Electric power equipment purchase bidding scheme evaluation method, device and equipment

Abstract

The invention discloses an electric power equipment purchase bidding scheme evaluation method, device and equipment, and relates to the technical field of electric power equipment bidding purchase evaluation. The method comprises the following steps: collecting historical bidding schemes and a plurality of evaluation index data of each bidder to form a local data set; selecting a local model and a global model in the asynchronous federated learning framework; on the basis of an asynchronous federated learning framework, asynchronously training a local model by using respective local data sets at each bidding party, training a global model by adopting a dynamic aggregation strategy according to parameters of the trained local model, and determining the trained global model as a federated model; and respectively inputting the current bidding scheme into the federal model, outputting a plurality of evaluation index predicted values of the current bidding scheme, determining a comprehensive evaluation result of the current bidding scheme of each bidding party, and further generating an evaluation report of the current bidding scheme of each bidding party. According to the method, the problem of data privacy protection in a traditional evaluation method can be solved, and meanwhile, the evaluation accuracy is improved.

Description

Technical Field

[0001] The present application relates to the technical field of power equipment bidding and procurement evaluation, and in particular to a method, device and equipment for evaluating power equipment procurement bidding plans. Background Art

[0002] Power equipment procurement often involves proposals submitted by multiple bidders, which require detailed evaluation to select the optimal supplier and solution. Traditional evaluation methods often rely on subjective judgment by reviewers and limited data analysis, which can be subject to bias and inaccuracy.

[0003] Machine learning techniques can be used to process and analyze large amounts of historical data, extracting useful insights and improving the evaluation process. However, traditional machine learning methods often require centralized data storage, which can raise data privacy and security issues. In the power equipment procurement process, traditional data processing methods cannot adequately address data privacy issues, especially when multiple entities and sensitive information are involved. The sensitive data of bidders and purchasers needs to be protected to prevent leakage and misuse. Summary of the Invention

[0004] The purpose of this application is to provide a method, device and equipment for evaluating bidding plans for power equipment procurement, which can solve the data privacy protection problem in traditional evaluation methods and improve the accuracy of evaluation.

[0005] To achieve the above objectives, this application provides the following solutions.

[0006] In the first aspect, the present application provides a method for evaluating bidding schemes for power equipment procurement, including: collecting historical bidding schemes of each bidder and multiple evaluation index data of each historical bidding scheme, and forming a local data set of each bidder; selecting a machine learning model as a local model and a global model in an asynchronous federated learning framework; based on the asynchronous federated learning framework, each bidder uses its own local data set to asynchronously train the local model, and adopts a dynamic aggregation strategy to train the global model based on the parameters of the trained local model, and determines the trained global model as the federated model; wherein the dynamic aggregation strategy includes the weight allocation of multi-supplier collaborative data in the joint bidding scenario, and the feature fusion parameters of multi-department data within the same bidder; the current bidding scheme of each bidder is input into the federated model respectively, and multiple evaluation index prediction values of the current bidding scheme of each bidder are output; based on the multiple evaluation index prediction values, a comprehensive evaluation result of the current bidding scheme of each bidder is determined; based on the comprehensive evaluation result, an evaluation report of the current bidding scheme of each bidder is generated.

[0007] Optionally, select a machine learning model, specifically including: if the prediction result is a numerical value, select a regression model; if it is used for prediction classification, select a classification model; if the size of the local data set is larger than a preset threshold, or features need to be extracted, select a deep learning model; if the data distribution of the local data set is nonlinear, select a nonlinear model.

[0008] Optionally, based on the predicted values of multiple evaluation indicators, the comprehensive evaluation results of the current bidding proposals of each bidder are determined, specifically including: based on the predicted values of multiple evaluation indicators, using a weighting method or a multi-objective decision-making method to obtain a comprehensive evaluation score of the current bidding proposals of each bidder.

[0009] Optionally, the evaluation report includes: comprehensive evaluation score, technical advantages and quotation rationality.

[0010] Optionally, the method for generating technical advantages includes: obtaining technical documents in the current bidding proposal of each bidder; extracting technical parameters in the technical documents and standardizing them into technical feature vectors; constructing an evaluation model based on a federated neural network; inputting the technical feature vectors into the evaluation model based on a federated neural network, and outputting the technical score of the current bidding proposal of each bidder; constructing a knowledge graph of power equipment based on the industry standards and historical optimal technical parameters of power equipment; based on the knowledge graph, using a graph neural network to compare the current bidding proposal of each bidder with the technical benchmark, and identifying the advantages of the current bidding proposal of each bidder.

[0011] Optionally, the method for generating the rationality of the quotation includes: using a trained regression model to predict a reasonable price range for the power equipment based on historical quotations of multiple bidders for the power equipment; comparing the quotation in the current bidding plan of each bidder with the reasonable price range to determine whether the quotation in the current bidding plan of each bidder is reasonable.

[0012] Optionally, when there is real-time market data, the method for generating the rationality of the quotation includes: obtaining the historical market price and external dynamic data of the power equipment, and constructing a multi-factor time series feature vector; the external dynamic data includes the power industry commodity price index, raw material price curve, international exchange rate and transportation cost; according to the multi-factor time series feature vector, the multi-factor time series model is federated learned to obtain a multi-factor time series prediction model; according to the current multi-factor time series feature vector, the multi-factor time series prediction model is used to dynamically predict the price of the power equipment at the time of delivery; according to the dynamically predicted price of the power equipment at the time of delivery and the current market price of the power equipment, the formula is used to Calculate the base price; where P bace is the base price, α is the weight coefficient, is the dynamically predicted price of power equipment at delivery, P marketis the real-time market price of the power equipment; according to the benchmark price, using the formula [P min , P max ]=[P bace (1-β), P bace ·(1+γ)], determine the reasonable price range of power equipment in real time; where, P min The lowest price in the reasonable price range, P max is the highest price in the reasonable price range, β is the price downward threshold, and γ is the price upward threshold; based on the reasonable price threshold, the isolation forest algorithm or the local outlier factor algorithm is used to compare the quotation of the current bidding scheme of each bidder with the real-time reasonable price range; if the quotation falls within the real-time reasonable price range, the quotation rationality of the current bidding scheme is reasonable; if the quotation does not fall within the real-time reasonable price range, the quotation rationality of the current bidding scheme is unreasonable.

[0013] Optionally, the method further includes: introducing a partner evaluation sub-model into the asynchronous federated learning framework to analyze the historical cooperation data of the joint bidders and output a cooperation risk score; the calculation formula of the cooperation risk score is: Where R is the cooperation risk score, ω i is the risk weight of the i-th partner, s i is the historical cooperation success rate, c i is the real-time credit indicator, f is the normalization function, and n is the number of partners.

[0014] Optionally, the method also includes: within the same bidder, dividing the local data sets of the technical department and the business department, performing model training through a multi-node federated learning framework within the enterprise, and using differential privacy technology to encrypt the gradient parameters between departments to obtain a department collaborative evaluation model; the local data set of the technical department includes power equipment performance parameters and technical scores; the local data set of the business department includes power equipment quotation data and quotation rationality scores; based on the power equipment performance parameters and power equipment quotation data of the same bidder in the current bidding plan, using the department collaborative evaluation model to generate technical scores, quotation rationality scores and department collaborative comprehensive scores, and generate a department collaborative evaluation report.

[0015] Optionally, the method also includes: during the equipment delivery and operation and maintenance stage of the successful bidder, continuously collecting the acceptance data, performance indicators and maintenance costs of the power equipment to form an incremental data set; based on the asynchronous federated learning framework, inputting the incremental data set into the asynchronously trained local model for re-training, updating the parameters of the federated model, and obtaining an updated federated model; according to the updated federated model, obtaining the reliability score of the power equipment and generating a full life cycle performance evaluation report.

[0016] In a second aspect, the present application provides a device for evaluating bidding schemes for power equipment procurement, comprising:

[0017] A collection module is used to collect historical bidding proposals of each bidder and multiple evaluation index data of each historical bidding proposal, and form a local data set of each bidder;

[0018] The selection module is used to select machine learning models as local models and global models in the asynchronous federated learning framework;

[0019] The training module is used to asynchronously train local models at each bidder using their own local datasets based on an asynchronous federated learning framework. Based on the trained local model parameters, a dynamic aggregation strategy is used to train a global model, which is then used as the federated model. The dynamic aggregation strategy includes weighting the collaborative data of multiple suppliers in a joint bidding scenario, as well as feature fusion parameters for data from multiple departments within the same bidder.

[0020] An application module, configured to input the current bidding proposals of each bidder into the federated model and output predicted values of multiple evaluation indicators of the current bidding proposals of each bidder;

[0021] A comprehensive evaluation module is used to determine the comprehensive evaluation results of each bidder's current bidding plan based on the predicted values of multiple evaluation indicators;

[0022] The optimization suggestion production module is used to generate an evaluation report of the current bidding scheme of each bidder based on the comprehensive evaluation results.

[0023] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned methods for evaluating bid proposals for procurement of electric power equipment.

[0024] According to the specific embodiments provided in this application, this application has the following technical effects.

[0025] This application provides a method, device and equipment for evaluating bid proposals for power equipment procurement. By utilizing a federated learning framework, different data holders are allowed to jointly train models without sharing original data, thereby protecting the data privacy of each data holder. At the same time, the federated learning framework can effectively integrate information from multiple parties, improve the comprehensiveness and accuracy of bid proposal evaluation, reduce the subjectivity and bias caused by relying entirely on evaluation experts during the evaluation process, and improve the accuracy of the evaluation.

[0026] This application expands from a single bidding evaluation to the full process management of power equipment procurement, operation and maintenance through full life cycle data feedback and dynamic model updates, helping bidders to continuously optimize products and services and providing long-term decision-making support for purchasers. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 A flowchart of a method for evaluating a bidding scheme for power equipment procurement is provided in one embodiment of the present application.

[0029] Figure 2 A schematic diagram of the data sharing process based on horizontal federated learning provided in another embodiment of the present application.

[0030] Figure 3 A schematic diagram of the implementation flow of a method for evaluating a power transformer procurement bidding scheme provided in another embodiment of the present application.

[0031] Figure 4 A schematic diagram of the structure of a power equipment procurement bidding scheme evaluation device provided in one embodiment of the present application.

[0032] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0034] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0035] Power equipment procurement refers to the purchase of various power equipment and technical services required by power companies or related organizations during the construction, maintenance, and upgrade of power systems. Power equipment procurement encompasses the entire process, from determining requirements, selecting suppliers, signing contracts, to equipment delivery and acceptance.

[0036] In an exemplary embodiment, Figure 1 As shown, a method for evaluating a bidding scheme for power equipment procurement is provided, including the following steps 101 to 106.

[0037] Step 101: Collect historical bidding proposals of each bidder and multiple evaluation index data of each historical bidding proposal, and form a local data set of each bidder.

[0038] Step 102: Select a machine learning model as the local model and the global model in the asynchronous federated learning framework.

[0039] Step 103: Based on the asynchronous federated learning framework, each bidder uses its own local data set to asynchronously train a local model, and uses a dynamic aggregation strategy to train a global model based on the trained local model parameters, and determines the trained global model as the federated model; wherein the dynamic aggregation strategy includes the weight allocation of multi-supplier collaborative data in the joint bidding scenario, and the feature fusion parameters of multi-department data within the same bidder.

[0040] Step 104: Input the current bidding proposals of each bidder into the federated model respectively, and output the predicted values of multiple evaluation indicators of the current bidding proposals of each bidder.

[0041] Step 105: Determine the comprehensive evaluation result of each bidder's current bidding proposal based on the predicted values of multiple evaluation indicators.

[0042] Step 106: Generate an evaluation report of each bidder's current bidding proposal based on the comprehensive evaluation results.

[0043] By implementing steps 101 through 106 above and jointly modeling data from multiple parties through federated learning, the accuracy and reliability of bid evaluation are improved. Evaluating bids based on federated learning and machine learning techniques overcomes the limitations of traditional power equipment procurement evaluation methods and meets data privacy requirements.

[0044] In another exemplary embodiment of the present application, step 101 collects historical bidding proposals and evaluation index data of each bidder, where the evaluation index data refers to evaluation data related to power equipment procurement. Purchase demand data of the purchaser may also be collected.

[0045] Preprocess the collected data to ensure data quality and consistency for subsequent model training and evaluation. Preprocessing includes cleaning, normalization, and standardization.

[0046] In another exemplary embodiment of the present application, an appropriate machine learning model (such as a regression model, a classification model, a deep learning model, etc.) is selected and designed according to the characteristics of power equipment procurement and the attributes of the data. Among them, according to the characteristics of the power equipment procurement target means: if the goal is to predict continuous values, such as the price, procurement cost or demand of power equipment, a regression model should be selected. If the goal is classification, such as judging the reliability of a supplier or the status of equipment, a classification model should be selected. For complex data relationships, large-scale data sets or the need to process unstructured data, a deep learning model is selected. According to data attributes means: large data sets are suitable for deep learning models, and small data sets are suitable for traditional machine learning models. Text or image data may require a deep learning model to extract features. If the data distribution is complex or nonlinear, a nonlinear model is required, such as a decision tree, support vector machine or neural network.

[0047] In another exemplary embodiment of the present application, federated learning is a distributed machine learning framework consisting of a federated server and multiple participants. Each participant holds its own local data. During the process, the data of each participant will not leave the local area of the participant. The federated model undergoes multiple rounds of training until it terminates after meeting the requirements. Using federated learning technology, model training is performed locally on each data holder. Through multiple rounds of iterations, the model parameters obtained by training by all parties are merged and the global model is updated. This ensures that the model fully utilizes decentralized data for training while protecting data privacy. Data sharing based on horizontal federated learning includes supply chain platforms and participants, and the data sharing process is as follows: Figure 2 As shown, it includes: initiating horizontal federated data sharing requirements → issuing global parameter models → each training parameter model locally → encrypting parameter models and sending encrypted gradients → securely aggregating models → sending aggregated encrypted gradients → decrypting gradients and updating them locally to obtain global model data.

[0048] In the asynchronous federated learning mechanism, each bidder immediately uploads parameters after completing local training, without waiting for other nodes. Upon receiving each node's parameters, the server immediately updates the global model and distributes them to online nodes. This mechanism addresses network latency and uneven data distribution among bidders in power procurement, improving the model's adaptability and real-time performance in dynamic environments.

[0049] A specific implementation process of the above step 103 may include federated learning environment configuration and model training.

[0050] (1) Federated learning environment configuration.

[0051] Participant setup: Identify the data holders participating in federated learning (for example, bidders, purchasers, etc.) and configure the corresponding computing resources and software environment.

[0052] Privacy protection technology configuration: Ensure that the federated learning framework supports privacy protection technologies, such as homomorphic encryption or differential privacy, to ensure data security.

[0053] (2) Model training.

[0054] Local model training: Each data holder independently and asynchronously performs local model training.

[0055] If the data holder is a joint bidder or an internal department of an enterprise, the following steps are further performed:

[0056] Joint bidding entity: Divide each partner's local dataset into independent subsets, train sub-models separately, aggregate sub-model parameters through the federated server, and assign weights to the collaborative data of multiple suppliers in the joint bidding scenario to generate a global risk assessment model for the joint bidding entity.

[0057] Internal enterprise departments: Noise is added to the data of the technical department and the business department respectively, and department-level local models are trained. The models are uploaded to the enterprise federated server through encrypted gradients, and the parameters of the features of the data from multiple departments within the enterprise are integrated to generate a cross-departmental collaborative optimization model.

[0058] Each participant updates the model using local data based on the global model parameters currently held locally, and submits the parameter update values after training is completed, without the need to keep in sync with other participants.

[0059] Global Model Update: The central server continuously receives and processes model updates from all participants in real time. Each update is immediately integrated with the current global model to form a new version of the global model. Updated parameters are dynamically distributed to all participants, providing real-time feedback to the node submitting the update or synchronizing to other nodes via periodic broadcasts or on-demand pull requests, supporting continuous iteration within heterogeneous training schedules.

[0060] The fusion mentioned in the previous paragraph, through dynamic weight allocation, time decay, and version control, effectively addresses network latency and uneven data distribution in power procurement while ensuring model convergence. Dynamic weighted aggregation: Nodes with large data volumes have a greater impact on the global model, and parameters with longer delays have lower weights. Incremental real-time updates: Each time a node parameter is received, it is immediately linearly superimposed with the current model according to its weight. Version control: Nodes submit parameters with the model version number, and outdated versions are discarded to prevent global model contamination.

[0061] In another exemplary embodiment of this application, a trained federated model is used to evaluate current bid proposals. The model predicts the pros and cons of each bid proposal and generates evaluation results, namely, predicted values for multiple evaluation indicators for each bidder's current bid proposal. These evaluation results reflect the performance of the bid proposal based on historical data and current conditions.

[0062] In another exemplary embodiment of the present application, a weighting method or a multi-objective decision-making method is used based on the predicted values of multiple evaluation indicators to obtain a comprehensive evaluation score of the current bidding proposal of each bidder.

[0063] The key algorithms involved in the above embodiments include model training and updating in federated learning, as well as bidding scheme evaluation.

[0064] (1) Model training and updating in federated learning.

[0065] Assume that each participant in the power equipment bidding (P1, P2, ..., P n ) each holds a local dataset (D1, D2, ..., D n ). Where D1 represents the local dataset held by the first participant P1, D2 represents the local dataset held by the first participant P2, and D n Represents the first participant P n The local dataset held, the subscript n indicates the number of participants.

[0066] 1) Federated learning algorithm: Use asynchronous federated learning algorithm.

[0067] Global model initialization: Initialize the global model parameters w0 on the federated server.

[0068] Local model initialization: Each participant copies the global model parameter w0 as the initial parameter of the local model.

[0069] 2) Federal training process.

[0070] Local training: Each participant P i Using its local dataset D i For the model parameter w i Perform E rounds of local training and update the local model parameters. The update formula is: Where η is the learning rate, L(·) is the loss function, is the gradient operator, is the model parameter obtained by the i-th participant in the t-th round of local training, P is the model parameter obtained by the i-th participant in the t+1 round of local training, t≤E. i represents the i-th participant, D i Represents the i-th participant Pi The local dataset held, w i represents the model parameters of the i-th participant.

[0071] Parameter upload: Each participant uploads the updated parameters of the local model The data is sent to the federation server without waiting for other nodes. The server continuously monitors parameter updates, allowing different participants to submit parameters at different frequencies.

[0072] Dynamic aggregation strategy: The federated server dynamically adjusts the aggregation weight based on the data volume, training timestamp, and network status of the participants. The global model update formula is:

[0073] in:

[0074] |D i | represents the i-th participant P i The size of the local dataset.

[0075] α i is the dynamic weight coefficient of the i-th participant, and the calculation formula is:

[0076]

[0077] Δt i is the training delay time of the i-th participant (current timestamp minus parameter generation timestamp); λ is the decay coefficient (the default value is 0.1), which is used to control the impact of delay on weight; β(v i ,v golbal ) is the version consistency factor.

[0078] Dynamic node management: When a new bidder joins, the federated server initializes its local model to the current global model and assigns a unique identifier. When an existing bidder exits due to a network outage or voluntarily, the federated server automatically removes its unfinished parameter updates and records the exit timestamp.

[0079] Global model delivery: After the federated server dynamically aggregates parameters and updates the global model, it delivers the latest model in the following ways:

[0080] Active push: Push the updated global model to all online participants in real time;

[0081] On-demand pull: Participants actively pull the latest model before the next round of training begins.

[0082] Version consistency guarantee: Participants save the currently used global model version number locally, and the server records the global model version number. Participants must attach the version number when uploading parameters. If the version numbers are inconsistent, the server refuses to aggregate or requires the participants to synchronize the latest model first.

[0083] Iteration: Repeat the above steps until the global model converges or the preset number of iterations T is reached.

[0084] Algorithm output: The converged global model w*, which is used to evaluate new bidding solutions.

[0085] (2) Evaluation of bidding proposals.

[0086] After the global model w* converges, the evaluation of new bidding proposals can be carried out. This process mainly depends on the global model's predictive ability for the new proposal and the calculation of evaluation indicators.

[0087] Input: Converged global model w*, and feature vector X of the new bidding solution new .

[0088] Plan prediction: Use the global model w* to predict the new bidding plan X new Make predictions and output evaluation indicators Such as cost, risk, performance score, etc. Among them, f represents the prediction function.

[0089] Evaluation indicator calculation: Based on the evaluation indicator Calculate the comprehensive evaluation score S new ,Common evaluation indicators may include: economic, feasibility and risk.

[0090] Economic efficiency (cost-benefit ratio): equal to the ratio of benefit to cost.

[0091] Feasibility (Technical Adaptability): A feasibility score calculated based on technical constraints.

[0092] Riskiness: A risk prediction indicator based on a global model.

[0093] Comprehensive evaluation: Use weighting method or multi-objective decision-making method to integrate various evaluation indicators and obtain the final comprehensive evaluation score S new : Among them, α k is the weight of the kth evaluation indicator, and m is the number of evaluation indicators.

[0094] Output: Comprehensive evaluation score S of the bidding proposal new , used for decision support.

[0095] Tender proposal evaluation refers to the process of conducting a comprehensive and systematic review and analysis of various proposals submitted by bidders during the bidding process to determine the best supplier or service provider. A comprehensive evaluation is conducted from multiple aspects including technology, economy, supplier credibility and capabilities, and contract terms and conditions. In another exemplary embodiment of the present application, a comprehensive scoring report is generated based on the evaluation results output by the model. This report provides decision makers with a scientific and objective evaluation basis to help them make reasonable procurement decisions. The evaluation report includes: comprehensive evaluation score, technical advantages and quotation rationality.

[0096] Exemplarily, the method for generating a technical advantage includes the following steps 201 to 206 .

[0097] Step 201: Obtain the technical documents in the current bidding proposal of each bidder.

[0098] Step 202: extract technical parameters from the technical document and standardize them into technical feature vectors.

[0099] Technical documents are unstructured technical documents, and key features can be extracted through NLP (Natural Language Processing) technology.

[0100] Step 203: Construct an evaluation model based on the federated neural network.

[0101] Step 204: Input the technical feature vector into the evaluation model based on the federated neural network, and output the technical score of the current bidding solution of each bidder.

[0102] In order to improve the output accuracy of technical scores, the hierarchical analysis method can be used to dynamically adjust the weights of technical indicators in the evaluation model based on the federated neural network through a global model of federated aggregation.

[0103] Step 205: Construct a knowledge graph of power equipment based on the industry standards and historical optimal technical parameters of power equipment.

[0104] Step 206: Based on the knowledge graph, use a graph neural network to compare the current bidding proposals of each bidder with the technical benchmark to identify the advantages of the current bidding proposals of each bidder.

[0105] In one example, a method for generating bid rationality includes: using a trained regression model to predict a reasonable price range for power equipment based on historical bids for power equipment from multiple bidders; comparing the bid in each bidder's current bid proposal with the reasonable price range to determine whether the bid in each bidder's current bid proposal is reasonable.

[0106] The training process for the trained regression model is to integrate historical bid data from multiple bidders and train the regression model without sharing the original data. This training process is called horizontal federated learning.

[0107] Power equipment procurement projects have long lead times, with equipment delivery potentially months in the future. This requires predicting reasonable price fluctuations at the time of delivery. However, real-time prices can be affected by short-term market fluctuations and may not reflect long-term trends. In another example, when real-time market data is available, the method for determining bid rationality can be replaced by steps 301 through 308.

[0108] Step 301: Obtain historical market prices and external dynamic data of power equipment and construct a multi-factor time series feature vector; the external dynamic data includes the power industry commodity price index, raw material price curve, international exchange rate and transportation cost.

[0109] Step 302: Based on the multi-factor time series feature vector, perform federated learning on the multi-factor time series model to obtain a multi-factor time series prediction model.

[0110] Under the federated learning framework, each participant uses local data to train a multi-factor time series prediction model, only uploads the encrypted model parameter gradient to the federated server, and generates a global prediction model through dynamic aggregation.

[0111] All participants share encrypted model parameter gradients, jointly adjust model parameters, and ultimately jointly train a unified multi-factor time series prediction model.

[0112] Step 303: Based on the current multi-factor time series feature vector, the multi-factor time series prediction model is used to dynamically predict the price of the power equipment at the time of delivery.

[0113] Based on the LSTM or Transformer model, a multi-factor time series prediction module is designed, and the input feature vector is:

[0114] X t =[P market ,I commodity ,C material ,R exchange ,T transport ];

[0115] Where, P market is the real-time market price; I commodity is the commodity price index; C material is the raw material price curve; R exchange is the international exchange rate; T transport For transportation costs.

[0116] Step 304: Based on the dynamically predicted price of the power equipment at delivery and the current market price of the power equipment, use the formula Calculate the base price; where P bace is the base price, α is the weight coefficient, is the dynamically predicted price of power equipment at delivery, P market The real-time market price of power equipment.

[0117] Step 305: Based on the base price, use the formula [P min , P max ]=[P bace (1-β), P bace ·(1+γ)], determine the reasonable price range of power equipment in real time; where, P min The lowest price in the reasonable price range, P max is the highest price in the reasonable price range, β is the price floating threshold, and γ is the price floating threshold.

[0118] Step 306: Based on the reasonable price threshold, the current bid price of each bidder is compared with the real-time reasonable price range using the isolation forest algorithm or the local outlier factor algorithm.

[0119] Step 307: If the bid price falls within the real-time reasonable price range, the bid price rationality of the current bidding proposal is reasonable.

[0120] Step 308: If the quotation does not fall within the real-time reasonable price range, the quotation rationality of the current bidding solution is unreasonable.

[0121] For example, a more specific implementation process of steps 301 to 308 is as follows.

[0122] 1. Data preprocessing stage.

[0123] (1) Each data holder standardizes local historical data.

[0124] (2) Market data stream processing: Obtain market prices in real time through API and parse JSON format market data.

[0125] 2. Federated model training.

[0126] Each supplier collects its own historical bidding proposals and quotation data to form an independent local data set, without having to share the original data with bidders.

[0127] Use historical quote data to train a regression model. Input features include device specifications, historical transaction prices, and timestamps.

[0128] Suppliers use their own data sets to train regression models (for predicting reasonable price ranges) or anomaly detection models (such as isolation forests) locally, encrypt the model parameters and upload them to the bidder's federal server. The server aggregates the parameters through an asynchronous federated learning algorithm to generate a global model.

[0129] Bidders use the global model, combined with real-time market prices, to calculate a reasonable price range and compare it with current bid prices.

[0130] 3. Dynamic price range calculation.

[0131] The base price calculation formula is:

[0132] Interval range: [P min , P max ]=[P bace (1-β), P bace (1+γ)]. β and γ are dynamically adjusted based on historical volatility.

[0133] 4. Abnormal quotation detection.

[0134] Use the isolation forest algorithm to compare the bid price with the real-time reasonable price range: if the bid price is within the range, it is judged as "reasonable"; if the deviation exceeds the threshold, it is marked as "unreasonable".

[0135] In another example, the cost decomposition model can also be used to identify abnormal cost items: through federated learning, the cost structure of each supplier is jointly analyzed, and the gradient boosting tree is used to establish a cost-quotation mapping relationship to identify abnormal cost items.

[0136] In another example, technical scores and quotation data are combined to build a Pareto frontier model, and federated reinforcement learning is used to find the optimal technology-cost balance point to assist in decision-making.

[0137] In complex scenarios such as joint bidding and subcontracting, traditional methods have difficulty in dynamically predicting the risks of partners or subcontractors, and lack the ability to integrate multi-supplier collaboration data. To solve this problem, in another exemplary embodiment of the present application, the above method also includes: introducing a partner evaluation sub-model in the asynchronous federated learning framework to analyze the historical collaboration data of the joint bidders and output a collaboration risk score; the calculation formula for the collaboration risk score is Where R is the cooperation risk score, ω i is the risk weight of the i-th partner, s i is the historical cooperation success rate, c i is the real-time credit indicator, f is the normalization function, and n is the number of partners.

[0138] In another exemplary embodiment of the present application, the method further includes: within the same bidder, dividing local data sets into the technical department and the business department, performing model training through a multi-node federated learning framework within the enterprise, and encrypting gradient parameters between departments using differential privacy technology to obtain a department collaborative evaluation model; the local data set of the technical department includes power equipment performance parameters and technical scores; the local data set of the business department includes quotation data and quotation rationality scores of the power equipment;

[0139] Based on the power equipment performance parameters and power equipment quotation data of the same bidder in the current bidding plan, the department collaborative evaluation model is used to generate technical scores, quotation rationality scores and department collaborative comprehensive scores, and a department collaborative evaluation report is generated.

[0140] The collaborative departmental evaluation report is implemented through an enterprise-wide federated learning framework and consists of three phases: data preparation, model training, and report generation. First, the technical department standardizes equipment performance parameters (such as efficiency and failure rate) and adds differential privacy noise, while the business department encrypts the quote data. Subsequently, each department trains its own local model: the technical model outputs a technical score, and the business model outputs a quote rationality score. The enterprise-wide federated server integrates the model parameters of both departments using a secure aggregation protocol to generate a composite score. The report is then automatically generated as a PDF or visualization report based on predefined rules.

[0141] The report contains four core modules:

[0142] Technical evaluation: technical score (0-100 points), advantage parameters (such as efficiency and compatibility), gap with industry benchmarks, and improvement suggestions;

[0143] Business evaluation: quotation score (percentage of deviation from market price), cost structure (raw materials, transportation ratio), real-time market risk warning (such as price fluctuations);

[0144] Synergy analysis: comprehensive score (technology-quote weighted balance), matching rating (such as "high cost-performance" or "technology surplus"), and synergy risk index (quantified supply chain or cost risk);

[0145] Optimization recommendations: Targeted measures, urgency level (high / medium / low). Reports are dynamically updated through a federated model to ensure data privacy and real-time decision-making.

[0146] Traditional evaluation methods primarily focus on selecting proposals during the bidding phase and lack continuous tracking and feedback of full-lifecycle data such as operational performance, maintenance costs, and failure rates after equipment delivery. This results in the evaluation system being limited to short-term decision-making and making it difficult to dynamically optimize the supplier's long-term service quality. Therefore, in another exemplary embodiment of the present application, the method further includes: during the equipment delivery and operation and maintenance phase of the winning bidder, continuously collecting acceptance data, performance indicators, and maintenance costs of the power equipment to form an incremental dataset; based on an asynchronous federated learning framework, inputting the incremental dataset into the asynchronously trained local model for retraining, updating the parameters of the federated model, and obtaining an updated federated model; based on the updated federated model, obtaining a power equipment reliability score and generating a full-lifecycle performance evaluation report. The updated federated learning model indirectly generates the final equipment reliability, long-term cost, and technical degradation indicators by predicting intermediate data and combining them with preset rules and formulas. For example, for equipment reliability prediction, the model outputs the probability of future failures; for long-term cost, the model estimates the total cost of the equipment over its entire lifecycle; and for technical degradation trends, the model predicts the attenuation curve of performance parameters.

[0147] During the equipment delivery phase, equipment acceptance data and initial operational indicators are collected, including equipment specification compliance, installation and commissioning efficiency, and initial failure rate. During the operation and maintenance phase, IoT sensors are used to obtain real-time equipment performance data and maintenance records. Incremental data sets are federated into the federated model, and global model parameters are dynamically updated through asynchronous federated learning to optimize the evaluation weights of subsequent bidding proposals. The calculation formula is:

[0148]

[0149] Where α is the learning rate, D 增量 For incremental datasets, is the gradient of the loss function.

[0150] This application uses a federated learning framework to support the full lifecycle assessment of power equipment, optimizing the accuracy of subsequent bid evaluations. By transmitting data back throughout the entire lifecycle and dynamically updating models, it expands from a single bidding evaluation to the full management of power equipment procurement, operation, and maintenance, helping bidders continuously optimize their products and services while providing long-term decision support for purchasers.

[0151] In another exemplary embodiment of the present application, the method also employs security and privacy protection throughout the data transmission and model training process. Data protection ensures that data privacy is fully protected during data transmission, model training, and evaluation to prevent data leakage. Compliance checks ensure that the entire process complies with relevant data protection regulations and standards.

[0152] The implementation method can fully utilize the data of all parties to conduct scientific evaluation of bidding proposals while protecting data privacy, thereby improving the accuracy and fairness of procurement decisions.

[0153] The method of this application can be summarized as the following steps: First, collect historical bidding data and evaluation data related to power equipment procurement from each bidder; then, use the federated learning framework to train the model between different data holders to ensure data privacy while jointly optimizing the model; then, use the trained federated model to evaluate the current bidding scheme, and evaluate the pros and cons of each bidding scheme through the prediction model; finally, generate a comprehensive scoring report based on the evaluation results output by the model to assist decision makers in making scientific and reasonable procurement decisions. This method can effectively integrate information from multiple parties, improve the comprehensiveness and accuracy of bidding scheme evaluation, and reduce the subjectivity and bias caused by relying entirely on evaluation experts during the evaluation process. The key to this method is to utilize the privacy protection capabilities of federated learning while integrating data from multiple parties for model training, thereby improving the effectiveness of bidding scheme evaluation.

[0154] The following provides a method for evaluating bidding proposals for power transformer procurement.

[0155] Background: A power company plans to purchase a batch of power transformers, involving multiple suppliers submitting bids. The company hopes to use a scientific and objective method to evaluate each bid and improve the accuracy and fairness of procurement decisions. Figure 3 shown.

[0156] (1) Data collection and preprocessing.

[0157] Data Collection: Collect historical bidding data from various suppliers, including previous quotes, delivery times, and equipment performance. Gather internal procurement demand data and transformer technical requirements from the power company. Summarize historical procurement records, including equipment performance, maintenance records, and costs.

[0158] Data preprocessing: Clean the data to remove duplicate or erroneous information. Normalize the data to maintain consistency during model training.

[0159] (2) Federated learning environment configuration.

[0160] Participant Setup: The power company is set as the data center, responsible for managing the overall federated learning process. Each supplier, as the data holder, stores its historical bidding data locally and does not expose it to other suppliers.

[0161] Privacy protection technology configuration: Configure homomorphic encryption technology to ensure that data remains encrypted during transmission and processing.

[0162] (3) Model design and training.

[0163] Model design: Select a regression model (such as linear regression or neural network) to predict the overall score of the bid proposal.

[0164] Local model training: Each vendor independently initiates local training without waiting for a global synchronization signal. Each participant uses local data to calculate parameter updates based on the currently available global model version and triggers upload immediately after training is complete, without aligning with the progress of other nodes.

[0165] Global Model Updates: A central server asynchronously receives and streams model updates submitted by various vendors. With each update, the global model is adjusted immediately through a dynamic fusion mechanism, without waiting for all nodes to submit. Updated global model parameters are distributed to vendors via on-demand push or event-driven delivery. Nodes that have submitted updates receive the latest version first, while inactive nodes can delay synchronization or pull updates at their own pace. Asymmetric training frequency and network latency are supported.

[0166] (4) Evaluation of bidding proposals.

[0167] Proposal input: The bidding proposals submitted by each supplier (such as quotation, delivery time, technical specifications, etc.) are input into the trained federated model.

[0168] Evaluation calculation: Use the global model to score each bidding proposal and generate a predicted score for each proposal.

[0169] Result analysis: Analyze the scoring results output by the model, compare the advantages and disadvantages of different solutions, and consider the performance and cost-effectiveness of various aspects.

[0170] (5) Report generation and decision support.

[0171] Comprehensive scoring report: Generate a detailed scoring report, including the score of each bidding scheme, technical advantage analysis, quotation rationality, etc.

[0172] Decision support: Submit the scoring report to the procurement decision-making team as a reference for procurement decisions.

[0173] (6) Full life cycle data feedback and model optimization.

[0174] During the equipment delivery and operation and maintenance phases, equipment acceptance data, performance indicators, and maintenance costs are continuously collected to form incremental data sets.

[0175] Based on the asynchronous federated learning framework, incremental datasets are fed into the local model for training and the parameters of the global model are updated.

[0176] Based on the updated federal model, a full life cycle performance evaluation report is generated, which includes equipment reliability score, long-term cost-effectiveness ratio and technology degradation trend.

[0177] (7) Security and privacy protection.

[0178] Data protection: Ensure the use of encryption technology during data transmission to protect data privacy and security.

[0179] Compliance checks: Ensure that data processing and model training processes comply with data protection regulations.

[0180] (8)Results.

[0181] Through the above embodiments, power companies can use federated learning technology to scientifically evaluate various bidding proposals while protecting the data privacy of each supplier, thereby improving the accuracy and fairness of procurement decisions.

[0182] The regression model code is as follows.

[0183] import numpy as np

[0184] from sklearn.linear_model importLinearRegression

[0185] from sklearn.metrics importmean_squared_error.

[0186] Line 1: Imports the numpy library, abbreviated as np. Numpy is a core library for scientific computing in Python, providing efficient array and matrix operations and commonly used in data preprocessing and numerical computation in machine learning. Line 2: Imports the LinearRegression class from the linear_model module of the scikit-learn library. scikit-learn is a Python machine learning library, and LinearRegression is an implementation class of the linear regression model. Line 3: Imports the mean_squared_error function from the metrics module of the scikit-learn library to calculate the mean squared error of the regression model.

[0187] The sample data (local dataset) code is as follows.

[0188] X = np.array([[1,2],[2,3],[4,5],[6,7]]) # Create sample input feature data (4 samples, 2 features per sample)

[0189] y = np.array([2,3,4,6]) #Create the corresponding target value data (the true value of the 4 samples).

[0190] The code for training the local regression model is as follows.

[0191] model=LinearRegression()#Initialize linear regression model object

[0192] model.fit(X,y)#Use training data (X and y) to train (fit) the regression model.

[0193] The code for prediction and calculation of loss is as follows.

[0194] predictions = model.predict(X) #Use the trained model to predict the training data

[0195] loss = mean_squared_error(y,predictions) #Calculate the mean squared error between the predicted value and the true value (Mean Squared Error)

[0196] print("Model parameters:",model.coef_)#Output the weight coefficient of the model (parameters corresponding to the features)

[0197] print("Prediction loss:",loss)#Output the prediction loss of the model on the training data.

[0198] This application solves the data privacy protection problem in traditional evaluation methods by innovatively introducing federated learning technology, while improving the accuracy and efficiency of the evaluation, providing an advanced method for bidding scheme evaluation in power equipment procurement and other similar scenarios.

[0199] Based on the same inventive concept, embodiments of the present application also provide a power equipment procurement bid scheme evaluation device for implementing the above-mentioned power equipment procurement bid scheme evaluation method. The solution provided by this device is similar to the solution described in the above-mentioned method. Therefore, the specific limitations of one or more power equipment procurement bid scheme evaluation device embodiments provided below can be found in the above-mentioned limitations of the power equipment procurement bid scheme evaluation method and will not be repeated here.

[0200] In an exemplary embodiment, Figure 4 As shown, a power equipment procurement bidding scheme evaluation device is provided, which includes: a collection module, a selection module, a training module, an application module, a comprehensive evaluation module, an optimization suggestion production module, a full life cycle data acquisition module, an incremental training module and a performance report generation module.

[0201] The collection module is used to collect the historical bidding plans of each bidder and multiple evaluation index data of each historical bidding plan, and form a local data set of each bidder.

[0202] Collection modules include Figure 4 The data collection and preprocessing module and the parameter aggregation module are used to construct the local data set of each bidder.

[0203] The selection module is used to select machine learning models as local models and global models in the asynchronous federated learning framework.

[0204] The training module is used to asynchronously train local models on each bidder using its own local data set based on the asynchronous federated learning framework, and to train the global model using a dynamic aggregation strategy based on the parameters of the trained local models, and to determine the trained global model as the federated model; the dynamic aggregation strategy includes the weight allocation of collaborative data from multiple suppliers in the joint bidding scenario, as well as the feature fusion parameters of data from multiple departments within the same bidder.

[0205] Training modules include Figure 4 The global model update module in . The global model update module refers to the parameter update of the global model during the training process.

[0206] The application module is used to input the current bidding scheme of each bidder into the federal model respectively, and output the predicted values of multiple evaluation indicators of the current bidding scheme of each bidder.

[0207] The comprehensive evaluation module is used to determine the comprehensive evaluation results of the current bidding schemes of each bidder based on the predicted values of multiple evaluation indicators.

[0208] The optimization suggestion production module is used to generate an evaluation report of the current bidding scheme of each bidder based on the comprehensive evaluation results.

[0209] The application module and the comprehensive evaluation module constitute the solution prediction module, the solution prediction module and the optimization suggestion production module form the global model evaluation module, and the parameter aggregation module, the global model update module and the global model evaluation module can be executed on the federated server (center).

[0210] The full life cycle data collection module is used to collect performance data in real time during the equipment delivery and operation stages.

[0211] The incremental training module, based on the asynchronous federated learning mechanism, inputs incremental data into the local model and updates the global parameters.

[0212] The performance report generation module outputs equipment reliability, long-term cost and technology degradation indicators based on the updated model.

[0213] In an exemplary embodiment, Figure 4 It is shown that the power equipment procurement bidding scheme evaluation device also includes: a privacy protection module, which ensures that the federated learning framework supports privacy protection technologies such as homomorphic encryption or differential privacy to ensure data security.

[0214] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store evaluation reports. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for evaluating a bidding scheme for the procurement of power equipment is implemented.

[0215] Those skilled in the art will understand that Figure 5 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application and does not constitute a limitation on the computer device to which the solution of the present application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps of the above-mentioned method embodiments when executing the computer program.

[0216] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0217] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for evaluating a bidding scheme for power equipment procurement, characterized in that: include: Collect the historical bidding proposals of each bidder and multiple evaluation indicator data of each historical bidding proposal, and form a local data set for each bidder; Select machine learning models as local models and global models in the asynchronous federated learning framework; Based on the asynchronous federated learning framework, each bidder asynchronously trains a local model using its own local dataset. Based on the trained local model parameters, a dynamic aggregation strategy is used to train a global model, which is then used as the federated model. The dynamic aggregation strategy includes weighting the collaborative data of multiple suppliers in joint bidding scenarios, as well as feature fusion parameters for data from multiple departments within the same bidder. Inputting the current bidding proposals of each bidder into the federated model respectively, and outputting predicted values of multiple evaluation indicators of the current bidding proposals of each bidder; Determine the comprehensive evaluation results of each bidder's current bidding proposal based on the predicted values of multiple evaluation indicators; Based on the comprehensive evaluation results, an evaluation report of each bidder's current bidding proposal is generated.

2. The method for evaluating a bidding scheme for power equipment procurement according to claim 1, characterized in that: Select a machine learning model, including: If the prediction result is a numerical value, the regression model is selected; If used for predictive classification, select the classification model; If the size of the local dataset is larger than the preset threshold, or if feature extraction is required, a deep learning model is selected; If the data distribution of the local dataset is nonlinear, a nonlinear model is selected.

3. The method for evaluating a bidding scheme for power equipment procurement according to claim 1, characterized in that: Based on the predicted values of multiple evaluation indicators, the comprehensive evaluation results of each bidder's current bidding proposal are determined, including: Based on the predicted values of multiple evaluation indicators, a weighting method or a multi-objective decision-making method is used to obtain a comprehensive evaluation score of each bidder's current bidding proposal.

4. The method for evaluating a bidding scheme for power equipment procurement according to claim 3, characterized in that: The evaluation report includes: comprehensive evaluation score, technical advantages and quotation rationality.

5. The method for evaluating a bidding scheme for power equipment procurement according to claim 4, characterized in that: The method for generating the technical advantage includes: Obtain technical documents from each bidder's current bid proposal; Extracting technical parameters from the technical document and standardizing them into technical feature vectors; Build an evaluation model based on federated neural networks; Inputting the technical feature vector into an evaluation model based on a federated neural network to output a technical score of each bidder's current bid proposal; Construct a knowledge graph of power equipment based on industry standards and historically optimal technical parameters of power equipment; Based on the knowledge graph, a graph neural network is used to compare the current bidding proposals of each bidder with the technical benchmark to identify the advantages of the current bidding proposals of each bidder.

6. The method for evaluating a bidding scheme for power equipment procurement according to claim 4, characterized in that: The method for generating the rationality of the quotation includes: Based on historical bids for power equipment from multiple bidders, the trained regression model is used to predict the reasonable price range for power equipment. The price quoted in the current bidding proposal of each bidder is compared with the reasonable price range to determine whether the price quoted in the current bidding proposal of each bidder is reasonable.

7. The method for evaluating a bidding scheme for power equipment procurement according to claim 4, characterized in that: When real-time market data is available, the method for generating the quote rationality includes: Obtain historical market prices and external dynamic data for power equipment and construct a multi-factor time series feature vector; the external dynamic data includes the power industry commodity price index, raw material price curve, international exchange rates, and transportation costs; According to the multi-factor time series feature vector, the multi-factor time series model is federated learned to obtain a multi-factor time series prediction model; Dynamically predicting the price of power equipment upon delivery using the multi-factor time series prediction model based on the current multi-factor time series feature vector; According to the dynamic forecast of the price of power equipment at delivery and the current market price of power equipment, the formula Calculate the base price; where P bace is the base price, α is the weight coefficient, is the dynamically predicted price of power equipment at delivery, P market The real-time market price of power equipment; According to the benchmark price, using the formula [P min , P max ]=[P bace (1-β), P bace ·(1+γ)], determine the reasonable price range of power equipment in real time; where, P min The lowest price in the reasonable price range, P max is the highest price in the reasonable price range, β is the price floating threshold, and γ is the price floating threshold; Based on the reasonable price threshold, using an isolation forest algorithm or a local outlier factor algorithm, the current bid price of each bidder is compared with a real-time reasonable price range; If the bid price falls within the real-time reasonable price range, the current bid is reasonable; If the quotation does not fall within the real-time reasonable price range, the quotation rationality of the current bidding proposal is unreasonable.

8. The method for evaluating a bidding scheme for power equipment procurement according to claim 1, wherein: The method further includes: introducing a partner evaluation sub-model into the asynchronous federated learning framework to analyze the historical cooperation data of the joint bidders and output a cooperation risk score; the calculation formula of the cooperation risk score is: Where R is the cooperation risk score, ω i is the risk weight of the i-th partner, s i is the historical cooperation success rate, c i is the real-time credit indicator, f is the normalization function, and n is the number of partners.

9. The power equipment scheme evaluation method according to claim 1, characterized in that: The method further includes: within the same bidder, dividing local data sets into the technical department and the commercial department, performing model training through a multi-node federated learning framework within the enterprise, and encrypting gradient parameters between departments using differential privacy technology to obtain a departmental collaborative evaluation model; the local data set of the technical department includes power equipment performance parameters and technical scores; the local data set of the commercial department includes quotation data and quotation rationality scores for power equipment; Based on the power equipment performance parameters and power equipment quotation data of the same bidder in the current bidding plan, the department collaborative evaluation model is used to generate technical scores, quotation rationality scores and department collaborative comprehensive scores, and a department collaborative evaluation report is generated.

10. The method for evaluating a bidding scheme for power equipment procurement according to claim 1, wherein: The method further comprises: During the equipment delivery and operation and maintenance phase of the winning bidder, the acceptance data, performance indicators, and maintenance costs of the power equipment are continuously collected to form an incremental data set; Based on the asynchronous federated learning framework, the incremental dataset is input into the asynchronously trained local model for retraining, and the parameters of the federated model are updated to obtain the updated federated model. Based on the updated federal model, the reliability score of power equipment is obtained and a full life cycle performance evaluation report is generated.

11. A device for evaluating bidding schemes for power equipment procurement, characterized in that: include: A collection module is used to collect historical bidding proposals of each bidder and multiple evaluation index data of each historical bidding proposal, and form a local data set of each bidder; The selection module is used to select machine learning models as local models and global models in the asynchronous federated learning framework; The training module is used to asynchronously train local models at each bidder using their own local datasets based on an asynchronous federated learning framework. Based on the trained local model parameters, a dynamic aggregation strategy is used to train a global model, which is then used as the federated model. The dynamic aggregation strategy includes weighting the collaborative data of multiple suppliers in a joint bidding scenario, as well as feature fusion parameters for data from multiple departments within the same bidder. An application module, configured to input the current bidding proposals of each bidder into the federated model and output predicted values of multiple evaluation indicators of the current bidding proposals of each bidder; A comprehensive evaluation module is used to determine the comprehensive evaluation results of each bidder's current bidding plan based on the predicted values of multiple evaluation indicators; The optimization suggestion production module is used to generate an evaluation report of the current bidding scheme of each bidder based on the comprehensive evaluation results.

12. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for evaluating a bid scheme for procurement of electric power equipment according to any one of claims 1 to 10.

Images