PCB procurement platform AI intelligent supply and demand precise matching method, system, terminal and medium

Abstract

The application discloses a PCB procurement platform AI intelligent supply and demand precise matching method and system, a terminal and a medium, and relates to the technical field of PCB procurement platforms. The method comprises the following steps: acquiring demand information and extracting structured demand parameters, and generating a demand feature vector by using a first neural network encoder; acquiring a historical production load sequence, process capability data and quality data of a supplier, predicting future available capacity by using an AI capacity prediction model, and generating a supplier feature vector by combining the process and quality data through a second neural network encoder; inputting the demand and supplier feature vectors into a deep matching network based on an attention mechanism, calculating a supply and demand matching probability; and outputting a matching result based on the probability. The application overcomes the defects of the existing static scoring method which ignores real-time load by combining dynamic capacity prediction with an attention mechanism deep matching method, realizes high-precision and load-aware supply and demand matching, and improves the matching success rate and platform capacity balance.

Description

Technical Field

[0001] This invention relates to the field of PCB platform trading technology, and in particular to an AI-powered intelligent supply and demand matching method, system, terminal, and medium for PCB procurement platforms. Background Technology

[0002] In PCB (Printed Circuit Board) procurement platforms, buyers typically need to fill out cumbersome requirement templates, manually entering information such as the number of layers, dimensions, materials, and processes. This process is complex and prone to errors. To reduce the burden on customers, existing technologies have attempted to introduce multimodal input and automatic parameter extraction technologies, which can identify structured PCB requirement parameters from voice, text, or engineering files.

[0003] However, in the supply and demand matching process, existing methods generally remain at the stage of weighted scoring based on static attributes. That is, parameters such as board type and quantity of the demand side are compared with fixed indicators (such as main product category, maximum capacity, geographical location, etc.) pre-registered by suppliers, and a comprehensive score is calculated through artificially set fixed weights, and recommendations are made based on this score. The drawback of this approach is that the production status of suppliers changes dynamically, and their "current available capacity" fluctuates in real time with the number of orders in progress. However, the existing matching model only relies on static upper limits such as maximum capacity, and cannot detect whether suppliers are about to reach full capacity or are already idle. This leads to extremely uneven order allocation. For example, some suppliers are overloaded with orders, resulting in delays, while others have idle capacity. At the same time, the artificially set weights cannot adapt to the actual preferences of different customers and the matching priorities under different demand scenarios. The matching accuracy is low, and the purchaser often finds it difficult to find suppliers that have spare capacity, process adaptability, and high delivery quality, resulting in increased procurement costs and uncontrollable supply cycles.

[0004] In summary, the main deficiency of existing technologies lies in the failure to integrate real-time forecasting of supplier dynamic capacity and deep interactive feature learning into the supply and demand matching process. Therefore, it is necessary to provide an AI-powered intelligent supply and demand matching method, system, terminal, and medium for PCB procurement platforms to overcome these shortcomings. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-powered intelligent supply and demand matching method, system, terminal, and medium for PCB procurement platforms. It aims to solve the problem of how to combine dynamic capacity forecasting of suppliers with deep matching based on attention mechanisms to achieve highly accurate and load-aware automatic matching between PCB procurement needs and suppliers, thereby overcoming the inherent defects of static scoring.

[0006] To achieve the above objectives, the first aspect of the present invention provides an AI-powered intelligent supply and demand matching method for a PCB procurement platform, comprising: Step S10: Obtain the requirement information submitted by the customer, extract the structured requirement parameters, and use the first neural network encoder to generate the requirement feature vector; Step S20: Obtain the attribute data of each candidate supplier, including historical production load sequence, process capacity data and quality data; use the AI ​​capacity prediction model to predict the future available capacity of the supplier based on the historical production load sequence, and combine the process capacity data and quality data to generate supplier feature vectors through a second neural network encoder. Step S30: Input the demand feature vector and the supplier feature vector into a deep matching network based on an attention mechanism to calculate the supply-demand matching probability of the supplier; Step S40: Based on the supply and demand matching probability, output the supply and demand matching result.

[0007] In a preferred embodiment, step S10, extracting the structured requirements parameters, includes: A multimodal AI perception model is used to identify at least one input from voice, text, images, or engineering files, and to extract the number of PCB layers, size, board type, and process requirements respectively. By comparing the recognition results of different modalities through a cross-validation mechanism, when conflicts exist, the recognition results of the engineering documents shall prevail or natural language clarification questions shall be generated to confirm with the customer, and conflict-free structured requirement parameters shall be output.

[0008] In a preferred embodiment, the AI ​​capacity prediction model is a Long Short-Term Memory (LSTM) network model, whose input is the supplier's work-in-process area sequence W over the past T time units. hist The output is the predicted future production capacity consumption. The future available capacity C avail Calculate using the following formula: , ; Among them, C max This represents the supplier's maximum production capacity.

[0009] In a preferred embodiment, the deep matching network based on the attention mechanism includes a demand self-attention layer, a supplier self-attention layer, a demand-supplier cross-attention layer, and a multilayer perceptron output layer; the supply-demand matching probability P is calculated as follows: ; Among them, h d h is the demand feature vector. s Here, denoted as the supplier feature vector, CrossAttend represents the cross-attention operation, MLP stands for Multilayer Perceptron, and σ is the sigmoid function.

[0010] In a preferred embodiment, step S40 includes: when there are multiple unmatched demands, using the real-time load rate ρi of all suppliers and the supply-demand matching probability as input, and using a trained reinforcement learning agent to output a global supply-demand allocation scheme. The reward function R of the reinforcement learning agent includes a load balancing term: ; Among them, S succ To indicate a successful match, T represents the variance of the load factor of each supplier. on To ensure timely delivery and fulfillment of rewards, a,b,c These are the preset weighting coefficients.

[0011] In a preferred embodiment, a process pre-screening step is included before step S30: using a pre-built PCB process knowledge graph, the process parameters in the requirements are semantically aligned with the process capability tags of the suppliers, and suppliers with non-coverable process conflicts are filtered out.

[0012] In a preferred embodiment, after step S40, a quotation anomaly verification step is further included: using a pre-trained quotation regression model, inputting features including at least area, number of layers, and process complexity, to calculate an estimated reasonable price; if the supplier's actual quotation exceeds the preset fluctuation range of the estimated reasonable price, it is marked as an abnormal quotation and a review is triggered.

[0013] A second aspect of this invention provides an AI-powered intelligent supply and demand matching system for a PCB procurement platform, comprising: The requirement encoding module is used to obtain the requirement information submitted by the customer, extract the structured requirement parameters, and generate the requirement feature vector using the first neural network encoder. The supplier coding module is used to acquire attribute data of each candidate supplier, including historical production load sequence, process capacity data, and quality data; using an AI capacity prediction model, the future available capacity of the supplier is predicted based on the historical production load sequence, and combined with the process capacity data and quality data, a supplier feature vector is generated through a second neural network encoder. The matching calculation module has a built-in deep matching network based on the attention mechanism, which is used to input the demand feature vector and the supplier feature vector into the deep matching network based on the attention mechanism to calculate the supply and demand matching probability of the supplier. The matching and allocation module is used to output the supply and demand matching result based on the supply and demand matching probability.

[0014] A third aspect of the present invention provides a terminal, the terminal including a memory, a processor, and a computer program stored in the memory, wherein when the computer program is executed by the processor, it implements the various steps of the AI ​​intelligent supply and demand precise matching method for PCB procurement platform as described in any of the above embodiments.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the various steps of the AI-powered intelligent supply and demand matching method for a PCB procurement platform as described in any of the above embodiments.

[0016] The fifth aspect of the present invention provides a computer program product, including a computer program or instructions, wherein when the computer program or instructions are processed and executed, they implement the various steps of the AI ​​intelligent supply and demand precise matching method for PCB procurement platform as described in any of the above embodiments.

[0017] The AI-powered intelligent supply and demand matching method, system, terminal, and medium provided by this invention for a PCB procurement platform integrates dynamic capacity forecasting into supplier characterization, overcoming the shortcomings of existing technologies that ignore real-time load. A deep matching network based on attention replaces manual fixed weights, automatically learning matching patterns from historical transaction data to accurately distinguish between suppliers that are "perfectly matched" and those that are "partially satisfied," significantly improving the matching success rate and reducing the trial-and-error costs and delivery risks for the purchaser due to inaccurate matching. Simultaneously, it encourages orders to flow to suppliers with surplus capacity, balancing the platform's production load. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of the AI-powered intelligent supply and demand matching method for the PCB procurement platform provided by this invention; Figure 2 This is a framework diagram of the AI-powered intelligent supply and demand matching system for the PCB procurement platform provided by this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described in this specification are merely for explaining the invention and are not intended to limit the invention.

[0021] It should be noted that when an element is referred to as being "fixed to" or "set on" another element, it can be directly on the other element or there may be an intervening element. When an element is considered to be "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0022] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0023] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0024] Example 1 In an embodiment of the present invention, an AI-powered intelligent supply and demand matching method for a PCB procurement platform is provided. This method can automatically sense the future available capacity of suppliers and learn the complex interactive matching relationship between demand and suppliers through a deep neural network, thereby outputting a more accurate supply and demand matching result.

[0025] It should be noted that this method can be deployed entirely or partially on cloud servers, edge computing nodes, or local terminal devices. Each AI model can operate using a centralized training and cloud inference, local lightweight inference, or a hybrid of both. The model type can be flexibly selected, such as neural networks, probabilistic models, or combinations thereof.

[0026] like Figure 1 As shown, the method includes at least steps S10-S40.

[0027] Step S10: Obtain the requirement information submitted by the customer, extract the structured requirement parameters, and use the first neural network encoder to generate the requirement feature vector.

[0028] Customers can submit their requirements through various channels, including but not limited to voice, text, images, and project files (such as Gerber files), and traditional preset templates are also compatible. The platform's unified interface (e.g., an API gateway) receives the raw data packets and uses a multimodal AI perception model for recognition.

[0029] Specifically, if the input is speech, an end-to-end Automatic Speech Recognition (ASR) engine based on a deep neural network is triggered to convert the speech waveform into a text sequence. For text information, a pre-trained natural language processing model is used for Named Entity Recognition (NER). This model can be BERT (Bidirectional Encoder Representations from Transformers) or RoBERTa based on the Transformer architecture, extracting entities such as "four layers", "FR4", "10cm×10cm", "immersion gold", and "500 pieces", which correspond to the number of layers L, material type M, size area A, and special process C, respectively. tech And the quantity Q. If the input contains images or project files, a computer vision analysis module based on Convolutional Neural Network (CNN) is launched in parallel to identify physical parameters such as line width, line spacing, minimum aperture, and overall dimensions from the image or Gerber rendering.

[0030] Furthermore, to ensure the accuracy of the requirement parameters, the system performs a cross-validation step. The text parameters extracted by NLP are compared with the physical feature parameters identified by CV. For example, if the text states "minimum aperture 0.3mm," but the engineering documents, analyzed by the Design for Manufacturability (DFM) model, show an actual minimum aperture of 0.2mm, a conflict arises. In this case, according to preset rules, the engineering documents are taken as the highest basis for physical fact, and the parameter is corrected to 0.2mm, or the parameter is marked as pending confirmation. A natural language clarification question, such as "Your required aperture is 0.2mm, do you confirm?", is generated and sent to the customer via instant messaging interface for confirmation, thus correcting the requirement data accordingly. The final output is a conflict-free structured set of requirement parameters, whose key-value pairs can be represented as {Number of layers: 4, Board type: FR4, Size: [100mm, 100mm], Process: Immersion gold, Quantity: 500}.

[0031] After obtaining the structured parameters, the first neural network encoder is used to generate the required feature vector h. d In this embodiment, the first neural network encoder uses a Transformer-based encoding network. The input is a sequence of parameters after passing through the embedding layer, and the output is a fixed-dimensional required embedding vector, such as h. d ∈R 256The encoder has been pre-trained with a large amount of PCB inquiry data and is able to capture the combined semantics between parameters. For example, the manufacturing difficulty implied by "four-layer + rigid-flex board" is much higher than that of ordinary four-layer board.

[0032] Through the above steps, even if customers submit their requirements in messy natural language or drawings, the system can automatically extract complete and accurate structured parameters and transform them into feature vectors suitable for deep matching network processing, greatly reducing the input burden on customers and laying a reliable data foundation for subsequent accurate matching.

[0033] Step S20: Obtain the attribute data of each candidate supplier, including historical production load sequence, process capacity data and quality data; use the AI ​​capacity prediction model to predict the future available capacity of the supplier based on the historical production load sequence, and combine the process capacity data and quality data to generate the supplier feature vector through the second neural network encoder.

[0034] In this step, attribute data is collected for each candidate supplier i on the platform. Attribute data includes both static and dynamic components: static data such as equipment list, set of producible board types, and historical order delivery rate R. del,i and historical quality score R qual,i Dynamic data mainly consists of historical production load sequences over a recent period, such as the work-in-process area sequence W over the past 12 weeks. hist,i At the same time, the maximum monthly production capacity C of the supplier is read from the database. max,i .

[0035] To accurately determine a supplier's current and near-term production capacity, the system utilizes an AI-powered capacity forecasting model to predict the supplier's future available capacity based on historical production load sequences. For example, this AI capacity forecasting model is preferably a Long Short-Term Memory (LSTM) network model, whose input is the work-in-process area sequence W of supplier i over the past T time units (e.g., T=12 weeks). hist The output is the predicted future production capacity consumption. Then the future available production capacity C avail Calculate using the following formula: ,in, .

[0036] This LSTM model is trained using historical load data from various suppliers, enabling it to learn the cyclical patterns of their production rhythms, such as the typical surge in orders at the end of the month, thereby outputting reliable capacity utilization forecasts.

[0037] Then, the process capacity data (e.g., a list of surface treatment types that can be processed, represented as a multi-thermal encoded vector), quality data (delivery rate and score normalization values), and the above-predicted available capacity C are combined. avail (After normalization) the data is concatenated and input into the second neural network encoder to generate the supplier feature vector h. s The second neural network encoder can also use the same Transformer network architecture as the first encoder, but its parameters are trained independently. The generated h... s ∈R 256 This implies information about the supplier's real-time process adaptability, remaining capacity, and delivery quality.

[0038] Therefore, by integrating dynamic capacity forecasting and coding, the supplier feature vector is no longer a static "maximum capacity" label, but a "capacity snapshot" that evolves in real time with the actual production status. This enables the matching model to distinguish between suppliers that are "idle" and those that are "about to be at full capacity," which is a key improvement for achieving accurate matching.

[0039] Furthermore, in a preferred embodiment, before executing step S30, suppliers with hard mismatches are first eliminated through process pre-screening to reduce subsequent network computation and avoid mismatches. The platform pre-constructs a PCB process knowledge graph, storing the semantic associations between standard processes and specific terms of each supplier in a graph database. For example, "immersion gold" is equivalent to "chemical gold" in another supplier's system. The process fields in the structured requirement parameters obtained in step S10 are queried into this knowledge graph and semantically aligned with the process capability tags of each supplier. If a supplier does not possess a certain key process in the requirement (such as being unable to produce rigid-flex boards), it is directly filtered out and does not enter the subsequent matching calculation. This mechanism solves the problem of matching deviation caused by different suppliers using synonymous processes, ensuring from the source that the matching object is qualified in terms of process.

[0040] Step S30: Transfer the demand feature vector h d and the feature vector h of each pre-screened candidate supplier s A deep matching network based on an attention mechanism is used to input pairs of inputs to calculate the supply-demand matching probability of the supplier.

[0041] The deep matching network comprises the following layers: a demand self-attention layer to enhance the interaction between dimensions within the demand vector; a supplier self-attention layer to perform the same processing on the supplier vector; a demand-supplier cross-attention layer to query relevant information from the supplier vector using the demand vector; and a supplier-demand cross-attention layer to query relevant parts of the demand vector using the supplier vector. The cross-attention operation allows the model to automatically focus on key matching dimensions such as "whether the supplier's capacity can meet the demand quantity" and "whether the process is highly compatible." The two cross-attention results are concatenated and fed into a multilayer perceptron (MLP). Finally, the matching probability P is output through a sigmoid function, calculated as follows: ; Here, CrossAttend represents the overall mapping between cross-attention and the concatenation operation, and σ is the sigmoid function, which restricts the output to between 0 and 1. The closer the probability value is to 1, the higher the matching degree. This network is trained using the platform's historical transaction data, with successful transactions as positive samples and exposure without a transaction as negative samples, and is optimized using the cross-entropy loss function.

[0042] Understandably, this network can learn nonlinear matching relationships that are difficult to characterize using traditional methods: for example, when the predicted available capacity is only slightly higher than the required area, the probability P decreases significantly; conversely, when capacity is ample and the process coverage is perfect, P approaches 1. This fine-grained quantification capability is far superior to fixed-weight summation, significantly improving matching accuracy.

[0043] Step S40: Based on the supply and demand matching probability, output the supply and demand matching result.

[0044] In simple scenarios with only a single matching requirement, the system can directly sort the suppliers by their matching probabilities P calculated in step S30 from high to low, and select the supplier with the highest probability or the top K (e.g., K=5) as the matching result output.

[0045] When multiple demands exist simultaneously on the platform, simple sorting may result in some popular suppliers being selected repeatedly for multiple demands, while other suppliers remain idle. Therefore, this embodiment introduces a reinforcement learning agent for global supply and demand allocation. This agent is deployed as the core decision engine of the matching and allocation module, and its training algorithm can employ a Deep Q-Network (DQN). The state space is defined as the embedding vectors of all demands to be matched and the real-time load rate ρ of all suppliers. i , load factor ρ i = (C max,i - C avail,i ) / C max,iThis refers to the percentage of capacity already utilized. The action space allows for selecting a supplier or executing "not allocating" for each demand. The reward function R is designed as follows: ; Among them, S succ A successful match indication (1 when the selected supplier's matching probability P is higher than a threshold such as 0.8, otherwise 0); T represents the variance of load rates across all suppliers; the more unbalanced the load, the larger the variance. The negative sign indicates a penalty for the imbalance. on To ensure timely delivery and thus reward the supplier, the reward can be calculated based on the supplier's historical on-time delivery rate. a,b,c The weighting coefficients are preset, for example, α=1, β=0.5, γ=0.3. By maximizing the cumulative reward, the reinforcement learning agent learns to prioritize allocating orders to suppliers with low load and high matching probability. This ensures a high matching success rate while effectively balancing the production load of each supplier, thus improving the overall delivery stability of the platform.

[0046] Furthermore, a quote anomaly verification step can be performed before or after outputting the final matching result. This utilizes a pre-trained quote regression model, such as an XGBoost model based on gradient boosting trees, with input features including at least the required area A, the number of layers L, and the process complexity coefficient C. tech (This can be obtained by clustering and assigning values ​​to historical order processes, such as 0.5 for immersion gold and 0.2 for ordinary tin plating) and using current raw material price indices such as copper clad laminates, to calculate the estimated reasonable price P. est When the supplier's actual quote P offer When the value exceeds the preset floating range, such as P offer >1.3·P est or P offer <0.7·P est The system marks the quote as abnormal and automatically triggers a review process, such as sending a message to the supplier to remind them to verify, thereby eliminating malicious price increases or false reports and protecting the interests of the purchaser.

[0047] It is understandable that the above thresholds and coefficients can be flexibly adjusted according to actual business needs. For example, the floating range can be set to 1.5 and 0.5 times, and the load rate variance penalty coefficient β can be dynamically learned based on the platform's current emphasis on balance.

[0048] Example 2 In an embodiment of the present invention, an AI-powered intelligent supply and demand matching system for a PCB procurement platform is provided. This system corresponds entirely to the methods described in the above embodiments, and therefore will not be repeated below.

[0049] like Figure 2 As shown, the system mainly includes: The requirement encoding module 10 is used to obtain the requirement information submitted by the customer, extract the structured requirement parameters, and generate the requirement feature vector using the first neural network encoder. The supplier coding module 20 is used to acquire attribute data of each candidate supplier, including historical production load sequence, process capacity data and quality data; using an AI capacity prediction model, it predicts the future available capacity of the supplier based on the historical production load sequence, and combines the process capacity data and quality data to generate supplier feature vectors through a second neural network encoder. The matching calculation module 30 has a built-in deep matching network based on the attention mechanism, which is used to input the demand feature vector and the supplier feature vector into the deep matching network based on the attention mechanism to calculate the supply and demand matching probability of the supplier. The matching and allocation module 40 is used to output the supply and demand matching result based on the supply and demand matching probability.

[0050] Example 3 Based on the same inventive concept, the present invention also provides a terminal device, which includes a memory, a processor, and a computer program stored in the memory. When the processor executes the program, it implements the various steps of the above-described method embodiments. For example, when a purchaser inputs a voice request on a local terminal, a lightweight inference engine in the terminal can perform a matching calculation between the request encoding and locally cached supplier characteristics. The result is uploaded to the cloud or displayed directly, making it suitable for network-restricted scenarios.

[0051] Example 4 Similarly, the present invention also provides a computer-readable storage medium, such as a hard disk, solid-state drive, optical disk, or flash memory, on which a computer program is stored, which, when executed by a processor, also implements the above-described method. In this way, the software product can be easily distributed and installed on different PCB procurement platforms or user-end devices.

[0052] In summary, the present invention provides an AI-powered intelligent supply and demand matching method, system, terminal, and medium for a PCB procurement platform. In the demand encoding step, a first neural network encoder maps structured demand parameters into semantically rich demand feature vectors, enabling the compact expression of complex demands such as "four-layer rigid-flex board." Secondly, in the supplier encoding step, an AI capacity prediction model is introduced to predict the future available capacity of suppliers based on their historical production load sequences. This dynamic capacity information, along with process capability and quality data, is fed into a second neural network encoder to generate a supplier feature vector that reflects the supplier's capacity to fulfill orders in real time. Then, the matching calculation step employs a deep matching network based on an attention mechanism, allowing the demand feature vector and supplier feature vector to interact through cross-attention, automatically capturing key matching signals such as "process gaps" and "insufficient capacity," ultimately outputting an accurate supply and demand matching probability. Finally, the matching result is output based on this probability.

[0053] Therefore, integrating dynamic capacity forecasting into supplier characterization solves the problem of existing technologies ignoring real-time load; attention-based deep matching networks replace manual fixed weights, which can automatically learn matching patterns from historical transaction data, accurately distinguish between suppliers that are "perfectly matched" and those that are "partially satisfied", significantly improve the matching success rate, reduce the trial-and-error costs and delivery risks caused by inaccurate matching for the purchaser, and at the same time encourage orders to flow to suppliers with surplus capacity, thus balancing the platform's production load.

[0054] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0055] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0056] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0057] In the embodiments provided by this invention, it should be understood that the disclosed systems, devices / terminal equipment, and methods can be implemented in other ways. For example, the system or device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of systems or units may be electrical, mechanical, or other forms.

[0058] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0059] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0060] The present invention is not limited to the description in the specification and embodiments, and thus other advantages and modifications can be readily realized by those skilled in the art. Therefore, the present invention is not limited to the specific details, representative devices and illustrated examples shown and described herein without departing from the spirit and scope of the general concept as defined by the claims and their equivalents.

Claims

1. A method for precise supply and demand matching using AI in a PCB procurement platform, characterized in that, include: Step S10: Obtain the requirement information submitted by the customer, extract the structured requirement parameters, and generate the requirement feature vector using the first neural network encoder; Step S20: Obtain the attribute data of each candidate supplier, including historical production load sequence, process capacity data and quality data; use the AI ​​capacity prediction model to predict the future available capacity of the supplier based on the historical production load sequence, and combine the process capacity data and quality data to generate supplier feature vectors through a second neural network encoder. Step S30: Input the demand feature vector and the supplier feature vector into a deep matching network based on an attention mechanism to calculate the supply-demand matching probability of the supplier; Step S40: Based on the supply and demand matching probability, output the supply and demand matching result.

2. The AI-powered intelligent supply and demand matching method for PCB procurement platforms as described in claim 1, characterized in that, In step S10, extracting structured requirement parameters includes: A multimodal AI perception model is used to identify at least one input from voice, text, images, or engineering files, and to extract the number of PCB layers, size, board type, and process requirements respectively. By comparing the recognition results of different modalities through a cross-validation mechanism, when conflicts exist, the recognition results of the engineering documents shall prevail or natural language clarification questions shall be generated to confirm with the customer, and conflict-free structured requirement parameters shall be output.

3. The AI-powered intelligent supply and demand matching method for PCB procurement platforms as described in claim 1, characterized in that, The AI ​​capacity prediction model is a Long Short-Term Memory (LSTM) network model, whose input is the work-in-process area sequence W of the supplier over the past T time units. hist The output is the predicted future production capacity consumption. The future available capacity C avail Calculate using the following formula: ， ； Among them, C max This represents the supplier's maximum production capacity.

4. The AI-powered intelligent supply and demand matching method for PCB procurement platforms as described in claim 1, characterized in that, The deep matching network based on the attention mechanism includes a demand self-attention layer, a supplier self-attention layer, a demand-supplier cross-attention layer, and a multilayer perceptron output layer; the supply-demand matching probability P is calculated as follows: ； Among them, h d h is the demand feature vector. s Here, denoted as the supplier feature vector, CrossAttend represents the cross-attention operation, MLP stands for Multilayer Perceptron, and σ is the sigmoid function.

5. The AI-powered intelligent supply and demand matching method for PCB procurement platforms as described in claim 1, characterized in that, Step S40 includes: when there are multiple unmatched demands, using the real-time load rate ρi of all suppliers and the supply-demand matching probability as input, and using a trained reinforcement learning agent to output a global supply-demand allocation scheme. The reward function R of the reinforcement learning agent includes a load balancing term: ； Among them, S succ To indicate a successful match, T represents the variance of the load factor of each supplier. on To ensure timely delivery and reward fulfillment, α,β,γ These are the preset weighting coefficients.

6. The AI-powered intelligent supply and demand matching method for PCB procurement platforms as described in claim 1, characterized in that, Before step S30, a process pre-screening step is also included: using a pre-built PCB process knowledge graph, the process parameters in the requirements are semantically aligned with the process capability tags of the suppliers, and suppliers with non-coverable process conflicts are filtered out.

7. The AI-powered intelligent supply and demand matching method for PCB procurement platforms as described in claim 1, characterized in that, Following step S40, a quotation anomaly verification step is also included: using a pre-trained quotation regression model, inputting features including at least area, number of layers, and process complexity, to calculate an estimated reasonable price; if the supplier's actual quotation exceeds the preset fluctuation range of the estimated reasonable price, it is marked as an abnormal quotation and a review is triggered.

8. A PCB procurement platform AI intelligent supply and demand precision matching system, characterized in that, include: The requirement encoding module is used to obtain the requirement information submitted by the customer, extract the structured requirement parameters, and generate the requirement feature vector using the first neural network encoder. The supplier coding module is used to acquire attribute data of each candidate supplier, including historical production load sequence, process capacity data, and quality data; using an AI capacity prediction model, the future available capacity of the supplier is predicted based on the historical production load sequence, and combined with the process capacity data and quality data, a supplier feature vector is generated through a second neural network encoder. The matching calculation module has a built-in deep matching network based on the attention mechanism, which is used to input the demand feature vector and the supplier feature vector into the deep matching network based on the attention mechanism to calculate the supply and demand matching probability of the supplier. The matching and allocation module is used to output the supply and demand matching result based on the supply and demand matching probability.

9. A terminal, characterized in that, The terminal includes a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the various steps of the AI-powered intelligent supply and demand matching method for the PCB procurement platform as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the AI-powered intelligent supply and demand matching method for the PCB procurement platform as described in any one of claims 1-7.

Images