Power material demand data-based procurement strategy optimization method and system

Abstract

The application discloses a procurement strategy optimization method and system based on power material demand data, relates to the technical field of intelligent procurement management of a power system, and comprises the following steps: fuzzy modeling is carried out by using spherical fuzzy values; preliminary causal strength is determined based on Granger causal analysis; an initial weight population is generated by using a search window and a Bezier curve; the weight is optimized in combination with NSGA-II; a parent node set is identified based on an MMHC algorithm; an outlying factor is calculated by using an FDPC-OF method; a KD-Tree index is constructed and updated; and classification is carried out by using Drools. The application improves the accuracy of demand prediction and the scientificity of priority ranking by combining spherical fuzzy modeling with causal analysis and NSGA-II optimization, improves the stability and adjustability of procurement strategies by using the MMHC algorithm and the Hill-Climbing algorithm in linkage optimization, and by combining the FDPC-OF density peak outlying identification method with the KD-Tree high-dimensional index structure.

Description

Technical Field

[0001] This invention relates to the field of intelligent procurement management technology for power systems, and in particular to a method and system for optimizing procurement strategies based on power material demand data. Background Technology

[0002] With the continuous expansion of the power system and the ongoing improvement of its intelligence level, power companies have placed higher demands on the response speed, forecast accuracy, and supply coordination capabilities of material procurement. Traditional power material procurement mainly relies on manual experience or rule-based decision-making methods based on static inventory thresholds, lacking in-depth integration and analysis of dynamic market environments, real-time grid operation data, and multi-source heterogeneous demand information. With the in-depth application of technologies such as artificial intelligence, machine learning, causal modeling, and intelligent optimization in power dispatching, operation and maintenance management, fuzzy modeling technology has shown advantages in uncertain decision-making, and can handle multi-source heterogeneous data more flexibly. The Granger causal analysis method can effectively mine the temporal dependencies between variables, providing more reliable causal support for demand forecasting. Multi-objective optimization algorithms have significant potential in solving multi-objective trade-offs such as procurement costs, delivery cycles, and inventory risks.

[0003] However, current technologies still have shortcomings. The multidimensional uncertainty representation advantage of spherical fuzzy values ​​lacks a comprehensive calculation method for scene membership based on Sugeno-Weber weighted average, resulting in insufficient adaptability and problems such as material mismatch and supply delays. Furthermore, it fails to fully integrate dynamic detection based on FDPC-OF outlier factors and KD-Tree indexes, making it difficult to optimize procurement strategies in real time. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a procurement strategy optimization method and system based on power material demand data. It addresses the lack of advantages in the multidimensional uncertainty representation of spherical fuzzy values, which leads to insufficient adaptability and problems such as material mismatch and supply delays. Furthermore, it fails to fully integrate dynamic detection based on FDPC-OF outlier factors and KD-Tree indexes, making it difficult to optimize procurement strategies in real time.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] Firstly, the present invention provides a procurement strategy optimization method based on power material demand data, which includes,

[0008] Collect power data and third-party demands, preprocess them, construct mapping rules, perform mapping through a rule engine, perform fuzzy modeling using spherical fuzzy values, calculate the comprehensive membership degree of the scenario using Sugeno-Weber weighted average, construct hybrid scenarios based on fuzzy rules, generate comprehensive scenario demand prediction intervals, determine preliminary causal strength based on Granger causality analysis, generate an initial weight population using search windows and Bézier curves, construct a multi-objective benefit function and combine it with NSGA-II to optimize weights, generate a smooth Pareto front, select the optimal optimization weights, calculate comprehensive scenario demand predictions and prioritize them, and generate a comprehensive scenario demand prediction report.

[0009] The PSM is used to calculate propensity scores to generate a balanced dataset. A directed acyclic graph is constructed using Gaussian distribution and Pearson correlation coefficient. The set of parent nodes is identified based on the MMHC algorithm. The optimal causal network is generated through Hill-Climbing optimization. The outlier factor is calculated using the FDPC-OF method. A KD-Tree index is constructed for updating. Drools is used for classification to generate a structured anomaly report. The collaborative cost of the update strategy is optimized based on the P2P collaborative network, and the optimized procurement strategy is output.

[0010] Implement optimized procurement strategies and store them in the database.

[0011] As a preferred embodiment of the procurement strategy optimization method based on power material demand data described in this invention, the following steps are included: constructing mapping rules, using spherical fuzzy values ​​for fuzzy modeling, and calculating the comprehensive membership degree of the scenario through Sugeno-Weber weighted average to generate a comprehensive scenario demand prediction interval, including:

[0012] Based on the characteristics of power data, mapping rules are constructed, and the power data characteristics are mapped to the corresponding scenarios through the rule engine to generate a scenario feature matrix.

[0013] Spherical fuzz values ​​are calculated for the power data features of each scenario, and the power data features of each scenario are fused using Sugeno-Weber weighted average. The comprehensive membership degree of the scenario is calculated, and combined with the scenario feature matrix, a comprehensive scenario demand prediction interval is generated.

[0014] As a preferred embodiment of the procurement strategy optimization method based on power material demand data described in this invention, the following steps are included: determining the preliminary causal strength based on Granger causality analysis, generating an initial weight population using Bézier curves, optimizing the weights using NSGA-II, and forming a comprehensive scenario demand forecast report.

[0015] Based on the multi-party interest coordination matrix, Granger causal analysis is used to calculate the preliminary causal strength. Combined with the comprehensive scenario demand prediction interval, the initial weight range is calculated. An initial weight population is generated through a search window and smoothed using a Bézier curve.

[0016] Based on the initial causal strength, a multi-objective benefit function is defined;

[0017] The initial weighted population is adjusted using preliminary causal strength. Combined with a multi-objective payoff function, the initial weighted population is optimized using the NSGA-II algorithm based on the multi-objective payoff function and causal adjustment weights to generate a dynamic Pareto front. The front is then smoothed using a Bézier curve. The optimal optimized weights are selected, and the comprehensive scenario demand prediction is calculated. The priority list is generated by sorting the data in descending order. The membership degrees of each scenario, the membership degrees of mixed scenarios, the optimal optimized weights, and the comprehensive scenario demand prediction are combined to form a comprehensive scenario demand prediction report.

[0018] As a preferred embodiment of the procurement strategy optimization method based on power material demand data described in this invention, the following steps are included: constructing a directed acyclic graph, optimizing it using Hill-Climbing, and calculating the outlier factor using the FDPC-OF method:

[0019] Extract the comprehensive scenario demand forecast from the comprehensive scenario demand forecast report, combine it with market price and supplier inventory, use PSM to calculate the propensity score, generate a balanced dataset, and construct a directed acyclic graph.

[0020] The Hill-Climbing algorithm is used to optimize the directed acyclic graph, generate the optimal causal relationship network, calculate the conditional probability, and calculate the policy conditional probability for each dimension.

[0021] Using FDPC-OF, we calculate the local density between each node, the centripetal relative distance between each node, and the outlier factor through K nearest neighbors;

[0022] Nodes with outlier factors greater than the outlier factor threshold (based on experimental analysis) are filtered out, marked as anomalies, and a list of anomaly events is output.

[0023] As a preferred embodiment of the procurement strategy optimization method based on power material demand data described in this invention, the steps of constructing and updating a KD-Tree index, classifying using Drools, and optimizing based on a P2P collaborative network include:

[0024] Construct a KD-Tree index, use K-nearest neighbors in the KD-Tree, recalculate and update the outlier factor, update the list of anomalous events, and calculate the anomaly priority;

[0025] The Drools algorithm is used to classify the anomaly types in the list of update anomalies, generate classification labels, update the conditional probabilities, optimize the edge weights using the Hill-Climbing algorithm, calculate the conditional probabilities of update strategies, generate an update strategy population, and build a P2P collaborative network among suppliers to optimize collaborative costs.

[0026] By optimizing costs and updating strategy populations, combined with structured anomaly reports, we can integrate them into an optimized procurement strategy.

[0027] As a preferred embodiment of the procurement strategy optimization method based on power material demand data described in this invention, the step of executing the optimized procurement strategy and storing it in the database includes:

[0028] The optimized procurement strategy is broken down into execution tasks through a message queue service, distributed via API, and indexed using Elasticsearch. The optimized procurement strategy and demand forecast report are then stored in the database.

[0029] As a preferred embodiment of the procurement strategy optimization method based on power material demand data described in this invention, the step of collecting power data and third-party requests, and performing preprocessing, includes:

[0030] Use API interfaces to collect power data and extract power data features;

[0031] Use API interfaces to collect the demands of power grid operators, suppliers, and procurement departments, and construct a multi-party interest coordination matrix.

[0032] Secondly, this invention provides a procurement strategy optimization system based on power material demand data, including:

[0033] The data collection and forecasting module is used to collect electricity data and third-party demands, preprocess the data, construct mapping rules, perform mapping through a rule engine, perform fuzzy modeling using spherical fuzzy values, calculate the comprehensive membership degree of the scenario using Sugeno-Weber weighted average, construct hybrid scenarios based on fuzzy rules, generate comprehensive scenario demand forecast intervals, determine preliminary causal strength based on Granger causality analysis, generate an initial weight population using a search window and Bézier curve, construct a multi-objective benefit function and combine it with NSGA-II to optimize the weights, generate a smooth Pareto front, select the optimal optimization weights, calculate the comprehensive scenario demand forecast, prioritize the forecasts, and generate a comprehensive scenario demand forecast report.

[0034] The detection optimization module is used to generate a balanced dataset by calculating propensity scores using PSM, construct a directed acyclic graph using Gaussian distribution and Pearson correlation coefficient, identify the set of parent nodes based on the MMHC algorithm, generate the optimal causal network through Hill-Climbing optimization, calculate the outlier factor using the FDPC-OF method, build and update the KD-Tree index, classify using Drools, generate a structured anomaly report, optimize the collaborative cost of the update strategy based on the P2P collaborative network, and output the optimized procurement strategy.

[0035] The execution storage module is used to execute optimized procurement strategies and store them in the database.

[0036] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the procurement strategy optimization method based on power material demand data as described in the first aspect of the present invention.

[0037] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the procurement strategy optimization method based on power material demand data as described in the first aspect of the present invention.

[0038] The beneficial effects of this invention are as follows: This invention uses spherical fuzzy values ​​for fuzzy modeling, determines preliminary causal strength based on Granger causality analysis, generates an initial weight population using a search window and Bézier curves, optimizes weights using NSGA-II, identifies the parent node set based on the MMHC algorithm, calculates outlier factors using the FDPC-OF method, constructs a KD-Tree index for updating, and uses Drools for classification; it improves the accuracy of demand forecasting and the scientific nature of priority ranking; and enhances the stability and adjustability of procurement strategies. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of the procurement strategy optimization method based on power material demand data in Example 1.

[0041] Figure 2 This is a schematic diagram of the procurement strategy optimization system based on power material demand data in Example 1.

[0042] Figure 3 This is a schematic diagram of the causal network for constructing the procurement strategy optimization method based on power material demand data in Example 1.

[0043] Figure 4 This is a schematic diagram illustrating the generation of the scenario feature matrix for the procurement strategy optimization method based on power material demand data in Example 1. Detailed Implementation

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0047] Example 1, referring to Figures 1 to 4 This is the first embodiment of the present invention, which provides a procurement strategy optimization method based on power material demand data, including the following steps:

[0048] S1. Collect power data and third-party demands, preprocess them, construct mapping rules, perform mapping through a rule engine, use spherical fuzzy values ​​for fuzzy modeling, calculate the comprehensive membership degree of the scenario using Sugeno-Weber weighted average, construct a hybrid scenario based on fuzzy rules, generate a comprehensive scenario demand prediction interval, determine the preliminary causal strength based on Granger causality analysis, generate an initial weight population using a search window and Bézier curve, construct a multi-objective benefit function and combine it with NSGA-II to optimize the weights, generate a smooth Pareto front, select the optimal optimization weights, calculate the comprehensive scenario demand prediction and prioritize it, and generate a comprehensive scenario demand prediction report.

[0049] Specifically, this involves collecting electricity data and third-party requests, and preprocessing them, including:

[0050] Electricity data is collected using API interfaces, including real-time and historical load, failure rate, supplier inventory, market price, temperature, price fluctuation frequency, carbon emission policies, and electricity industry regulations. Duplicate electricity data is removed, and missing data is filled using mean imputation.

[0051] The API interface is used to collect the demands of power grid operators (stability, cost, and environmental protection), suppliers (profit margin, delivery time, and long-term cooperation), and procurement departments (budget, timeliness, quality, and environmental protection). The demands of the three parties are quantified by weighted average method and normalized to construct a multi-party interest coordination matrix.

[0052] Based on power data, power data features are extracted, including calculating monthly average load, monthly average temperature, monthly failure rate, and monthly average delivery time using the mean method, calculating monthly price volatility using the standard deviation method, and normalizing them in conjunction with monthly carbon emission limits (based on carbon emission policies).

[0053] By collecting and cleaning multi-source power data, the cost of manual data collection is significantly reduced, and the timeliness and breadth of data are improved. Duplicate data is removed and missing values ​​are filled to ensure data consistency and integrity, providing high-quality input for subsequent analysis. A multi-party interest coordination matrix is ​​constructed to realize the numerical and controllable multi-party game relationship. Monthly indicators are extracted by the mean method and standard deviation method, and normalized by combining carbon emission quotas to form multi-dimensional time series features. This not only improves the prediction accuracy of the model, but also enhances the responsiveness to carbon policy sensitivity.

[0054] Furthermore, mapping rules are constructed, fuzzy modeling is performed using spherical fuzzy values, and the comprehensive membership degree of the scene is calculated using Sugeno-Weber weighted average to generate a comprehensive scene demand prediction interval, including:

[0055] The formula for calculating the interval representation of power data characteristics is as follows:

[0056] [F min,i ,F max,i ] = [F i -σ i ,F i +σ i ],

[0057] Among them, F min,i and F max,i F represents the upper and lower bounds of feature i, respectively. i σ is the normalized value of feature i. i The standard deviation of feature i;

[0058] Based on interval representation, the priority among power data features is calculated using interval likelihood theory, with the following formula:

[0059]

[0060] Among them, [p ij [This indicates the priority of feature i over feature j;]

[0061] Based on the multi-party interest coordination matrix and the priority among power data features, the weight of each power data feature is calculated using the following formula:

[0062] ω i =∑ k w k,i ·p ij ,

[0063] Where, ω i w represents the weight of feature i. k,i The weights of participant k (including grid operators, suppliers, and procurement departments) for feature i (obtained based on the weighted average method);

[0064] Based on power data, four demand scenarios are defined through empirical analysis, including routine scenarios, emergency scenarios, peak scenarios, and green scenarios.

[0065] Based on the characteristics of power data, load, temperature, failure rate and carbon emission thresholds are set (the first three thresholds are the corresponding monthly averages, and the carbon emission threshold is the monthly carbon emission limit). Based on empirical rules, mapping rules are constructed (if the load is greater than the threshold and the temperature is greater than the temperature threshold, it is mapped to the peak scenario; if the failure rate is greater than the failure rate threshold, it is mapped to the emergency scenario; if the carbon emission target is less than the carbon emission threshold, it is mapped to the green scenario; otherwise, it is mapped to the normal scenario).

[0066] Based on mapping rules, power data features are mapped to corresponding scenarios through a rule engine to generate a scenario feature matrix;

[0067] The spherical blur value is calculated for the power data features of each scenario, using the following formula:

[0068]

[0069] Among them, O s,i Let μ be the spherical blur value of feature i in scene s. s,i Let η be the membership degree of feature i to scene s (based on the matching degree between the feature value and the scene threshold; for example, if the carbon emission target is less than the carbon emission threshold, then the matching degree of the green scene is the carbon emission threshold). s,i Let ν represent the neutrality of feature i with respect to scene s (expressed as half the membership value). s,i Let i be the degree of non-membership of feature i with respect to scene s;

[0070] Based on spherical fuzzy values, the power data features of each scene are fused using a Sugeno-Weber weighted average to calculate the comprehensive membership degree of the scene. The formula is as follows:

[0071]

[0072] Among them, Q s Let τ be the overall membership degree of scenario s, τ be the Sugeno-Weber parameter (set based on carbon emission policy), and n be the number of features;

[0073] Based on the characteristics of power data, fuzzy rules are defined (for example, if the load is greater than the load threshold and the temperature is greater than the temperature threshold, the membership degree of the peak scenario is increased; if the load is less than the load threshold, it belongs to both the regular and peak scenarios and is defined as a mixed scenario).

[0074] The membership degree of a mixed scene is calculated using the following formula:

[0075] Q 1 mix =∑ s∈S α s ·Q s ,

[0076] Among them, Q 1 mix For the membership degree of mixed scenes, α s The blending coefficient for scene s (set based on mapping rules);

[0077] Based on the membership degree and scene feature matrix of each scene, a comprehensive scene demand prediction interval is generated. Then, combining the membership degree of the mixed scenes, the comprehensive mixed scene demand is calculated using the following formula:

[0078]

[0079] D mix =∑ s∈S α s ·[D s,min D s,max ],

[0080] Among them, D s,min and D s,max These are the lower and upper limits of the comprehensive scenario demand prediction range for scenario s, respectively. mix To meet the needs of a comprehensive mix of scenarios, S represents the number of scenarios.

[0081] By representing power data characteristics as upper and lower limit intervals and introducing standard deviation as a dynamic uncertainty factor, the model's adaptability to data fluctuations is improved, maintaining stability under noise or measurement errors and providing a foundation for high-reliability data analysis. Prioritization between features is calculated based on interval probability, enhancing the accuracy and agility of response decisions. The preferences of grid operators, suppliers, and purchasers are incorporated into weight calculation, reflecting the interests of each participant and improving the model's relevance to real-world decision-making. This contributes to building a win-win power market mechanism. Spherical fuzzy representation is used in scenario determination, enabling the system to provide accurate modeling and analysis even under conditions of incomplete information or fuzzy judgments. Expanding its applicability in highly uncertain scenarios such as new energy access, the Sugeno-Weber weighted average method enhances feature fusion capabilities. This method has strong modeling capabilities for additive and multiplicative relationships and can simulate nonlinear interactions between multiple variables such as carbon emissions, load, and temperature, enhancing the flexibility of response to policy changes. By defining a mixing coefficient and linking it with scenario membership to calculate mixed scenarios, it effectively solves the problems of blurred scenario boundaries and cross-scenario coupling in reality, enabling the prediction system to more realistically simulate the actual operating environment. The proposed comprehensive scenario demand prediction interval allows the dispatch center to allocate resources with risk awareness, making capacity reserves and flexible allocations, thus helping to achieve the dual goals of load balancing and low-carbon operation.

[0082] Furthermore, based on Granger causality analysis, preliminary causal strength is determined, and an initial weighted population is generated using Bézier curves. The weights are then optimized using NSGA-II to generate a comprehensive scenario demand prediction report, including:

[0083] Based on the multi-party interest coordination matrix, Granger causality analysis is used to calculate the preliminary causal strength.

[0084] Based on the preliminary causal strength and the comprehensive scenario demand prediction range, the initial weight range is calculated using the following formula:

[0085]

[0086] Among them, w s,min and w s,max These are the upper and lower bounds of the weight interval for scene s, respectively, and CS s Let be the initial causal strength of scenario s;

[0087] Based on the initial weight range, an initial weight population is generated through a search window and smoothed using a Bézier curve.

[0088] The power grid operator, suppliers, and procurement department are defined as participants, while fuzzy rules are defined as strategists.

[0089] Based on the initial causal strength, a multi-objective benefit function is defined as follows:

[0090] U i ={U i,acc U i,cost U i,time},

[0091] U i,acc =w 1 i,acc ·(1-err s )·CS s ,

[0092]

[0093] U i,time =w 3 i,time ·T s ,

[0094] err s =1-μ s,i ,

[0095] Among them, U i Let U be the multi-objective payoff vector for participant i. i,acc For participant i's prediction accuracy gain, U i,cost For participant i's cost-benefit, U i,time For participant i's time-sensitive benefits, w 1 i,acc w 2 i,cost and w 3 i,time These are the weights for the gains in prediction accuracy, cost, and timeliness, respectively. s Let C be the membership error of scene s. s For the standardized cost of scenario s (obtained by normalization based on market prices), T s Standardize the timeliness of scenario s (obtained based on delivery time normalization);

[0096] The initial weighted population is adjusted using the preliminary causal strength, as shown in the formula:

[0097] w ′ s =w s ·(1+β·CS s ),

[0098] Among them, w ′ s Adjust the causal weights for scene s, w sβ is the initial weight of scenario s (obtained based on the initial weight range), and β is the causal adjustment factor (set based on the volatility of real-time data);

[0099] Using the NSGA-II algorithm, based on a multi-objective payoff function and causal weight adjustment, the initial weight population is optimized to generate a dynamic Pareto front. A Bézier curve is then used for smoothing to select the optimal weights. The formula is as follows:

[0100] w″ s =argmax(∑ i U i ·μ s CS s ),

[0101] Among them, w″ s The optimal weights for scenario s;

[0102] Based on the optimal weights, the membership degree of each scenario, and the membership degree of the mixed scenario, the comprehensive scenario demand prediction is calculated using the following formula:

[0103] D′ t =∑ s∈S μ s ·w″ s ·[D s,min D s,max ],

[0104] Among them, D′ t The comprehensive scenario demand forecast interval for time t;

[0105] The optimal optimization weights of the comprehensive scenario demand forecast are sorted in descending order to generate a priority list; the membership degree of each scenario, the membership degree of the mixed scenario, the optimal optimization weight, the comprehensive scenario demand forecast, and the priority list are integrated into a comprehensive scenario demand forecast report.

[0106] By identifying time-series causal relationships between variables through statistical modeling methods, theoretical support is provided for predictive modeling. Granger analysis is more suitable for handling dynamically evolving causal mechanisms, improving the interpretability and foresight of the model. Smooth interpolation avoids the instability of population quality caused by discrete initialization, enhancing the population diversity and local optimization ability of the genetic algorithm. A multi-objective benefit function is constructed, refining the benefit structure of participants and considering prediction accuracy, cost control, and timeliness response respectively, reflecting the real interest preferences of the participants and making the optimization results realistically feasible. A causal adjustment factor is introduced for dynamic weight correction, which can flexibly adjust the model weights according to real-time data changes, making the prediction process environmentally adaptable and robust. NSGA-II is used to optimize the multi-objective benefit function and select the Pareto optimal solution to achieve a balanced optimization of the prediction scheme between accuracy, cost, and timeliness, avoiding the bias caused by single-objective optimization. A prediction report is generated by integrating membership degree and weight information, outputting a comprehensive scenario demand prediction report, covering optimal weights, scenario priorities, prediction intervals, etc., providing users with structured and operable decision-making basis.

[0107] S2. Use PSM to calculate propensity scores and generate a balanced dataset. Construct a directed acyclic graph using Gaussian distribution and Pearson correlation coefficient. Identify the set of parent nodes based on the MMHC algorithm. Optimize the network using Hill-Climbing. Calculate the outlier factor using the FDPC-OF method. Build and update the KD-Tree index. Use Drools for classification. Generate a structured anomaly report. Optimize the collaborative cost of the update strategy based on the P2P collaborative network. Output the optimized procurement strategy.

[0108] Specifically, a directed acyclic graph is constructed, optimized using Hill-Climbing, and the outlier factor is calculated using the FDPC-OF method, including:

[0109] Extract the comprehensive scenario demand forecast from the comprehensive scenario demand forecast report, combine it with market prices and supplier inventory, normalize it, and then map it to procurement time, procurement quantity and supplier selection through linear transformation.

[0110] PSM is used to calculate propensity scores and generate a balanced dataset. The formula is as follows:

[0111] PS = P(B = 1 | X),

[0112] Where PS is the propensity score, B is the scenario label (emergency scenarios are set to 1, and other scenarios are set to 0), X is the covariate, which represents the combined scenario demand forecast, market price, and supplier inventory, and P is the conditional probability, which represents the probability of the scenario occurring under the covariate.

[0113] The equilibrium dataset is defined as BN nodes, and the probability distribution of each node is calculated using a Gaussian distribution. The relationship between the equilibrium datasets is defined as an edge (e.g., market price affects demand). The Pearson correlation coefficient between nodes is calculated using the Pearson correlation coefficient formula, and then normalized and defined as the edge weight. A directed acyclic graph is constructed.

[0114] The MMHC algorithm is used to identify the parent node of each node, which is then concatenated into a set of parent nodes. The Hill-Climbing algorithm is used to optimize the directed acyclic graph and generate the optimal causal relationship network.

[0115] Based on the optimal causal relationship network, the conditional probability is calculated using the following formula:

[0116]

[0117] Among them, P 1 (H|Pa(H)) represents the conditional probability under the set Pa(H) of the parent node H, and P 2 Let P be a joint probability distribution. 3 The marginal probability of the parent node set;

[0118] A BN model was constructed, including an output layer and an input layer, and the model was trained using power grid reference data provided by IEEE.

[0119] We define procurement time, procurement quantity, and supplier selection as dimensions of the strategy space. Based on conditional probabilities, we calculate the conditional probabilities of strategies for each dimension to generate an initial strategy population. The formula is as follows:

[0120] P″(T ′ Q ′ ,V′∣D′ t C s ) = P 1 (V′∣D′ t C s )·P 1 (T ′ D′ t ,V′)·P 1 (Q ′ |D′ t ,V′),

[0121] Where P” is the policy conditional probability, T ′ For procurement time, Q ′ V′ represents the purchase volume, and V′ represents the supplier selection.

[0122] Based on BN nodes, using FDPC-OF, the local density between each node, the centripetal relative distance between each node, and the outlier factor are calculated through K nearest neighbors. The formula is as follows:

[0123]

[0124] Where, ρ I Let be the local density of node I. Let be the Euclidean distance between node I and node J (calculated based on the Euclidean formula). The cutoff distance (automatically calculated by FDPC-OF), δ I Let OF be the centripetal relative distance of node I. I Let N be the outlier factor of node I, kl be the number of K nearest neighbors, and N be the number of K nearest neighbors. kl (I) is the set of K nearest neighbors of node I (obtained based on K nearest neighbors);

[0125] Nodes with outlier factors greater than the outlier factor threshold (based on experimental analysis) are filtered out, marked as anomalies, and a list of anomaly events is output.

[0126] By normalizing data and performing linear transformation mapping, raw data of different dimensions can be compared and modeled on a unified scale, enhancing the interpretability and modeling accuracy of relationships between variables. Pearson coefficients are used to establish correlations between variables, and the optimal causal structure is obtained by combining MMHC and Hill-Climbing algorithms, significantly improving the model's ability to identify deep dependencies between variables. Conditional probability modeling of variables such as procurement time, batch size, and supplier selection generates a strategy population that conforms to historical data logic, providing prior distribution support for multi-objective optimization. Local density and K-nearest neighbor information are combined to accurately identify nodes that do not conform to the overall distribution pattern, improving the sensitivity and accuracy of anomaly detection. Dynamic thresholds set in the experiment are used to accurately screen detected high outlier nodes, avoiding false alarms and improving reliability in actual deployment.

[0127] Furthermore, a KD-Tree index is constructed and updated, Drools is used for classification, and optimization is performed based on a P2P collaborative network, including:

[0128] A KD-Tree index is built based on BN nodes. The median of the node with the largest variance (the variance of each node is calculated and sorted in descending order) is set as the split point, and the high-dimensional space is divided into a binary tree structure.

[0129] Based on BN nodes, K-nearest neighbors are used in KD-Tree to obtain the updated K-nearest neighbor set of BN nodes, recalculate and update the outlier factor, filter and update the list of anomalous events;

[0130] Based on the updated outlier factor, the outlier priority is calculated using the following formula:

[0131]

[0132] Among them, PriI Let W be the outlier priority of node I, and W be the outlier factor weight (adjusted based on historical data). 1 Ran is used to influence the range of weights. I The influence range of node I is obtained by collecting historical anomaly data through the API interface, extracting the minimum value of the maximum influence range, and then normalizing it. Let be the updated outlier factor for node I;

[0133] Use Drools to categorize the anomaly types in the update anomaly event list (including price anomalies, inventory anomalies, and demand anomalies) and generate category labels;

[0134] Using NLG, exception types and exception priorities are integrated into a structured exception report;

[0135] Based on the structured anomaly report, the conditional probability of the BN node is updated, and the edge weights are optimized using the Hill-Climbing algorithm, with the following formula:

[0136]

[0137] Among them, PO(X) i |Pa(X) i A) represents the update conditional probability of the parent node H set Pa(H) and the exception type A.

[0138] Based on the updated BN nodes, calculate the conditional probability of the update policy and generate an update policy population, using the following formula:

[0139]

[0140] Where PA is the conditional probability of the update policy;

[0141] Based on the updated strategy population and structured anomaly reports, a P2P collaborative network among suppliers is constructed to optimize collaborative costs, as shown in the formula:

[0142] Cξ SE =∑Θ,ξm Θ→ ξ·Γ Θ→ξ ,

[0143] Among them, CB SE To update the P2P collaboration cost of strategy SE, m Θ→ξ The transportation cost from supplier Θ to supplier ξ (obtained by retrieving the average transportation cost records of historical purchase orders from the ERP system), Γ Θ→ξ The shared quantity of materials from supplier Θ to supplier ξ (based on supplier inventory);

[0144] By optimizing costs and updating strategy populations, combined with structured anomaly reports, we can integrate them into an optimized procurement strategy.

[0145] By introducing a spatial indexing mechanism into high-dimensional supply chain data, subsequent query efficiency is significantly improved. Variance, as the partitioning criterion, has strong discriminative properties, ensuring that the partitioning preserves the data distribution characteristics to the greatest extent. Outlier factors are calculated using the updated KNN set, enabling the capture of abnormal dynamic changes caused by environmental or node state variations, avoiding misjudgments caused by static models. Combined with the conditional probability adjustment mechanism of BN nodes, real-time and refined anomaly detection is achieved. Anomaly priority is calculated based on outlier factors and impact range, enabling quantitative assessment of anomaly severity. Classified anomaly events can be quickly identified and handled by system or operations personnel. Through a dynamically updated rule base, Drools adapts to actual business changes, achieving maintainability and scalability of the anomaly detection mechanism. By leveraging NLG to generate standardized reports, not only is human-machine collaboration efficiency improved, but operational inputs are also provided for subsequent conditional probability learning based on Hill-Climbing. This enables feedback optimization between anomalies and causal chains. The Hill-Climbing algorithm is used for local optimum search, and the BN structure is dynamically adjusted based on anomaly type and frequency to improve inference. The optimized strategy probability and anomaly information are input into a P2P collaborative network. Combined with historical transportation costs and shared inventory data extracted from ERP, the allocation of materials and transportation costs among suppliers are optimized. The optimized cost structure, strategy population, and anomaly reports are integrated to generate interpretable and executable procurement strategies, empowering enterprise intelligent decision-making systems and enhancing supply chain resilience and risk resistance.

[0146] S3. Execute the optimized procurement strategy and store it in the database;

[0147] Specifically, the optimized procurement strategy is executed and stored in the database, including:

[0148] The optimized procurement strategy is broken down into execution tasks through a message queue service, distributed via API, and indexed using Elasticsearch. The optimized procurement strategy and demand forecast report are then stored in the database.

[0149] By deconstructing procurement strategies into schedulable tasks, the system standardizes and automates task processes, supports parallel task processing, improves response speed, reduces the risk of human intervention in strategy execution, issues tasks through API interfaces to improve system collaboration efficiency, facilitates cross-system process management, uses Elasticsearch to build data indexes to improve data retrieval efficiency and accuracy, supports multi-dimensional data analysis and visualization, stores structured optimized procurement strategies and demand forecast reports in the database, supports historical data tracking and regression testing, and realizes the accumulation and reuse of data assets.

[0150] This embodiment also provides a procurement strategy optimization system based on power material demand data, including:

[0151] The data collection and forecasting module is used to collect electricity data and third-party demands, preprocess the data, construct mapping rules, perform mapping through a rule engine, perform fuzzy modeling using spherical fuzzy values, calculate the comprehensive membership degree of the scenario using Sugeno-Weber weighted average, construct hybrid scenarios based on fuzzy rules, generate comprehensive scenario demand forecast intervals, determine preliminary causal strength based on Granger causality analysis, generate an initial weight population using a search window and Bézier curve, construct a multi-objective benefit function and combine it with NSGA-II to optimize the weights, generate a smooth Pareto front, select the optimal optimization weights, calculate the comprehensive scenario demand forecast, prioritize the forecasts, and generate a comprehensive scenario demand forecast report.

[0152] The detection optimization module is used to generate a balanced dataset by calculating propensity scores using PSM, construct a directed acyclic graph using Gaussian distribution and Pearson correlation coefficient, identify the set of parent nodes based on the MMHC algorithm, generate the optimal causal network through Hill-Climbing optimization, calculate the outlier factor using the FDPC-OF method, build and update the KD-Tree index, classify using Drools, generate a structured anomaly report, optimize the collaborative cost of the update strategy based on the P2P collaborative network, and output the optimized procurement strategy.

[0153] The execution storage module is used to execute optimized procurement strategies and store them in the database.

[0154] This embodiment also provides a computer device applicable to the procurement strategy optimization method based on power material demand data, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the procurement strategy optimization method based on power material demand data proposed in the above embodiment.

[0155] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0156] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the procurement strategy optimization method based on power material demand data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0157] In summary, this invention uses spherical fuzzy values ​​for fuzzy modeling, determines preliminary causal strength based on Granger causality analysis, generates an initial weight population using a search window and Bézier curves, optimizes weights using NSGA-II, identifies the parent node set based on the MMHC algorithm, calculates outlier factors using the FDPC-OF method, constructs a KD-Tree index for updating, and uses Drools for classification; thus improving the accuracy of demand forecasting and the scientific nature of priority ranking; and enhancing the stability and adjustability of procurement strategies.

[0158] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A procurement strategy optimization method based on power material demand data, characterized in that: include, Collect power data and third-party demands, preprocess them, construct mapping rules, perform mapping through a rule engine, perform fuzzy modeling using spherical fuzzy values, calculate the comprehensive membership degree of the scenario using Sugeno-Weber weighted average, construct hybrid scenarios based on fuzzy rules, generate comprehensive scenario demand prediction intervals, determine preliminary causal strength based on Granger causality analysis, generate an initial weight population using search windows and Bézier curves, construct a multi-objective benefit function and combine it with NSGA-II to optimize weights, generate a smooth Pareto front, select the optimal optimization weights, calculate comprehensive scenario demand predictions and prioritize them, and generate a comprehensive scenario demand prediction report. The PSM is used to calculate propensity scores to generate a balanced dataset. A directed acyclic graph is constructed using Gaussian distribution and Pearson correlation coefficient. The set of parent nodes is identified based on the MMHC algorithm. The optimal causal network is generated through Hill-Climbing optimization. The outlier factor is calculated using the FDPC-OF method. A KD-Tree index is constructed for updating. Drools is used for classification to generate a structured anomaly report. The collaborative cost of the update strategy is optimized based on the P2P collaborative network, and the optimized procurement strategy is output. Implement optimized procurement strategies and store them in the database; The collection of power data and the demands of the three parties refer to using API interfaces to collect power data and extract power data features, using API interfaces to collect the demands of power grid operators, suppliers and procurement departments, and constructing a multi-party interest coordination matrix. The constructed mapping rules are mapped using a rule engine, fuzzy modeling is performed using spherical fuzzy values, and the comprehensive membership degree of the scene is calculated using Sugeno-Weber weighted average. Based on the fuzzy rules, a hybrid scene is constructed, generating a comprehensive scene demand prediction interval, including: Based on the characteristics of power data, mapping rules are constructed to map the power data characteristics to the corresponding scenarios through the rule engine, generating a scenario feature matrix. Spherical fuzzy values ​​are calculated for the power data features of each scenario, and the power data features of each scenario are fused using Sugeno-Weber weighted average to calculate the comprehensive membership degree of the scenario; fuzzy rules are defined based on the power data features to calculate the membership degree of the mixed scenario, and combined with the scenario feature matrix to generate a comprehensive scenario demand prediction interval; The process involves determining preliminary causal strength based on Granger causality analysis, generating an initial weight population using a search window and Bézier curves, constructing a multi-objective payoff function and optimizing the weights using NSGA-II, generating a smooth Pareto front, selecting the optimal optimized weights, calculating and prioritizing comprehensive scenario demand predictions, and generating a comprehensive scenario demand prediction report, including: Based on the multi-party interest coordination matrix, Granger causal analysis is used to calculate the preliminary causal strength. Combined with the comprehensive scenario demand prediction interval, the initial weight range is calculated. An initial weight population is generated through a search window and smoothed using a Bézier curve. Based on the initial causal strength, a multi-objective benefit function is defined; The initial weight population is adjusted using preliminary causal strength. The initial weight population is then optimized using the NSGA-II algorithm based on a multi-objective benefit function and causal adjustment weights to generate a dynamic Pareto front. The front is then smoothed using a Bézier curve. The optimal optimized weights are selected, and the comprehensive scenario demand prediction is calculated. The optimal optimized weights are sorted in descending order to generate a priority list. The membership degree of each scenario, the membership degree of the mixed scenario, the optimal optimized weights, and the comprehensive scenario demand prediction are combined to form a comprehensive scenario demand prediction report. The process involves constructing a directed acyclic graph using Gaussian distribution and Pearson correlation coefficient, identifying the set of parent nodes based on the MMHC algorithm, generating an optimal causal network through Hill-Climbing optimization, and calculating the outlier factor using the FDPC-OF method, including: Extract the comprehensive scenario demand forecast from the comprehensive scenario demand forecast report, combine it with market prices and supplier inventory, normalize it, and then map it to procurement time, procurement quantity and supplier selection through linear transformation. The PSM is used to calculate the propensity score, generate a balanced dataset, define the balanced dataset as BN nodes, calculate the probability distribution of each node using Gaussian distribution, define the relationship between the balanced datasets as edges, calculate the Pearson correlation coefficient between nodes using the Pearson correlation coefficient formula, normalize it, define it as edge weight, and construct a directed acyclic graph. The MMHC algorithm is used to identify the parent node of each node, which is then concatenated into a set of parent nodes. The Hill-Climbing algorithm is used to optimize the directed acyclic graph and generate the optimal causal relationship network. Based on the optimal causal relationship network, the conditional probability is calculated, and the procurement time, procurement batch and supplier selection are defined as the strategy space dimensions. Based on the conditional probability, the policy conditional probability of each dimension is calculated to generate the initial policy population. Based on BN nodes, using FDPC-OF, we calculate the local density between each node, the centripetal relative distance between each node, and the outlier factor through K nearest neighbors; Filter nodes whose outlier factor is greater than the outlier factor threshold, mark them as anomalies, and output a list of anomaly events; The process involves constructing and updating a KD-Tree index, classifying data using Drools, generating structured anomaly reports, optimizing the collaborative cost of the update strategy based on a P2P collaborative network, and outputting an optimized procurement strategy, including: A KD-Tree index is built based on BN nodes. K-nearest neighbors are used in the KD-Tree to recalculate and update the outlier factor, update the list of anomalous events, and calculate the anomaly priority based on the updated outlier factor. Drools is used to classify the anomaly types in the update anomaly event list, generate classification labels, integrate anomaly types and anomaly priorities into a structured anomaly report, update the conditional probability of BN nodes based on the structured anomaly report, optimize edge weights using the Hill-Climbing algorithm, calculate the conditional probability of update strategies based on the updated BN nodes, and generate an update strategy population. A P2P collaboration network among suppliers is built based on updated strategy populations and structured anomaly reports to optimize collaboration costs; By combining cost optimization and strategy population updates with structured anomaly reports, we can integrate them into an optimized procurement strategy. The comprehensive scenario demand prediction range is expressed as follows: ， In the formula, and Scenes The lower and upper limits of the comprehensive scenario demand prediction range. Features The weight, and Features lower bound and characteristics of the interval The upper limit of the interval, Features For the scene membership degree The number of features; The initial weight range is calculated using the following formula: ， in, and Scenes The upper and lower limits of the weight interval, For the scene The initial causal strength; Define a multi-objective benefit function, as follows: ， ， ， ， ， in, For participants The multi-objective benefit vector. For participants Benefits from prediction accuracy For participants Cost and benefit, For participants Time-sensitive benefits , as well as The weights for the benefits of prediction accuracy, cost benefits, and timeliness benefits are respectively. For the scene Membership error For the scene Standardized costs, For the scene Standardization and timeliness; The initial weighted population is adjusted using the preliminary causal strength, as shown in the formula: ， in, For the scene Causal adjustment weights, For the scene The initial weights, As a causal adjustment factor; The formula for selecting the optimal weight is: ， in, For the scene The optimal weight; The formula for calculating the demand forecast for the overall scenario is: ， in, For time The comprehensive scenario demand prediction range, This represents the number of scenes.

2. The procurement strategy optimization method based on power material demand data as described in claim 1, characterized in that: The execution of the optimized procurement strategy and its storage in the database includes: The optimized procurement strategy is broken down into execution tasks through a message queue service, distributed via API, and indexed using Elasticsearch. The optimized procurement strategy and demand forecast report are then stored in the database.

3. A procurement strategy optimization system based on power material demand data, based on the procurement strategy optimization method based on power material demand data as described in any one of claims 1 to 2, characterized in that: include, The data collection and forecasting module is used to collect electricity data and third-party demands, preprocess the data, construct mapping rules, perform mapping through a rule engine, perform fuzzy modeling using spherical fuzzy values, calculate the comprehensive membership degree of the scenario using Sugeno-Weber weighted average, construct hybrid scenarios based on fuzzy rules, generate comprehensive scenario demand forecast intervals, determine preliminary causal strength based on Granger causality analysis, generate an initial weight population using a search window and Bézier curve, construct a multi-objective benefit function and combine it with NSGA-II to optimize the weights, generate a smooth Pareto front, select the optimal optimization weights, calculate the comprehensive scenario demand forecast, prioritize the forecasts, and generate a comprehensive scenario demand forecast report. The detection optimization module is used to generate a balanced dataset by calculating propensity scores using PSM, construct a directed acyclic graph using Gaussian distribution and Pearson correlation coefficient, identify the set of parent nodes based on the MMHC algorithm, generate the optimal causal network through Hill-Climbing optimization, calculate the outlier factor using the FDPC-OF method, build and update the KD-Tree index, classify using Drools, generate a structured anomaly report, optimize the collaborative cost of the update strategy based on the P2P collaborative network, and output the optimized procurement strategy. The execution storage module is used to execute optimized procurement strategies and store them in the database.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the procurement strategy optimization method based on power material demand data as described in any one of claims 1 to 2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the procurement strategy optimization method based on power material demand data as described in any one of claims 1 to 2.

Images