demand time in the demand time window of the material; the constraint condition is a unified constraint with the global budget, capacity and time boundary, which limits the local decision of the node to break through the upper limit of the global resource, and avoids the contradiction that the procurement scheme is not executable due to no constraint; In combination with the optimization objective function and the constraint condition, multi-objective optimization is carried out based on the NSGA-II algorithm to obtain an optimal procurement scheme. The intelligent algorithm balances the multi-objective conflict, converts the local optimal demand of each node into a globally unified optimal procurement scheme, and completely solves the decision conflict of the node; the optimal procurement scheme provides a globally consensus procurement list, strategy and plan, and provides a unified action basis for the execution layer, so as to avoid the procurement execution confusion caused by the difference in node demand.

[0027] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0028] Finally: the above is only a preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction, characterized in that, include: Data sensing module: collects equipment status data, environmental market data, and planning data; The analysis yielded labeled data and data quality scores; Intelligent decision-making module: Constructs a material association map, analyzes it in conjunction with a set of digital demand cells, and obtains a trustworthy demand pool composed of a set of trustworthy demand cells; Intelligent decision-making module: Constructs a material association map, analyzes it in conjunction with a set of digital demand cells, and obtains a trustworthy demand pool composed of a set of trustworthy demand cells; Demand generation module: Based on the trusted demand pool and labeled data, analyze and obtain the optimal procurement plan for power grid materials.

2. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 1, characterized in that, Methods for obtaining labeled data include: A predefined set of core semantic concepts for the power grid, along with the attributes and relationships of these core semantic concepts; Based on NLP technology, the collected equipment status data, environmental market data, and planning data are analyzed and standardized, and labeled according to preset semantic labeling rules to obtain labeled data about core semantic concepts.

3. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 1, characterized in that, Methods for obtaining data quality scores include: Calculate the ratio of the actual number of labeled data entries to the number of data entries that should be acquired to obtain the data integrity index; Calculate the ratio of the number of labeled data entries that conform to the semantic annotation rules to the total number of labeled data entries to obtain the data consistency index; The data timeliness index is obtained by calculating the ratio of the number of data entries whose difference between the original data generation time and the system reception time is not higher than the difference threshold to the total number of data entries. The data quality score is calculated by weighting data integrity, data consistency, and data timeliness indicators.

4. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 1, characterized in that, Methods for generating standardized digital demand cell sets include: Generate cell header: Generate a unique cell ID for each power grid material procurement requirement, a unique identifier for edge nodes, and generate a timestamp and version number; Generate demand ontology: Generate standardized codes, predicted quantities of power grid material procurement needs, demand time windows, and spatial coordinates; Based on the total cost of procurement at different times, including procurement price cost, inventory cost, and stockout risk cost, calculate the optimal demand time that minimizes the total cost. Generate a chain of evidence: obtain the set of hash values ​​of the original data, the unique identifier of the model used for prediction, the decision logic, and the confidence score; Generate dynamic evolution identifiers: mark the initial state as a newborn.

5. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 4, characterized in that, Methods for obtaining forecasts of material demand include: Based on the current inventory of similar equipment, average health index, and standard service life of the equipment in the labeled data, a predicted value based on the equipment status is obtained. Specifically, the standardized operating data of each piece of equipment is calculated, and then the standardized operating data is weighted and summed to obtain the health index of a single piece of equipment. The arithmetic mean of the health indices of all similar equipment is taken to obtain the average health index. The predicted quantity of materials required is calculated based on the predicted values ​​based on equipment status and the predicted values ​​based on power grid planning.

6. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 4, characterized in that, Methods for obtaining demand time windows include: Calculate the earliest required time based on the current time and the supplier's shortest delivery cycle; The latest required time is calculated based on the predicted equipment failure time and emergency buffer time. Specifically, the operating data of equipment whose service life exceeds the corresponding service life threshold or whose operating data exceeds the corresponding operating threshold is collected in real time. The operating data is used as the input of the failure prediction model to obtain the predicted equipment failure time.

7. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 1, characterized in that, Methods for obtaining a trusted demand pool consisting of a set of trusted demand cells include: The edge nodes that generate dynamic evolution identifiers as newly formed standardized digital demand cells are broadcast to all other edge nodes in the consensus network. The receiving node verifies the integrity of the evidence chain of the standardized digital requirement cell and directly marks the standardized digital requirement cell with an incomplete evidence chain as invalid; Query the relationships between material nodes and other nodes in the standardized digital demand cell. If there is a logical conflict, mark it as a conflicting cell and record the corresponding conflict data. If the local data of the receiving node supports the requirements of the standardized digital requirement cell, then the standardized digital requirement cell is endorsed and endorsement information is generated. The final credibility is calculated by weighting the node's historical prediction accuracy and the node's current data quality score. The total voting score is calculated based on the final credibility and voting results of all edge nodes. If the total voting score is not lower than the consensus threshold, it is marked as confirmed, a trusted requirement cell is generated, and the cell ID, confidence score, voting score, and status are recorded. Otherwise, it is marked as to be optimized and returned to the generation node for recalibration. All trusted demand cells are categorized and stored according to material type and demand time window to obtain a trusted demand pool.

8. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 7, characterized in that, Methods for verifying the integrity of the standardized digital requirement cellular evidence chain include: Extract the set of hash values ​​of the original data used in the generation of the standardized digital requirement cell. Check whether the standardized digital demand cell evidence chain contains the hash value of at least one piece of original data, and compare the hash value with the corresponding hash value in the preset benchmark library one by one to determine whether they are consistent. Check whether the decision logic description field exists in the evidence chain of the standardized digital demand cell and whether it is a structured rule statement. Combine NLP technology to analyze and determine whether it conforms to the verification rules in the verification rule benchmark library. Check whether the confidence score exists in the evidence chain of the standardized digital demand cell and whether the value is within the range of [0,1]. Recalculate the verification confidence score based on the data quality score of the standardized digital demand cell generation node and the local test accuracy of the model. Compare whether the difference between the verification confidence score and the confidence score is within the allowable error. The evidence chain is considered complete if all data in the chain meets the standardized digital requirements of all the above judgment results; otherwise, it is considered incomplete.

9. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 1, characterized in that, Methods for obtaining the optimal procurement plan for power grid materials include: The optimization objective function is set to minimize total procurement cost, minimize supply chain risk, optimize inventory turnover rate, and maximize emergency response capability; among which, the total procurement cost is obtained by summing the total cost based on the different types of materials in the reliable demand pool and the total cost based on the optimal demand time. Supply chain risk is calculated by weighting the risks of different suppliers for different goods; Inventory turnover rate is calculated based on the forecast of material demand, the unit purchase price at the corresponding candidate purchase time, and the existing inventory quantity of the material. Emergency response capability is calculated based on the predicted delivery time of materials, the earliest demand time and the latest demand time in the demand time window of materials; Set constraints: Budget constraint: The procurement budget for all materials shall not exceed the total procurement budget; the procurement budget for materials shall be calculated based on the required quantity of materials and the unit purchase price of the materials. Capacity constraints: The demand for materials does not exceed the maximum capacity of the suppliers of the materials. When there are no fewer than two suppliers of the same material, the total supply meets the demand for the materials. Time constraint: The predicted delivery time of the material is between the earliest and latest demand times within the material's demand time window; By combining the objective function and constraints, the NSGA-II algorithm is used to perform multi-objective optimization to obtain the optimal procurement plan.

10. The power grid material procurement demand generation system based on multi-source data fusion and intelligent prediction according to claim 9, characterized in that, Methods for obtaining supplier risk include: Supplier risk is calculated using standardized data from four indicators: delivery delay rate, quality non-conformity rate, performance deviation, and price volatility. Specifically, the delivery delay rate is calculated as the ratio of the number of delayed deliveries to the total number of purchases in the past; the quality non-conformity rate is calculated as the ratio of the total number of non-conforming goods received in the past to the total number of goods delivered in the past; the performance deviation is calculated based on the actual quantity delivered in the j-th purchase and the quantity stipulated in the contract for the j-th purchase; and the price volatility rate is calculated as the ratio of the number of purchases during periods of abnormal price fluctuations in the past to the total number of purchases in the past. An abnormality is defined as the ratio of the difference between the actual settlement unit price and the contracted unit price to the contracted unit price exceeding a volatility threshold.

Images