Product scheme generation method and device, equipment and storage medium

Abstract

The invention provides a product scheme generation method and device, equipment and a storage medium, and the method comprises the steps: obtaining first text data of a product, and extracting demand information in the first text data based on a first processing model; constructing a knowledge graph based on the demand information, and reasoning the knowledge graph by using a first neural network to generate associated data of the demand information; performing multi-objective solution optimization on the associated data based on a first optimization algorithm, and determining a candidate scheme set corresponding to the first text data; obtaining a product template matched with the candidate scheme set, and filling the product template based on the candidate scheme set to obtain a product scheme set conforming to a product template format; and obtaining an application result of the product scheme set, training a first feedback model based on the application result, and optimizing the first optimization algorithm and the product template based on the trained first feedback model.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and in particular to a product solution generation method, apparatus, device, and storage medium. Background Technology

[0002] Currently, the process for securing orders or projects involves bidders participating in a bidding process, with the tendering party selecting the final partner from among the bidders. Therefore, responding promptly and quickly to the tendering party's needs and developing a competitive bidding proposal is a crucial issue for bidders. During the RFQ (Request for Quotation) stage of the bidding process, issues such as misunderstandings of requirements, low efficiency in proposal generation, and a lack of competitiveness in proposals exist. Many bidders have corresponding IT systems for managing the RFQ process, but their functionality is limited to process management, approval, and document backup during the bidding process. The output of product or project proposals still relies on traditional methods. Based on this situation, there is an urgent need for a more intelligent, faster, and more efficient method or system for generating product and project proposals to help bidders gain a competitive advantage in the RFQ stage. Summary of the Invention

[0003] This disclosure provides a product solution generation method, apparatus, device, and storage medium to at least solve the above-mentioned technical problems existing in the prior art.

[0004] A first aspect of this disclosure provides a method for generating a product solution, the method comprising: Obtain the first text data of the product, and extract the demand information from the first text data based on the first processing model; A knowledge graph is constructed based on demand information, and the first neural network is used to reason about the knowledge graph to generate related data of demand information. Based on the first optimization algorithm, multi-objective optimization is performed on the associated data to determine the candidate solution set corresponding to the first text data; Obtain product templates that match the candidate solution set, populate the product templates based on the candidate solution set, and obtain a product solution set that conforms to the product template format; Obtain the application results of the product solution set, train the first feedback model based on the application results, and optimize the first optimization algorithm and product template based on the trained first feedback model.

[0005] In one possible implementation, extracting demand information from the first text data based on a first processing model includes: Demand information includes explicit and implicit demands; The explicit requirements in the first text data are extracted based on the first model in the first processor model; The first implicit requirement and requirement association weights associated with the explicit requirement are determined based on the second model in the first processor model.

[0006] In one possible implementation, a knowledge graph is constructed based on demand information, including: Obtain technical and resource information corresponding to explicit requirements and primary implicit requirements; The demand layer is constructed based on explicit needs and primary implicit needs; the technology layer is constructed based on technical information; and the resource layer is constructed based on resource information. A knowledge graph is constructed based on the demand layer, technology layer, and resource layer.

[0007] In one possible implementation, a first neural network is used to reason about the knowledge graph to generate related data of demand information, including: The first inference algorithm in the first neural network is used to perform correlation inference on the demand layer, the technology layer and the demand correlation weights to generate correlation supplementary data. Related supplementary data includes: the second implicit requirement and the requirement association matrix; The second implicit requirement is used to supplement the requirement nodes of the requirement layer, and the requirement association matrix is ​​used to supplement the association between the requirement layer and the resource layer. Based on the resource layer to determine the constraint data, the second inference algorithm in the first neural network is used to perform correlation inference on the technical layer, resource layer and constraint data to generate correlation constraint data; Association constraint data includes: resource association matrix and resource constraint matrix; The associated data for demand information includes associated supplementary data and associated constraint data.

[0008] In one possible implementation, multi-objective optimization is performed on the associated data based on a first optimization algorithm to determine a set of candidate solutions corresponding to the first text data, including: A demand weight vector is established based on the demand layer, and a technical parameter vector is established based on the technical layer. First engineering data is generated based on demand weight vector, technical parameter vector, related supplementary data, related constraint data, and knowledge graph. The first optimization function is constructed based on the demand weight vector and the technical parameter vector. The constraints of the first optimization function are set based on the resource constraint matrix, the technical parameter vector, and the demand correlation matrix. Based on the first optimization algorithm, constraints, and first engineering data, the first optimization function is solved to generate a first solution set. The first solution set includes a set of candidate solutions corresponding to the first text data.

[0009] In one possible implementation, a product template is populated based on a set of candidate solutions to obtain a product solution that conforms to the product template format, including: Fill the structured requirements-technology mapping table with the candidate solution set to generate mapping table data; Using the mapping table data and product template as input to the solution generation model, a product solution conforming to the product template format is generated by the solution generation model. The product solution includes: technical solution, business solution, and competitive analysis.

[0010] In one possible implementation, the application results of the product solution are obtained, a first feedback model is trained based on the application results, and a first optimization algorithm and product template are optimized based on the trained first feedback model, including: The application results of the product solution include: historical solution data, external solution result feedback data, and industry dynamic data. Historical solution data includes: resource correlation matrix, demand correlation matrix, and solution-related data. Input the resource association matrix, demand association matrix, solution-related data, external solution result feedback data, and industry dynamic data into the first feedback model to obtain feedback optimization parameters; Feedback optimization parameters include: requirements analysis optimization parameters, technology mapping optimization parameters, and solution generation optimization parameters; Based on the demand analysis optimization parameters, the weight parameters of the first processing model and the association rules of the knowledge graph are adjusted; based on the technology mapping optimization parameters, the first optimization algorithm is adjusted; and based on the solution generation optimization parameters, the product template is adjusted. Customer ratings are obtained based on historical solution data, bidding results are obtained based on external solution feedback data, and the proportion of new technologies is obtained based on industry dynamics. A reward function is constructed based on preset weighting coefficients, winning bid results, customer ratings, and the proportion of new technologies. The feedback optimization parameters are scored based on a reward function, and updated based on the scores and a first feedback model. A second aspect of this disclosure provides a product solution generation apparatus, comprising: The demand analysis unit is used to acquire the first text data of the product, extract the demand information from the first text data based on the first processing model, construct a knowledge graph based on the demand information, and use the first neural network to reason about the knowledge graph to generate related data of the demand information. The technology mapping unit is used to perform multi-objective optimization on the associated data based on the first optimization algorithm to determine the candidate solution set corresponding to the first text data. The solution generation unit is used to obtain product templates that match the candidate solution set, fill the product templates based on the candidate solution set, and obtain a product solution set that conforms to the product template format. The dynamic optimization unit is used to obtain the application results of the product solution set, train the first feedback model based on the application results, and optimize the first optimization algorithm and product template based on the trained first feedback model.

[0011] A third aspect of this disclosure provides an electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods described in this disclosure.

[0012] A fourth aspect of this disclosure provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the methods described in this disclosure.

[0013] This disclosed method, apparatus, device, and storage medium for generating product solutions utilize Graph Convolutional Nueral Network (GCN) and attention mechanisms to parse RFQ text and obtain requirements. It then performs technical mapping on these requirements, generates bidding solutions based on the mapping results and templates, and optimizes the algorithm by incorporating feedback from historical data and solution results. This achieves fully automated processing from requirements analysis to bid document generation, significantly improving bidder response speed and solution competitiveness, thereby increasing the success rate of bids and perfectly meeting the high-efficiency and competitive demands of bidding processes.

[0014] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0015] The above and other objects, features, and advantages of this disclosure will become readily apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. Several embodiments of this disclosure are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0016] Figure 1 A schematic diagram illustrating the implementation flow of a product solution generation method according to an embodiment of this disclosure is shown; Figure 2 A schematic diagram illustrating the implementation flow of another product solution generation method according to an embodiment of this disclosure is shown; Figure 3 This illustration shows a schematic diagram of a product project quotation request quality house according to an embodiment of the present disclosure; Figure 4 A schematic diagram of a product solution generation apparatus according to an embodiment of the present disclosure is shown; Figure 5 A schematic diagram of the composition structure of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0017] To make the objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0018] Figure 1 This illustration shows a schematic diagram of the implementation flow of a product solution generation method according to an embodiment of the present disclosure, such as... Figure 1 As shown, the implementation process of a product solution generation method according to an embodiment of this disclosure includes the following steps: Step 101: Obtain the first text data of the product, and extract the demand information from the first text data based on the first processing model.

[0019] In this embodiment of the disclosure, the product definition and related requirements are as follows: a 15.6-inch laptop is required, weighing ≤1.5kg, with a battery life of ≥12 hours, supporting fast charging, suitable for mobile office, with a budget of ≤8000 yuan, and the first text data (RFQ text) is to be generated within 180 days. The text includes the aforementioned key information.

[0020] In this embodiment of the disclosure, demand information is extracted from the first text data based on a first processing model. The first processing model can be a Natural Language Processing (NLP) model. The demand information includes explicit and implicit demands. Specifically, explicit demands are extracted from the first text data based on a first model within the NLP model. Preferably, the first model is a domain-adaptive Bidirectional Encoder Representations from Transformers (BERT) model based on neural networks. A second model within the NLP model determines the first implicit demand associated with the explicit demand and the demand association weights. The second model can be an attention mechanism model. The first implicit demand is derived solely from the explicit demand through the attention mechanism model. The demand association weights are the association weights between the explicit and implicit demands, used to filter out low-relevance implicit demands. The demand information can be keywords from the aforementioned first text data, such as: size, weight, battery life, fast charging (performance), budget (cost), and delivery (cycle).

[0021] Step 102: Construct a knowledge graph based on the demand information, and use the first neural network to reason about the knowledge graph to generate related data of the demand information.

[0022] In this embodiment, technical and resource information corresponding to explicit requirements and a first implicit requirement are obtained. A requirement layer is constructed based on the explicit and first implicit requirements; a technology layer is constructed based on the technical information; a resource layer is constructed based on the resource information; and a knowledge graph is constructed based on the requirement, technology, and resource layers. The requirement layer is used to classify requirement information, and the technology layer is used to associate engineering characteristics (such as "battery energy density" corresponding to "range time") within the Quality Function Deployment (QFD) house of quality. Figure 3 This illustration shows a schematic diagram of a product item quotation request quality house according to an embodiment of the present disclosure, such as... Figure 3 As shown, the technical requirements and customer requirements in the House of Quality are linked through the content in the technical layer mentioned above. The resource layer is used to obtain technical information from the database, such as accessing the product database, marking technical feasibility (such as the accuracy of existing production lines and the delivery time of supply chain materials), and also to verify the feasibility of parameters in the technical solution and to constrain the solution generation process.

[0023] In this embodiment of the disclosure, after constructing the aforementioned knowledge graph, a first neural network is used to reason about the knowledge graph to generate associated data of demand information. Specifically, the first neural network is a Graph Convolutional Nueral Network (GCN). The first reasoning algorithm (demand-technology association reasoning algorithm) in the first neural network is used to perform association reasoning on the demand layer, the technology layer, and the demand association weights to generate supplementary association data. The supplementary association data includes second implicit demands (i.e., implicit demand nodes in the demand layer) and a demand association matrix (i.e., a demand-technology association matrix). This involves supplementing the demand nodes in the demand layer based on the second implicit demand, and supplementing the relationships between the demand layer and the resource layer based on the demand association matrix. Specifically, the demand-technology association reasoning is implemented through the following algorithm formula:

[0024] in, σ represents the feature vector of technology node t at layer l+1, representing the feature representation of the technology node after calculation at layer l+1; σ represents the activation function (preferably, a rectified linear unit (ReLU), sigmoid, etc. can be used) to introduce nonlinear transformations, enabling the model to learn complex patterns; N(t) represents the set of neighboring nodes of technology node t, which here refers to the set of demand nodes c associated with technology node t; The probability of demand-technology association (calculated through an attention mechanism); The weight matrix of the l-th layer performs a linear transformation on the features of the demand node c, thereby realizing the transformation of the feature space. Let be the feature vector of demand node c at layer l, which is the feature representation of demand node at layer l. Through the above algorithm, the features of adjacent demand nodes c are aggregated (weighted summation and linear transformation), and then processed by the activation function to obtain the feature representation of technology node t at the new layer (layer l+1, i.e. the layer below the current layer), realizing the automatic mapping of implicit demand to technical parameters (e.g., "lightweight" → "carbon fiber material" with a probability of 0.85).

[0025] In this embodiment, based on the resource layer constraint data, a second inference algorithm in the first neural network is used to perform correlation inference on the technology layer, resource layer, and constraint data to generate correlation constraint data. The correlation constraint data includes a resource correlation matrix and a resource constraint matrix. The second inference algorithm is a technology-resource correlation inference, and the constraint data are resource-related requirements in the RFQ, such as "delivery cycle ≤ 180 days" and "cost ≤ 8000 yuan". By performing technology-resource correlation inference on the technology layer data, resource layer data, and constraint data, a resource correlation matrix and a resource constraint matrix are generated. The resource correlation matrix is ​​the same as the technology-resource correlation matrix, and the resource constraint matrix is ​​used to mark the resource bottlenecks of infeasible technologies, such as "carbon fiber shell → insufficient production line precision" and "LG 80Wh battery → delivery time exceeds 60 days". The correlation data for the demand information includes correlation supplementary data and correlation constraint data.

[0026] Step 103: Based on the first optimization algorithm, perform multi-objective optimization on the associated data to determine the candidate solution set corresponding to the first text data.

[0027] In this embodiment, the first optimization algorithm is the Pareto optimization algorithm. Multi-objective optimization of the associated data is performed based on the Pareto optimization algorithm. Specifically, a demand weight vector is established based on the demand layer, and a technical parameter vector is established based on the technology layer. The demand weights can be adjusted according to the importance of different demands. For example, the demand weight vector... =[ The technical parameter vector T = [t1, t2, ..., tn]. First engineering data is generated based on the demand weight vector, technical parameter vector, associated supplementary data, associated constraint data, and knowledge graph. Preferably, the first engineering data is the relevant data of the House of Quality in the aforementioned quality function deployment, i.e., a House of Quality corresponding to the product solution is generated based on the demand weight vector, technical parameter vector, and knowledge graph. A first optimization function is constructed based on the demand weight vector and technical parameter vector. Constraints of the first optimization function are set based on the resource constraint matrix, technical parameter vector, and demand association matrix. Specifically, the first optimization function is a multi-objective optimization function, for example:

[0028] Set constraints:

[0029] Where R is the resource constraint matrix, T is the technical parameter vector, and R max This is the resource upper limit vector. A C-T This is a demand-technology correlation matrix, representing the strength of the correlation between customer needs and technical parameters; T is a demand weight vector, representing the importance of customer needs; min This is the minimum threshold vector for technical parameters, representing the minimum requirements that the technical parameters must meet.

[0030] In this embodiment, a first optimization function is solved based on a first optimization algorithm, constraints, and first engineering data to generate a first solution set. The first optimization algorithm is the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) algorithm within the Pareto optimization algorithm. The first solution set is a Pareto front solution set, which includes a set of candidate solutions corresponding to the first text data. Specifically, the NSGA-II algorithm is used to solve the Pareto front solution set. Through fast non-dominated sorting and crowding calculation, a Pareto front solution set P={T} is generated. p T c T b …T n}, generate multiple versions of the solution (such as the high-performance solution T) p Low-cost solution T c Equilibrium Scheme T b (This is for later selection.)

[0031] Step 104: Obtain the product template that matches the candidate solution set, fill the product template based on the candidate solution set, and obtain the product solution that conforms to the product template format.

[0032] In this embodiment of the disclosure, a structured demand-technology mapping table is filled based on the candidate solution set to generate mapping table data. Preferably, the candidate solution set (Pareto front solution set) is decomposed into "technical parameters + solution type + feasibility data" and associated with demand weights and resource constraint matrices to form row data of the mapping table, such as "low-cost solution" corresponding to "magnesium alloy shell + 70Wh domestic battery", and marked "resource feasibility 0.85". Furthermore, the mapping table data and product templates are used as input to the solution generation model to obtain a product solution that conforms to the product template format generated by the solution generation model. The product template can be a dynamic template, used to provide a standardized format framework and industry-specific modules, constraining the content structure and format. The solution generation model is an attention mechanism (Transformer sequence) sequence generation model. By inputting the structured mapping results, a product solution conforming to the product template format (preferred, conforming to the RFQ format) generated by the attention mechanism sequence generation model is obtained. The product solution includes: a technical solution, a business solution, and a competitive analysis. For example: Technical solution: core parameter table (e.g., "battery energy density ≥ 200Wh / kg"), design drawing links; Business solution: cost decomposition formula: total cost = ∑(material cost × (1 + process complexity coefficient)), delivery plan Gantt chart; Competitive analysis: comparison matrix with competitors (e.g., "This solution reduces cost by 15% and increases range by 20%").

[0033] Step 105: Obtain the application results of the product solution, train the first feedback model based on the application results, and optimize the first optimization algorithm and product template based on the trained first feedback model.

[0034] In this embodiment, the first feedback model is a Proximal Policy Optimization (PPO) algorithm. PPO includes a state space and an action space. The state space is used to obtain: the application results of the product solution, including: historical solution data, external solution result feedback data, and industry dynamic data. The historical solution data includes: the aforementioned resource association matrix, the aforementioned demand association matrix, and solution-related data. The solution-related data includes: the structured text vector of the tender document, cost decomposition coefficient, differentiated labels of competitive analysis, feature vector of demand information, demand weight vector, association probability, the aforementioned technical parameter vector, weight of the first optimization function, and feasibility probability. The solution result feedback data includes: the winning bid result and contract amount. The industry dynamic data includes: competitor parameter update data, policy and standard change data, and customer demand trend data. The resource association matrix, demand association matrix, solution-related data, external solution result feedback data, and industry dynamic data are input into the first feedback model to obtain feedback optimization parameters. These parameters include demand analysis optimization parameters, technology mapping optimization parameters, and solution generation optimization parameters. Based on the demand analysis optimization parameters, the weight parameters of the first processing model and the association rules of the knowledge graph are adjusted. Based on the technology mapping optimization parameters, the first optimization algorithm is adjusted. Based on the solution generation optimization parameters, the product template is adjusted. Specifically, historical solution data, external solution result feedback data, and industry dynamic data are used as environmental information inputs to the state space to achieve a complete observation of the environment and provide a computational basis for the reward function. The action space in the PPO is based on the input of the above environmental information to obtain demand analysis optimization parameters, technology mapping optimization parameters, and solution generation optimization parameters.

[0035] In this embodiment of the disclosure, customer scores are obtained based on historical solution data, bidding results are obtained based on external solution result feedback data, and the proportion of new technologies is obtained based on industry dynamics; a reward function is constructed based on preset weight coefficients, bidding results, customer scores, and the proportion of new technologies; the feedback optimization parameters are scored based on the reward function, and the feedback optimization parameters are updated based on the scores of the feedback optimization parameters and the first feedback model.

[0036] Figure 2 This illustration shows a schematic diagram of the implementation flow of another product solution generation method according to an embodiment of this disclosure, such as... Figure 2 As shown, the implementation process of another product solution generation method in this disclosure includes the following steps: Step 201: Analyze the demand information of the quotation request.

[0037] In this embodiment, an RFQ for a laptop project is used as an example. A laptop development project requires bidding, and the client needs to provide a product development plan as soon as possible to start the project. First, a requirements analysis is performed. Based on the product definition documents (i.e., the RFQ text) provided by the client, a domain-adaptive BERT model is used to parse the RFQ text, extracting explicit requirements (such as "operating temperature -20℃~80℃") and associating implicit requirements (such as "outdoor scenarios" → "waterproof rating IP67") through an attention mechanism. Further, a knowledge graph is constructed, and the knowledge graph is associated with the aforementioned laptop RFQ's quality house (or similar structure). Figure 3 The illustration shows a schematic diagram of a product project quotation request quality house (QFD) according to an embodiment of this disclosure. Based on the explicit requirements, the QFD quality house, and the RFQ text, the implicit requirements of the laptop computer are inferred. For example, the knowledge graph nodes include: requirements (mobile office → eye-protection screen, interface expansion), technologies (battery capacity, CPU power consumption, casing material), and resources (production line cycle, material inventory). Then, the association weights α are calculated: mobile office → eye-protection screen (α=0.6), interface expansion (α=0.7), generating a requirement-technology matrix. .

[0038] Step 202: Match requirements and generate multiple preliminary solutions.

[0039] In this embodiment, the objectives and constraints are first defined, and the requirement weights C are set as follows: performance (0.4), cost (0.3), and delivery (0.3). Technical parameters T are defined as follows: battery (80Wh / LG vs. 70Wh / domestic), CPU (i7-1360U vs. Si5-1340U), casing (carbon fiber vs. magnesium alloy), screen (2.8K vs. 2K eye-protection), interfaces (4 vs. 3), cost (≤7500 yuan), and delivery (≤65 days). Based on the requirement weights and technical parameters, the NSGA-II algorithm is used to solve for the following: f 1 (Cost): Minimize the Bill of Materials (BOM) (7500 yuan constraint); f 2 (Performance): Battery life × 0.4 + Interfaces × 0.3 + Eye protection × 0.3; f3(Delivery): |Cycle-180| / 180 (Minimize latency). Constraints: CPU power consumption ≤15W, casing weight ≤0.3kg (total weight ≤1.5kg). The final Pareto solution set is generated as follows: Option 1 (Performance Priority): 80Wh battery, i7 CPU, carbon fiber casing, 2.8K screen (cost 7500, delivery 170 days, performance 0.92). Option 2 (Cost Priority): 70Wh battery, i5 CPU, magnesium alloy casing, 2K screen (cost 6900, delivery 185 days, performance 0.85). The effects of the new technology application are simulated using a GAN model, for example, by setting a filter condition: check material inventory (LG battery is sufficient, magnesium alloy casing weight can be optimized to 0.3kg, pass).

[0040] Step 203: Generate a structured scheme.

[0041] In this embodiment, based on the preliminary plan in step 202, a laptop bidding template is loaded and automatically filled in to generate the following plan: Technical plan: parameter table (weight 1.48kg, battery life 11.5h), process (magnesium alloy unibody molding, domestic battery cell fast charging). Business plan: price of 7999 yuan (within budget), delivery time of 172 days (optimized production line priority, generated adversarial network (GAN) and re-verified). Simultaneously, competitive analysis is performed: comparison with similar competing models (0.04kg lighter, 500 yuan lower price, 7-day faster delivery). Based on the filled-in plan, a tender document is generated, including a technical white paper, price list, competitor comparison, and compliance verification is performed, such as: budget (7999≤8000), delivery time (59≤60), and technical parameters (all meet RFQ).

[0042] Step 204: Use reinforcement learning to drive policy evolution.

[0043] In this embodiment, a closed-loop feedback mechanism is constructed, namely, a PPO-based agent training framework is built to achieve a self-optimizing cycle of "strengthening effective strategies for successful bids and eliminating inefficient strategies for unsuccessful bids". The PPO agent training framework includes: a state space and an action space. The state space (S) is the input of environmental information, that is, the agent's complete observation of the current environment. The environmental information specifically includes: historical solution characteristics (S1): demand analysis results (e.g., extracted weight ≤ 1.5kg, battery life ≥ 12h), technology mapping parameters (e.g., battery type in the Pareto solution set), and solution text vectors (e.g., term frequency-inverse document frequency (TF-IDF) features of the tender documents). Solution result feedback (S2): successful / unsuccessful bid (I), customer rating (1-5 points), contract amount (quantified as cost deviation rate). Industry dynamics (S3): competitor parameter updates (e.g., the number of new interfaces added to a certain brand's products), policy and standard changes (e.g., environmental certification updates), and customer demand trends (e.g., the requirement for battery life of thin and light laptops has increased to 14h). The state space provides input to the action space (e.g., when the customer rating is low, the action tends to adjust the weight of the demand analysis); at the same time, the winning bid result (S2) and industry dynamics (S3) in S directly affect the output of the reward function.

[0044] In this embodiment, the Action Space (A) is the agent's decision output, i.e., the set of policy adjustment operations that the PPO algorithm can execute based on the input of the state space, used to optimize the solution generation process. This part includes: Demand parsing adjustment (A1): adjusting NLP model weights (e.g., increasing the attention weight θnlp for the keyword "battery life") and knowledge graph association rules (e.g., adding edge weights for "fast charging → high-power adapter"). Technology mapping optimization (A2): modifying multi-objective optimization parameters (e.g., increasing the cost weight λcost) and updating the resource constraint matrix (e.g., compressing the production line delivery cycle from 65 days to 60 days). Solution generation template update (A3): loading new industry templates (e.g., adding a "drop-proof design" module to the education industry) and adjusting competitive analysis dimensions (e.g., adding a comparison of "after-sales service response time"). The output of the Action Space is a policy mapping of the input of the state space. For example, when a customer repeatedly fails to win bids for cost-sensitive projects, the action triggers an increase in cost weight; after the action is executed, a new state is generated (e.g., generating a low-price solution after adjusting the cost weight), which in turn affects the reward calculation.

[0045] In this embodiment of the disclosure, the reward function is designed using the success rate, customer rating, and innovation as indicators:

[0046] Where R represents the reward value obtained by the agent, which is used to guide the optimization of the reinforcement learning policy (the higher the value, the better the corresponding action). , , These are weighted coefficients for the success rate, customer rating, and innovation, respectively. The sum of these three coefficients is 1. These coefficients can be adjusted according to the bidder's priority; for example, if the bidder is given priority, a weighting can be set. ≥0.5. This is an indicator function (1 for a successful bid, 0 for a failed bid), which directly rewards successful bids (the reward is negative for unsuccessful bids, indicated by a symbol). Control, if If the score is >0 and the bid is not won, then R's contribution to the corresponding weight will be reduced. The client's quantitative evaluation of the solution (e.g., 1-5 points, standardized to 0-100 points before being included in the calculation to ensure consistency with other indicators) and the proportion of technologies used for the first time in the solution (e.g., technologies not recorded in the bidder's patent database) in the total technology portfolio (0-1, the higher the proportion, the higher the reward, encouraging solution innovation). Traditional bidding solution generation tools (such as manually-driven or simple template tools) lack this data-driven reward mechanism. This solution quantifies the winning result into a reward through an indicator function, driving the intelligent agent to self-optimize and reflecting the core advantages of AI technology (learning from historical data, iterating strategies).

[0047] In this embodiment of the disclosure, the agent policy parameter θ is updated using the PPO algorithm to optimize the demand parsing weights, technology mapping rules, etc., and the agent policy is updated using the following formula:

[0048] in, This is a function of the policy parameters θ, used to measure the quality of the current policy. Optimizing this function updates the policy parameters, allowing the policy to better adapt to the environment and obtain more rewards. θ represents the agent's policy parameters (such as demand parsing weights and technology mapping rule hyperparameters), which are optimized through gradient updates. The expression represents the expectation for time step t, which means averaging the subsequent expressions over multiple time steps, taking into account the overall interaction between the agent and the environment at different time steps. This represents the probability ratio between the old and new strategies, measuring the difference between the current strategy and historical strategies. It is the advantage function (e.g., calculated using GAE), representing the value advantage of an action (reward minus baseline). Clip is the clipping probability ratio (...). To prevent excessive policy updates and ensure stable training, a range (e.g., 0.2) is set. Updating θ allows the agent to learn better actions (e.g., adjusting demand analysis weights and optimizing technology mapping rules), improving the quality of solution generation (e.g., success rate and customer satisfaction). This formula uses the PPO algorithm to stably iterate the policy parameter θ, optimizing rules such as demand analysis and technology mapping, enabling the agent to continuously evolve in RFQ solution generation and improve bidding competitiveness. The PPO algorithm continuously adjusts the policy parameter θ to achieve the objective function... Maximize and optimize the agent's strategy. By using a pruning mechanism and leveraging the advantage function, the algorithm can effectively learn new strategies while avoiding model collapse due to overly aggressive policy updates. This allows for more stable and efficient training of the agent, enabling it to gradually learn better behavioral strategies in complex environments (such as dealing with various demands and competitive situations in generating RFQ solutions).

[0049] In this embodiment, the application of the above algorithm is as follows: Reward calculation is performed on the solution: If the bid is successful (I=1), the customer score is 4.5. Combining this with the set weighting coefficient, the reward function R=0.6×1+0.3×4.5=1.95 (innovation degree 0, due to lack of new technology). The PPO updates according to the reward function: The cost weight (ω) of "budget-sensitive" customers is increased. cost =0.35), the next priority is the cost-based solution. Then, strategy optimization is performed: if the bid is not won (e.g., by a competitor with a 2K resolution screen), the screen resolution weight is adjusted (+0.1), a 2.5K resolution screen solution is generated, and the solution is verified and calculated to yield: the laptop project cost is 7400 yuan, the quoted price is 7999 yuan, and the performance is 0.9, which meets the budget.

[0050] In this embodiment, the above steps achieve end-to-end intelligent processing from requirements analysis to bid document generation in a laptop project RFQ scenario, significantly improving the bidder's response speed and solution competitiveness, thereby increasing the success rate and perfectly adapting to the high-efficiency and competitive requirements of bidding. The intelligent agent built based on this method only requires targeted updates or modifications to the agent's knowledge graph and reinforcement learning training to be applicable to products or projects in other industries. Any modifications based on this framework according to industry or product characteristics should be considered within the scope of protection of this solution.

[0051] This disclosed embodiment belongs to a multi-technology integrated intelligent agent architecture. It uses NLP and knowledge graphs to achieve requirement parsing and implicit requirement mining, then solves for Pareto optimization results, and finally generates multiple version solutions through GANs. Simultaneously, it drives closed-loop self-optimization through reinforcement learning (PPO). This solution requires the construction of a three-layer dynamic knowledge graph of "requirements-technology-resources," supporting implicit requirement reasoning and automatic completion of technical paths. Multiple strategy solutions (such as performance-first and cost-first) are dynamically generated based on real-time RFQ requirements. Competitive analysis and compliance verification are also embedded, ultimately generating the optimal RFQ project solution.

[0052] Figure 4 A schematic diagram of a product solution generation apparatus according to an embodiment of the present disclosure is shown, such as... Figure 4 As shown, an embodiment of this disclosure provides a product solution generation apparatus comprising: The requirement analysis unit 401 is used to acquire the first text data of the product, extract the requirement information from the first text data based on the first processing model, construct a knowledge graph based on the requirement information, and use the first neural network to reason about the knowledge graph to generate related data of the requirement information.

[0053] The requirement analysis unit 401 is also used to: extract explicit requirements and implicit requirements from the first text data based on the first model in the first processor model; and determine the first implicit requirements associated with the explicit requirements and the requirement association weights based on the second model in the first processor model.

[0054] The demand analysis unit 401 is also used to obtain technical and resource information corresponding to explicit demands and the first implicit demand; construct a demand layer based on explicit demands and the first implicit demand, construct a technology layer based on technical information, construct a resource layer based on resource information; and construct a knowledge graph based on the demand layer, technology layer, and resource layer.

[0055] The demand analysis unit 401 is further configured to: use the first inference algorithm in the first neural network to perform correlation inference on the demand layer, the technology layer, and the demand correlation weights to generate correlation supplementary data; the correlation supplementary data includes: a second implicit demand and a demand correlation matrix; supplement the demand nodes of the demand layer based on the second implicit demand, and supplement the correlation between the demand layer and the resource layer based on the demand correlation matrix; determine the constraint data based on the resource layer, and use the second inference algorithm in the first neural network to perform correlation inference on the technology layer, the resource layer, and the constraint data to generate correlation constraint data; the correlation constraint data includes: a resource correlation matrix and a resource constraint matrix; the correlation data of demand information includes correlation supplementary data and correlation constraint data.

[0056] The technology mapping unit 402 is used to perform multi-objective optimization on the associated data based on the first optimization algorithm to determine the candidate scheme set corresponding to the first text data.

[0057] The technology mapping unit 402 is further configured to: establish a demand weight vector based on the demand layer and a technology parameter vector based on the technology layer; generate first engineering data based on the demand weight vector, technology parameter vector, associated supplementary data, associated constraint data, and knowledge graph; construct a first optimization function based on the demand weight vector and technology parameter vector, and set constraints on the first optimization function based on the resource constraint matrix, technology parameter vector, and demand association matrix; solve the first optimization function based on the first optimization algorithm, constraints, and first engineering data to generate a first solution set; the first solution set includes a set of candidate solutions corresponding to the first text data.

[0058] The solution generation unit 403 is used to obtain product templates that match the candidate solution set, fill the product templates based on the candidate solution set, and obtain a product solution set that conforms to the product template format.

[0059] The solution generation unit 403 is also used to fill the structured requirement-technology mapping table based on the candidate solution set and generate mapping table data; and to use the mapping table data and product template as input to the solution generation model to obtain a product solution that conforms to the product template format generated by the solution generation model; the product solution includes: technical solution, business solution and competitive analysis.

[0060] The dynamic optimization unit 404 is used to obtain the application results of the product solution set, train the first feedback model based on the application results, and optimize the first optimization algorithm and product template based on the trained first feedback model.

[0061] The dynamic optimization unit 404 is also used to: The application results of the product solution include: historical solution data, external solution result feedback data, and industry dynamic data. Historical solution data includes: resource association matrix, demand association matrix, and solution-related data. The resource association matrix, demand association matrix, solution-related data, external solution result feedback data, and industry dynamic data are input into the first feedback model to obtain feedback optimization parameters. These parameters include: demand analysis optimization parameters, technology mapping optimization parameters, and solution generation optimization parameters. Based on the demand analysis optimization parameters, the weight parameters of the first processing model and the association rules of the knowledge graph are adjusted; based on the technology mapping optimization parameters, the first optimization algorithm is adjusted; and based on the solution generation optimization parameters, the product template is adjusted. Customer scores are obtained based on historical solution data; bidding results are obtained based on external solution result feedback data; and the proportion of new technologies is obtained based on industry dynamics. A reward function is constructed based on preset weight coefficients, bidding results, customer scores, and the proportion of new technologies. The feedback optimization parameters are scored based on the reward function, and the feedback optimization parameters are updated based on the scores and the first feedback model.

[0062] In an exemplary embodiment, the requirements analysis unit 401, technology mapping unit 402, solution generation unit 403, dynamic optimization unit 404, etc., may be implemented by one or more central processing units (CPUs), graphics processing units (GPUs), application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontroller units (MCUs), microprocessors, or other electronic components.

[0063] Regarding the apparatus in the above embodiments, the specific manner in which each module and unit performs its operations has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0064] According to embodiments of this disclosure, this disclosure also provides an electronic device and a readable storage medium.

[0065] Figure 5 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0066] like Figure 5As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0067] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0068] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as a product proposal generation method. For example, in some embodiments, a product proposal generation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of a product proposal generation method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform a product proposal generation method by any other suitable means (e.g., by means of firmware).

[0069] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0070] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0071] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0072] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0073] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0074] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0075] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0076] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0077] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A method for generating a product solution, characterized in that, The method includes: Obtain the first text data of the product, and extract the demand information from the first text data based on the first processing model; A knowledge graph is constructed based on the demand information, and the first neural network is used to reason about the knowledge graph to generate associated data of the demand information. Based on the first optimization algorithm, multi-objective optimization is performed on the associated data to determine the candidate solution set corresponding to the first text data; Obtain a product template that matches the candidate solution set, and populate the product template based on the candidate solution set to obtain a product solution set that conforms to the product template format; Obtain the application results of the product solution set, train a first feedback model based on the application results, and optimize the first optimization algorithm and the product template based on the trained first feedback model.

2. The product solution generation method according to claim 1, characterized in that, The step of extracting the demand information from the first text data based on the first processing model includes: The demand information includes explicit demands and implicit demands; The explicit requirements in the first text data are extracted based on the first model in the first processor model; The first implicit requirement and requirement association weight associated with the explicit requirement are determined based on the second model in the first processor model.

3. The product solution generation method according to claim 2, characterized in that, The construction of the knowledge graph based on the demand information includes: Obtain the technical and resource information corresponding to the explicit requirement and the first implicit requirement; A requirement layer is constructed based on the explicit requirements and the first implicit requirements; a technology layer is constructed based on the technical information; and a resource layer is constructed based on the resource information. The knowledge graph is constructed based on the demand layer, the technology layer, and the resource layer.

4. The product solution generation method according to claim 2, characterized in that, The step of using a first neural network to reason about the knowledge graph and generate associated data for the demand information includes: The first inference algorithm in the first neural network is used to perform correlation inference on the demand layer, the technology layer and the demand correlation weight to generate correlation supplementary data; The supplementary data includes: a second implicit requirement and a requirement association matrix; The requirement nodes of the requirement layer are supplemented based on the second implicit requirement, and the association between the requirement layer and the resource layer is supplemented based on the requirement association matrix; Based on the resource layer, constraint data is determined, and the second inference algorithm in the first neural network is used to perform correlation inference on the technology layer, the resource layer, and the constraint data to generate correlation constraint data. The associated constraint data includes: a resource association matrix and a resource constraint matrix; The associated data of the demand information includes the associated supplementary data and the associated constraint data.

5. The product solution generation method according to claim 4, characterized in that, The step of performing multi-objective optimization on the associated data based on the first optimization algorithm to determine the candidate solution set corresponding to the first text data includes: A demand weight vector is established based on the demand layer, and a technical parameter vector is established based on the technology layer; First engineering data is generated based on the demand weight vector, the technical parameter vector, the associated supplementary data, the associated constraint data, and the knowledge graph; A first optimization function is constructed based on the demand weight vector and the technical parameter vector, and the constraints of the first optimization function are set based on the resource constraint matrix, the technical parameter vector and the demand correlation matrix. Based on the first optimization algorithm, the constraints, and the first engineering data, the first optimization function is solved to generate a first solution set. The first solution set includes a set of candidate solutions corresponding to the first text data.

6. The product solution generation method according to claim 1, characterized in that, The step of filling the product template with the candidate solution set to obtain a product solution that conforms to the product template format includes: Based on the candidate solution set, populate the structured requirements-technology mapping table to generate mapping table data; Using the mapping table data and the product template as input to the solution generation model, a product solution that conforms to the product template format is generated by the solution generation model. The product plan includes: technical solutions, business solutions, and competitive analysis.

7. The product solution generation method according to claim 4, characterized in that, The steps of obtaining the application results of the product solution, training a first feedback model based on the application results, and optimizing the first optimization algorithm and the product template based on the trained first feedback model include: The application results of the product solution include: historical solution data, external solution result feedback data, and industry dynamic data. The historical solution data includes: the resource association matrix, the demand association matrix, and solution-related data. Input the resource association matrix, the demand association matrix, the solution-related data, the external solution result feedback data, and the industry dynamic data into the first feedback model to obtain feedback optimization parameters; The feedback optimization parameters include: demand analysis optimization parameters, technology mapping optimization parameters, and solution generation optimization parameters; The weight parameters of the first processing model and the association rules of the knowledge graph are adjusted based on the demand parsing optimization parameters; the first optimization algorithm is adjusted based on the technology mapping optimization parameters; and the product template is adjusted based on the scheme generation optimization parameters. Customer ratings are obtained based on the historical solution data; bidding results are obtained based on the external solution result feedback data; and the proportion of new technologies is obtained based on the industry dynamics. A reward function is constructed based on preset weighting coefficients, the winning bid results, the customer rating, and the proportion of new technologies. The feedback optimization parameters are scored based on the reward function, and the feedback optimization parameters are updated based on the scores of the feedback optimization parameters and the first feedback model.

8. A product solution generation device, characterized in that, The device includes: The demand analysis unit is used to acquire the first text data of the product, extract the demand information from the first text data based on the first processing model, construct a knowledge graph based on the demand information, and use a first neural network to reason about the knowledge graph to generate the associated data of the demand information. The technology mapping unit is used to perform multi-objective optimization on the associated data based on the first optimization algorithm to determine the candidate scheme set corresponding to the first text data; A solution generation unit is used to obtain a product template that matches the candidate solution set, and fill the product template based on the candidate solution set to obtain a product solution set that conforms to the product template format. The dynamic optimization unit is used to obtain the application results of the product solution set, train a first feedback model based on the application results, and optimize the first optimization algorithm and the product template based on the trained first feedback model.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images