A technology avoidance design method based on large language model

Abstract

The present application provides a technology avoidance design method based on a large language model, including: using natural language processing methods to perform semantic analysis on existing technical texts, combining the decentralized characteristics of blockchain, using a hierarchical clustering algorithm to perform multi-dimensional and hierarchical division of technical data, converting technical data into ordered functionally distinct knowledge modules, and constructing a knowledge base from a number of knowledge modules; continuously tracking the upgrading of technology, dynamically managing and optimizing the knowledge base, and continuously enriching the optimization strategies of local coordination and global diversion by regularly updating patent data, re-evaluating technical solutions, and adjusting optimization algorithm parameters.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a technology avoidance design method based on a large language model. Background Art

[0002] For product design companies, innovation ability is specifically reflected in the comprehensive ability to generate innovative design solutions. Although the experience and knowledge of designers are the source of innovative design solutions, they also limit the innovative design solutions generated by designers to a certain extent, because designers are easily limited to existing knowledge and experience, and it is difficult to generate highly innovative design solutions. In addition, the country now attaches importance to intellectual property protection. If the products designed with great effort fall into the scope of patent protection of others, it will be a great loss for the company. In the process of avoiding coexistence of patent technology based on large models, it is necessary to balance the functional complementarity and independence between various patent technical solutions. Functional complementarity helps to give full play to their respective technical advantages and achieve overall performance improvement. However, excessive pursuit of functional complementarity may lead to insufficient independence between patent technologies, thereby causing patent conflicts. Specifically, when we try to combine multiple patent technologies to build a new technical solution, there may be a certain degree of overlap in the technical details between the patents. If handled improperly, it may trigger a patent conflict. This contradiction brings great challenges to the design and optimization of technical solutions. On the one hand, we hope to maximize the complementary advantages of each patent through clever design. On the other hand, we must always be vigilant to avoid conflicts caused by overlap. While pursuing performance, we must also maintain the uniqueness of the implementation as much as possible. This requires us to carefully analyze and weigh all aspects of the technical solution to find the best balance point to achieve patent coexistence. Summary of the invention

[0003] The present invention provides a technology avoidance design method based on a large language model, which mainly includes:

[0004] Use natural language processing methods to perform semantic analysis on existing technical texts, combine the decentralized characteristics of blockchain, use hierarchical clustering algorithms to divide technical data into multiple dimensions and hierarchies, transform technical data into ordered knowledge modules with different functions, and construct a knowledge base from several knowledge modules;

[0005] Preset the knowledge base technology similarity calculation method, use the DBSCAN algorithm to perform technical analysis on the patent group, identify the core characteristics of the technology, split the technology into principle set and key set, use the Bloom filter algorithm to build a potential conflict set, map the key parameters of each technology to the conflict set, obtain the conflict technology conflict set, use the ant colony algorithm to locally explore, improve the technical parameter values ​​through multiple iterations, and obtain a local coordination temporary solution;

[0006] Based on the temporary solution obtained, the particle swarm optimization algorithm is used to generate an optimized adjustment plan by locally adjusting the value range of technical parameters, find the best parameter combination to solve the technical conflict, build a local harmonious coexistence technical solution, add it to the knowledge base, and mark the resolved technical conflicts in the technical conflict set;

[0007] Determine whether the patent conflict set is empty. If not, use the support vector machine algorithm to analyze the remaining technical characteristics, establish a multi-dimensional weighted evaluation function to calculate and predict the optimal application scenario and combination method, and obtain an evaluation score. According to the evaluation score, the conflicting technologies are introduced to different application fields to achieve complementary coexistence at the global level, obtain the optimized coexistence technical solution, and add the optimized coexistence technical solution to the knowledge base;

[0008] Using the random forest algorithm, we construct an evaluation index system from several key perspectives, including technological innovation and market acceptance. We quantitatively score the coexistence technology solutions in the knowledge base, simulate the application scenarios of the coexistence technology solutions, build a virtual environment model, simulate the performance of the coexistence technology solutions under different conditions, calculate the evaluation score, measure the deviation between the evaluation score and the actual score, and adjust the evaluation model parameters.

[0009] Adopting multi-objective optimization genetic algorithm, by constructing a fitness function including technical reliability, cost control, and market competitiveness, the optimal coexistence technical solution set is optimized, and the optimal technical combination solution that takes into account local coordination and global complementarity is searched and identified. According to specific application requirements, the most suitable solution is selected from the optimal patent technology combination solution, and the technical knowledge base is updated;

[0010] We continue to track technological updates, dynamically manage and optimize the knowledge base, and constantly enrich the optimization strategies of local coordination and global diversion by regularly updating patent data, re-evaluating technical solutions, and adjusting optimization algorithm parameters.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a technology avoidance design method based on a large language model. The method first performs semantic analysis on the technical text and constructs a knowledge module library using a hierarchical clustering algorithm. Then, the core features of the technology are identified by the DBSCAN algorithm, and a potential conflict set is constructed using a Bloom filter. For conflicting technologies, an ant colony algorithm and a particle swarm optimization algorithm are used to generate local coordination solutions. For the remaining conflicts, a support vector machine is used to analyze the technical characteristics, and a multi-dimensional evaluation function is designed to predict the optimal application scenario to achieve global complementary coexistence. The present invention also uses a random forest algorithm to construct an evaluation index system and simulates the performance of the solution through digital twin technology. Finally, a multi-objective optimization genetic algorithm is used to search for the optimal technology combination that takes into account local coordination and global complementarity. The present invention realizes the intelligent solution of patent technology conflicts by integrating multiple algorithms, and promotes technological innovation and coordinated development. The method can not only effectively resolve local technical conflicts, but also achieve technical complementarity at the global level, greatly improving the application value and market competitiveness of patent technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention is a flowchart of a technical avoidance design method based on a large language model.

[0014] Figure 2 A schematic diagram of a technical avoidance design method based on a large language model of the present invention.

[0015] Figure 3 This is another schematic diagram of a technology avoidance design method based on a large language model of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1-3 In this embodiment, a technical avoidance design method based on a large language model may specifically include:

[0018] Step S101, use natural language processing methods to perform semantic analysis on existing technical texts, combine the decentralized characteristics of blockchain, and use hierarchical clustering algorithms to divide technical data into multiple dimensions and hierarchies. Convert technical data into ordered knowledge modules with different functions, and construct a knowledge base from several knowledge modules.

[0019] Perform part-of-speech tagging and dependency syntax analysis on technical texts, extract keywords and core concepts, and generate text semantic vector representation. Based on the text semantic vector, use a hierarchical clustering algorithm to perform multi-level clustering on technical data to obtain a hierarchical technical knowledge tree structure. For the hierarchical technical knowledge tree structure, divide the technical data into independent knowledge modules, assign a unique identifier to each knowledge module, define the relationship between knowledge modules and the execution order, and construct a knowledge module dependency graph. Serialize the knowledge modules into serialized data; use a distributed hash table to store the serialized knowledge modules, and synchronously update the knowledge base status in the blockchain network.

[0020] Exemplarily, the technical text is tagged with part of speech and parsed with dependency syntax. The NLTK library is used to perform part of speech tagging, and Stanford Parser is used to build a semantic relationship network between words, extract keywords and core concepts, and generate a text semantic vector representation. Based on the text semantic vector, the AGNES agglomerative hierarchical clustering algorithm is used to perform multi-level clustering of technical data, calculate the Euclidean distance as the distance measure between classes, set the clustering threshold and iteratively calculate to form a hierarchical technical knowledge tree structure. According to the knowledge tree structure, the technical data is divided into independent knowledge modules, each module is assigned a unique identifier, the relationship between modules and the execution order are defined, and the knowledge module dependency graph is constructed. At the same time, metadata information and version control mechanism are added to each module. The knowledge module is serialized into JSON format, and the Merkle tree structure is used to organize the data to ensure data integrity and verifiability. The serialized knowledge module is stored in a distributed hash table, and the knowledge base status is synchronously updated in the blockchain network through the PoW consensus algorithm to realize a decentralized knowledge base management and sharing mechanism. The blockchain node verifies the hash value of the knowledge module and the Merkle tree root hash to ensure data consistency and immutability. In the process of technical text processing, we first use the NLTK library to segment and tag the text. For example, we can tag the sentence "Natural language processing technology is widely used" as [('Natural language

[0021] ','n'),('processing','v'),('technology','n'),('application','v'),('extensive','a')]. Then, Stanford Parser was used to construct a dependency syntactic tree, extract the subject-verb-object relationship, and generate a semantic relationship network. The importance of words was calculated by the TF-IDF algorithm, and the 10 words with the highest weight were selected as keywords. Then, the text was converted into a 300-dimensional semantic vector based on the Word2Vec model. Then, the AGNES hierarchical clustering algorithm was used to process the semantic vector. Initially, each vector was regarded as an independent category, and the Euclidean distance was calculated as the distance between classes. The clustering threshold was set to 0.5. When the distance between classes was less than the threshold, the categories were merged. The iteration was carried out until the merging could not be continued, and finally a multi-level knowledge tree was formed. According to the knowledge tree structure, the technical data was divided into independent knowledge modules, each module was assigned a 32-bit unique identifier, the pre-, post- and parallel relationships between modules were defined, and a directed acyclic graph was constructed to represent the knowledge module dependency. Metadata such as creation time, author, and version number were added to each module to implement the version control mechanism. The knowledge module is serialized into JSON format, the module hash value is calculated using the SHA-256 algorithm, and a Merkle tree is constructed to ensure data integrity. The serialized knowledge module is stored using a distributed hash table (DHT). The key in the DHT is the module identifier and the value is JSON data. In the blockchain network, nodes compete for the right to record accounts through the PoW consensus algorithm. The winning node packages the new knowledge module into a block and broadcasts it to the network. Other nodes verify the hash value of the knowledge module in the block and the Merkle tree root hash to ensure data consistency and immutability.

[0022] Step S102, preset the knowledge base technology similarity calculation method, use the DBSCAN algorithm to perform technical analysis on the patent group, identify the core characteristics of the technology, split the technology into principle set and key set, use the Bloom filter algorithm to build a potential conflict set, map the key parameters of each technology to the conflict set, obtain the conflict technology conflict set, use the ant colony algorithm for local exploration, improve the technical parameter values ​​through multiple iterations, and obtain a local coordination temporary solution.

[0023] Based on the patent text feature vector, the cosine similarity matrix between each patent in the patent group is calculated to generate the patent similarity value; the patent group is clustered using the DBSCAN density clustering algorithm, wherein the clustering parameters include the neighborhood radius adaptively determined according to the data distribution characteristics and the preset minimum sample number threshold; if there is a high-density area in the clustering result, the high-density area is identified to form a technology cluster, and the core patents are extracted from the technology cluster as the core technology features; the core technology features are disassembled to construct the principle set and the key set, wherein the key set contains the technical parameters; the Bloom filter algorithm is used to establish a fixed-size bit array, and the size of the bit array is the key set element A preset multiple of the number is used, and multiple independent hash functions are used to map the technical parameters in the key set to the bit array to generate a potential conflict set; a conflict relationship graph is constructed, wherein the technical parameters in the potential conflict set are used as nodes of the graph, and the conflict relationships between the parameters are used as edges of the graph; the strongly connected components in the conflict relationship graph are identified by depth-first search to obtain a specific technical conflict set; an ant colony algorithm is applied to perform local exploration on the technical conflict set, wherein the initial concentration and volatility of pheromones are set as preset values, and heuristic information is designed according to the value range of the technical parameters; the technical parameter values ​​are updated through multiple iterations, and if the number of iterations reaches the preset value or the parameters converge, a local coordination temporary solution is obtained.

[0024] Exemplarily, according to the technology similarity calculation mechanism in the preset knowledge base, the TF-IDF algorithm is used to extract patent text features, calculate the cosine similarity matrix between each patent in the patent group, and generate the similarity value between patent text vectors. The DBSCAN density clustering algorithm is used to cluster the patent group, and the neighborhood radius is adaptively determined by the data distribution characteristics. The minimum sample number threshold is set to 3, and the high-density area is identified to form a technology cluster, and the core patents of each cluster are extracted as the core technology features. The identified core technology features are disassembled, and the principle set and key set are constructed respectively. The Bloom filter algorithm is used to establish a fixed-size bit array, and the bit array size is selected to be 5 times the number of key set elements. Three independent hash functions are used to map the technical parameters in the key set to the bit array to generate a potential conflict set. The graph algorithm is used to construct a conflict relationship graph, and the technical parameters in the potential conflict set are used as nodes, and the conflict relationship between the parameters is used as an edge. The strongly connected components are identified through depth-first search to obtain a specific conflict technology conflict set. Based on the ant colony algorithm, the conflict set is locally explored, the initial pheromone concentration is set to 1, the volatility is set to 0.5, the heuristic information is designed according to the value range of the technical parameters, and the technical parameter values ​​are updated through multiple iterations. When the number of iterations reaches 100 or the parameters converge, a local coordination temporary solution is obtained.

[0025] In the process of patent text processing, the TF-IDF algorithm is first used to extract patent text features, and each patent document is represented as a term frequency-inverse document frequency vector. For a group containing 1000 patents, a 1000×1000 similarity matrix is ​​generated. Each element in the matrix represents the cosine similarity value between two patents, ranging from 0 to 1. Then the DBSCAN algorithm is used to cluster the patent group. By calculating the k-distance graph of the patent vector, the inflection point is selected as the neighborhood radius ε, which is set to 0.2, and the minimum number of samples is set to 3. After clustering, multiple technical clusters are formed, and each cluster selects the patent with the largest density of samples as the core patent. The keywords and topic sentences of the core patents are extracted as the core technical features, and the principle set and key set are constructed. The Bloom filter algorithm is used to create a bit array, and the size is set to 5 times the number of key set elements. If the key set has 200 elements, the bit array size is 1000 bits. Three independent hash functions, such as MurmurHash, FNV and Jenkins hash, are selected to map the technical parameters in the key set to the bit array. Based on the mapping results, a conflict relationship graph is constructed, with nodes as technical parameters and edges as conflict relationships. The Tarjan algorithm is used for depth-first search to identify strongly connected components and obtain a specific set of conflicting technologies. Finally, the ant colony algorithm is used to optimize the conflict parameters, with 10 ants, an initial pheromone concentration of 1, a volatilization rate of 0.5, and 100 iterations. In each iteration, the ants select paths based on the pheromone concentration and heuristic information and update the technical parameter values. The heuristic information is designed based on the parameter value range. If the parameter range is [0,100], the heuristic information is set to 1 / (100-current value + 1). After the iteration, the optimal solution is selected as a temporary solution for local coordination to achieve optimal adjustment of the technical parameters.

[0026] Step S103, based on the obtained temporary solution, a particle swarm optimization algorithm is used to generate an optimized adjustment solution by locally adjusting the value range of the technical parameters, find the best parameter combination to resolve the technical conflict, build a local harmonious coexistence technical solution, add it to the knowledge base, and mark the resolved technical conflict in the technical conflict set.

[0027] Initialize the particle swarm algorithm, the initialization includes: setting the number of particles, the number of iterations, the inertia weight and the learning factor, and taking the technical parameters in the temporary solution as the initial position of the particle swarm; iteratively updating the particle position and speed through the particle swarm algorithm, the iterative update includes: calculating the fitness function value, the fitness function is designed based on the conflict degree and performance index of the technical parameters; updating the individual optimal position and the global optimal position according to the fitness function value; judging whether the iteration termination condition is met, if the iteration termination condition is met, extracting the best parameter combination from the global optimal position; generating a specific technical implementation plan according to the best parameter combination, the technical implementation plan includes parameter explanation and application instructions; adding the technical implementation plan to a preset knowledge base; for the technical conflict set, marking the resolved technical conflicts and updating the conflict set state; calculating the number and severity of the remaining conflicts; judging whether the remaining number of conflicts is greater than a preset threshold or whether the maximum conflict severity exceeds a preset value, if so, performing the next round of optimization, if not, ending the optimization process.

[0028] Exemplarily, according to the obtained temporary solution, the parameters of the particle swarm algorithm are initialized, the number of particles is set to 50, the number of iterations is 100, the inertia weight is 0.7, the learning factors c1 and c2 are both 2, and the technical parameters in the temporary solution are used as the initial position of the particle swarm. The particle position and speed are iteratively updated by the particle swarm algorithm, and the fitness function value is calculated in each iteration. The fitness function is designed based on the conflict degree and performance index of the technical parameters, and a weighted summation method is used for comprehensive evaluation to update the individual optimal position and the global optimal position, and dynamically adjust the value range of the technical parameters. When the number of iterations reaches 100 or the fitness function value changes by less than 0.001 for 10 consecutive iterations, the best parameter combination is extracted from the global optimal position, the parameter value is mapped to the actual technical indicator, a specific technical implementation plan is generated, including parameter explanations and application instructions, a local harmonious coexistence technical plan is constructed, and the plan is added to the preset knowledge base. Mark the resolved technical conflicts in the technical conflict set, update the conflict set status, calculate the number and severity of the remaining conflicts, and if the number of remaining conflicts is greater than the preset threshold of 10 or the maximum conflict severity exceeds 0.8, the next round of optimization is performed, otherwise the optimization process is terminated. In the process of technical conflict resolution, the particle swarm algorithm is first initialized based on the obtained temporary solution, and the number of particles is set to 50, representing 50 different combinations of technical parameters. The number of iterations is set to 100, the inertia weight is 0.7, and the learning factors c1 and c2 are both 2. These parameters affect the movement and convergence speed of particles. The five key technical parameters in the temporary solution are used as the initial particle positions, such as material strength, working temperature, and processing accuracy. In the iterative process, the composite fitness function f = 0.6 × (1-conflict degree) + 0.4 × performance index is designed, where the conflict degree is calculated by the correlation matrix between the parameters, and the performance index is obtained by weighting the normalized values ​​of each parameter. The particle position and speed are updated in each iteration, such as the material strength is adjusted from the initial 500MPa to 520MPa, and the working temperature is changed from 150℃ to 145℃. When the fitness function value changes less than 0.001 for 10 consecutive iterations, the global optimal position is extracted as the best parameter combination. The optimal parameters are mapped to actual technical indicators, such as material strength of 520MPa corresponding to a specific alloy formula, and operating temperature of 145℃ corresponding to a specific heat treatment process. The generated technical solution is added to the knowledge base and the resolution status is marked in the conflict set. If there are more than 10 unresolved conflicts remaining or the maximum conflict severity exceeds 0.8, a new round of optimization is started and the above process is repeated until the termination condition is met.

[0029] Step S104, determine whether the patent conflict set is empty. If it is not empty, use the support vector machine algorithm to analyze the remaining technical characteristics, establish a multi-dimensional weighted evaluation function to calculate and predict the optimal application scenario and combination method, and obtain an evaluation score. According to the evaluation score, the conflicting technologies are introduced to different application fields to achieve complementary coexistence at the global level, obtain the optimized coexistence technical solution, and add the optimized coexistence technical solution to the knowledge base.

[0030] A non-empty judgment is made on the patent conflict set. If it is not empty, the remaining technical feature vectors are extracted from the patent conflict set; based on the remaining technical feature vectors, a support vector machine algorithm is used for classification and clustering to obtain a technical feature distribution map; based on the technical feature distribution map, a multi-dimensional weighted evaluation function is established, and the multi-dimensional weighted evaluation function includes innovation, market potential and cost-effectiveness, and an evaluation score is calculated for each technical solution; based on the evaluation score, the technical solution is divided into different application fields using a clustering algorithm, and the optimal number of clusters is determined by the elbow rule; for the divided technical solutions, the complementarity index between the technical solutions is calculated, and the complementarity index is defined as the sum of the degree of complementarity of two technical solutions in different dimensions; by adjusting the combination of the technical solutions, the maximized overall complementarity index is obtained; the obtained co-existing technical solutions are added to the knowledge base, the knowledge graph structure is updated, new nodes are added to the knowledge graph to represent the technical solutions, edges represent the association relationship between technologies, and the weight of the edge reflects the strength of the association.

[0031] Exemplarily, the patent conflict set is judged as non-empty. If it is not empty, the remaining technical feature vectors are extracted. The support vector machine algorithm is used to classify and cluster the feature vectors. The RBF kernel function is used, and the kernel function parameter γ is set to 0.1 to obtain the technical feature distribution map. According to the technical feature distribution map, a multi-dimensional weighted evaluation function f = 0.4 innovation + 0.3 market potential + 0.3 * cost-effectiveness is designed, where innovation, market potential and cost-effectiveness are all normalized to the range of 0-1, and the evaluation score is calculated for each technical solution. Based on the evaluation score, the clustering algorithm is used to divide the technical solutions into different application fields, and the optimal clustering number K is determined by the elbow rule to achieve complementary coexistence of technology at the global level. The complementarity index between technical solutions is calculated. The complementarity index is defined as the sum of the degree of complementarity of two technical solutions in different dimensions. The overall complementarity index is maximized by adjusting the combination of technical solutions. The optimized coexisting technical solutions are added to the knowledge base, the knowledge graph structure is updated, new nodes are added to the graph to represent the technical solutions, and the edges represent the association relationship between technologies. The weight of the edge reflects the strength of association, and the dynamic optimization and expansion of the knowledge base are completed. In the process of resolving patent conflicts, we first check the remaining 10 patent technologies in the conflict set and extract the 20-dimensional feature vector of each technology, including keyword frequency, citation number and other indicators. The support vector machine algorithm with RBF kernel function is used for classification and clustering, and the kernel function parameter γ is set to 0.1 to obtain a two-dimensional distribution map of technical characteristics. Based on the distribution map, the evaluation function f = 0.4*innovation + 0.3*market potential + 0.3*cost-effectiveness is designed, where innovation is calculated by normalizing the number of patent citations, market potential is estimated by relevant market size data, and cost-effectiveness is obtained based on the difficulty score of technical implementation. The evaluation scores of the 10 technical solutions are calculated, ranging from 0.5 to 0.9. The K-means clustering algorithm is used to divide the technologies into different application fields. By calculating the silhouette coefficients of different K values, 2 to 6, the optimal cluster number K is determined to be 3, corresponding to three major application fields. The complementarity index between technical solutions is calculated, such as the sum of the differences between technologies A and B in the three dimensions of innovation, market potential, and cost-effectiveness, and a 10×10 complementarity matrix is ​​obtained. The technology combination with the greatest complementarity is selected through a greedy algorithm to obtain the optimized coexistence solution. The new solution is added to the knowledge base, and 3 nodes representing new application areas, 10 nodes representing technical solutions, and weighted edges between them are added to the knowledge graph. The weight range is 0.1 to 1, reflecting the strength of association.

[0032] Assign a weight to each evaluation dimension, collect relevant data for each conflicting technology, and quantify and score it according to each evaluation dimension; design a multi-dimensional weighted sum evaluation function to calculate the comprehensive evaluation score of each technology, and predict the application scenario of each technology and its combination with other technologies based on the comprehensive evaluation score; combine technologies with high scores and complementary application scenarios to form a complementary coexistence technology solution; optimize the combination solution to obtain the target combination solution as the final global complementary coexistence technology solution, and add it to the knowledge base.

[0033] An evaluation matrix is ​​constructed based on the three dimensions of technology maturity, application difficulty, and market potential, and the weight vector of each dimension is calculated using the hierarchical analysis method. After obtaining the weight vector, relevant data is extracted from patent documents and market reports, and each conflicting technology is quantitatively scored using the fuzzy comprehensive evaluation method. After obtaining the quantitative score, a multi-dimensional weighted sum evaluation function f(x)=w1*x1+w2*x2+w3*x3 is designed, where w1, w2, w3 are the weight vectors, and x1, x2, x3 are the quantitative scores, and the comprehensive evaluation score of each technology is calculated. According to the comprehensive evaluation score and the technical feature vector, the K-nearest neighbor algorithm is used to predict the best application scenario of each technology, and the K value is determined by cross-validation. After obtaining the best application scenario, the technology combination method is optimized by the genetic algorithm to generate a complementary and coexisting technology combination scheme. For the technology combination scheme, the complementarity index of the technology combination is calculated, and the complementarity index is defined as the sum of the degree of complementarity of the technologies in the combination in different dimensions. If the complementarity index meets a preset threshold, the technology combination scheme is subjected to multi-objective optimization, and a non-dominated solution set is screened out using the Pareto optimal principle.

[0034] Exemplarily, an evaluation matrix is ​​constructed based on the three dimensions of technology maturity, application difficulty, and market potential. The weights of each dimension are calculated by the hierarchical analysis method, a judgment matrix is ​​constructed and a consistency test is performed to obtain a weight vector. Relevant data are extracted from patent documents and market reports, and each conflicting technology is quantitatively scored using a fuzzy comprehensive evaluation method. A membership function is designed and a fuzzy relationship matrix is ​​constructed. A multi-dimensional weighted sum evaluation function f(x)=w1*x1+w2*x2+w3*x3 is designed, where w1, w2, and w3 are the weights of each dimension, ranging from 0 to 1 and summing to 1, and x1, x2, and x3 are the scores of the corresponding dimensions. The comprehensive evaluation score of each technology is calculated. The K-nearest neighbor algorithm is used to predict the best application scenario of each technology based on the comprehensive evaluation score and the technical feature vector, and the K value is determined by cross-validation. The technology combination method is optimized by a genetic algorithm, and the population size is set to 100, the crossover rate is 0.8, the mutation rate is 0.1, and 500 generations are iterated to generate a complementary and coexisting technology combination solution. Calculate the complementarity index of the technology combination, which is defined as the sum of the degree of complementarity of the technologies in the combination in different dimensions. Perform multi-objective optimization on the generated technology combination scheme, design objective functions including maximizing the overall score and maximizing the complementarity index, adjust the technology direction and weight, use the Pareto optimality principle to screen out the non-dominated solution set, and select the optimal solution from the non-dominated solution set through the TOPS IS method to obtain the final global complementary coexistence technology solution, add it to the knowledge base and update the association relationship. In the process of resolving technology conflicts, first construct a 3×3 evaluation matrix, which includes three dimensions: technology maturity, application difficulty and market potential. Calculate the weights through the hierarchical analysis method to obtain the weight vector (0.4, 0.3, 0.3). Extract data from 1,000 patent documents and 50 market reports, and use the fuzzy comprehensive evaluation method to score the 10 conflicting technologies. Design the membership function, map the original data to the [0,1] interval, and construct a 10×3 fuzzy relationship matrix. Apply the evaluation function

[0035] f(x)=0.4*x1+0.3*x2+0.3*x3 calculates the comprehensive score, with a score range of 0.6 to 0.9. The K-nearest neighbor algorithm is used to predict the application scenario, and the optimal K value is determined to be 5 through 5-fold cross validation. The genetic algorithm optimizes the technology combination, with a population size of 100, a crossover rate of 0.8, a mutation rate of 0.1, and 500 iterations to generate 20 candidate combination solutions. The complementarity index is calculated, with a value range of 0 to 1. Two objective functions are set for multi-objective optimization, f1 is the maximization of the overall score, and f2 is the maximization of the complementarity index. The Pareto optimal principle is used to screen out 10 non-dominated solutions, and the TOPS IS method is used to calculate the proximity of each solution to the ideal solution. The solution with the highest proximity is selected as the final global complementary coexistence technology solution. The solution contains 3 complementary technologies, with a total score of 2.5 and a complementarity index of 0.8. This solution is added to the knowledge base, and the nodes and edges in the association graph are updated to complete the entire technology conflict resolution process.

[0036] Step S105, using the random forest algorithm, constructing an evaluation index system from several key perspectives, including technological innovation and market acceptance, quantitatively scoring the coexistence technology solutions in the knowledge base, simulating the application scenarios of the coexistence technology solutions, building a virtual environment model, simulating the performance of the coexistence technology solutions under different conditions, calculating the evaluation score, measuring the deviation between the evaluation score and the actual score, and adjusting the evaluation model parameters.

[0037] Obtain characteristic variables from the degree of technological innovation and market acceptance, the characteristic variables including the number of patent citations, technological complexity, market size and user feedback; construct an evaluation index system for the random forest algorithm based on the characteristic variables. Obtain characteristic parameters of the coexistence technology solution, the characteristic parameters including market capacity, competition intensity and technological maturity; construct a virtual environment model based on the characteristic parameters, and set up simulation scenarios in the virtual environment model. Run simulation experiments for the simulation scenarios to obtain performance data; calculate the evaluation scores of the coexistence technology solution in each simulation scenario based on the performance data; determine the actual score yi and the predicted score Get the actual score yi and predicted score After that, according to the mean absolute error loss function Calculate the deviation between the evaluation score and the actual score, where n is the number of samples. Use the gradient descent method to iteratively optimize the parameters of the random forest model until the iterative optimization is stopped, and adjust the feature weights and decision tree structure.

[0038] Exemplarily, an evaluation index system is constructed based on the random forest algorithm, and characteristic variables are selected from two key perspectives: technological innovation and market acceptance, including the number of patent citations, technological complexity, market size, and user feedback, to extract features and quantify scores for coexistence technology solutions in the knowledge base. When constructing a decision tree, the Gini coefficient is used as the node splitting criterion, and the maximum depth of the tree is set to 10, and the minimum number of sample splits is set to 5. A virtual environment model is constructed using digital twin technology, and simulation scenarios are set according to the characteristic parameters of the coexistence technology solution. Key parameters include market capacity, competition intensity, and technological maturity. Simulation experiments are run under different conditions to collect performance data. Based on the simulation experiment data, the evaluation scores of the coexistence technology solution in each simulation scenario are calculated, and the original data is mapped to the range of 0-1 using normalization processing. The evaluation score calculation formula is evaluation score Score = w1*innovation+w2*market acceptance, where w1 and w2 are weight coefficients. Use the mean absolute error loss function Measures the deviation between the evaluation score and the actual score, where yi is the actual score, is the predicted score, and n is the number of samples. According to the loss function calculation results, the gradient descent method is used to iteratively optimize the parameters of the random forest model. The learning rate is set to 0.01. When the loss function changes less than 0.001 for 5 consecutive iterations or reaches the maximum number of iterations of 1000, the model is stopped. The feature weights and decision tree structure are adjusted to improve the accuracy and generalization ability of the evaluation model. In the process of evaluating the coexistence technology solutions, a random forest containing 100 decision trees is first constructed, and 10 feature variables are selected. Among them, the ones related to technological innovation include the number of patent citations, technological complexity, and innovation index, and those related to market acceptance include market size, user ratings, and product life cycle. Feature extraction is performed on the 1000 coexistence technology solutions in the knowledge base, and the original data is mapped to the 0-1 interval using normalization processing. When constructing the decision tree, the Gini coefficient is used as the node splitting criterion, and the maximum depth of the tree is set to 10 and the minimum number of sample splits is set to 5. A virtual environment model was constructed using digital twin technology, and 100 different simulation scenarios were set. Key parameters included market capacity, ranging from 100 million to 10 billion, competition intensity 1-10 levels, and technology maturity 0-100%. The simulation experiment was run 1,000 times in each scenario to collect performance data. When calculating the evaluation score, the formula Score = 0.6*innovation + 0.4*market acceptance was used to obtain the evaluation scores of 1,000 samples. The deviation was calculated using the mean absolute error loss function, and the initial MAE value was 0.15. The model parameters were optimized by the gradient descent method, and the learning rate was set to 0.01. After 500 iterations, the MAE dropped to 0.05, and the model accuracy was significantly improved. Finally, the optimized random forest model was obtained for reliable evaluation of new coexistence technology solutions.

[0039] Step S106, using a multi-objective optimization genetic algorithm, by constructing a fitness function that includes technical reliability, cost control, and market competitiveness, the optimal set of coexisting technical solutions is optimized, and the optimal technical combination solution that takes into account local coordination and global complementarity is searched and identified. According to specific application requirements, the most suitable solution is selected from the optimal patent technology combination solution, and the technical knowledge base is updated.

[0040] A multi-objective fitness function is constructed, including three sub-objectives: technical reliability, cost control, and market competitiveness. The maximum and minimum value normalization process is adopted, and the weight coefficients are set to the preset values ​​of the technical reliability, the cost control, and the market competitiveness respectively; according to the multi-objective fitness function, the comprehensive score of the technical solution is calculated; the population is initialized, the technical solution is encoded as a chromosome, the crossover probability and the mutation probability are set to the preset values, and the parent individual is selected through the roulette wheel selection algorithm; according to the crossover probability and the mutation probability, the parent individual is crossover and mutation operations are performed to generate a child population; Calculate the fitness value of the offspring population, and use the non-dominated sorting method to determine the dominance relationship between individuals; calculate the crowding degree of each individual in the offspring population according to the non-dominated sorting result to maintain the population diversity; select elite individuals from the offspring population to be retained to the next generation, and repeat the iteration until the termination condition is reached; if the termination condition is reached, extract the Pareto optimal solution set from the final population; select the most suitable technology combination scheme from the Pareto optimal solution set according to the preset screening condition; add the most suitable technology combination scheme to the technology knowledge base, and update the association relationship in the technology knowledge base.

[0041] Exemplarily, a multi-objective fitness function is constructed, including three sub-objectives: technical reliability, cost control, and market competitiveness. The maximum and minimum values ​​are normalized, and the weight coefficients are set to 0.4, 0.3, and 0.3, respectively, to obtain a comprehensive score. Initialize the population, encode the coexisting technology solutions as chromosomes, set the crossover probability to 0.8, and the mutation probability to 0.1. Select the parent individuals through the roulette selection algorithm, and perform crossover and mutation operations to generate the offspring population. Calculate the fitness value of the offspring population, use the non-dominated sorting method to determine the dominance relationship between individuals, calculate the crowding degree to maintain the diversity of the population, select elite individuals to be retained in the next generation, and repeat the iteration until the termination condition is reached. Evaluate the local coordination and global complementarity of the technology combination, calculate the synergy index and complementarity index between the technologies in the combination, and ensure that the preset threshold is met. Extract the Pareto optimal solution set from the final population, set the screening conditions according to the specific application requirements, including technology maturity not less than 0.7, expected market share not less than 10%, and return on investment not less than 15%, etc., select the most suitable technology combination solution, add the selected solution to the technology knowledge base, and update the association relationship.

[0042] In the multi-objective optimization process, the fitness function is first constructed, which includes three sub-objectives: technical reliability, cost control, and market competitiveness. For the 100 candidate technical solutions, the maximum and minimum values ​​are normalized to map the sub-objective values ​​to the range of 0-1. The weight coefficients are set to 0.4, 0.3, and 0.3, respectively, and the comprehensive score is calculated. A population of 1000 individuals is initialized, and the length of each chromosome is 100 bits, indicating the selection of technical solutions. The crossover probability is set to 0.8 and the mutation probability is set to 0.1, and the parent individual is selected by the roulette algorithm. Single-point crossover and bit flip mutation are performed to generate the offspring population. The offspring are non-dominated sorted, the crowding degree is calculated, and the top 500 elite individuals are selected. After 500 generations of iteration, the final population is obtained. The local coordination and global complementarity of the technology combination are evaluated, and the synergy index and complementarity index between the technologies in the combination are calculated. The synergy index is required to be greater than 0.6 and the complementarity index is required to be greater than 0.5. 50 Pareto optimal solutions were extracted from the final population, and the screening conditions were set: technology maturity not less than 0.7, expected market share not less than 10%, and return on investment not less than 15%. Finally, 5 optimal technology combination solutions were selected and added to the technology knowledge base containing 10,000 records, and the nodes and edges of the association graph were updated to complete the entire optimization process.

[0043] Step S107, continuously track the upgrading of technology, dynamically manage and optimize the knowledge base, and continuously enrich the optimization strategies of local coordination and global diversion by regularly updating patent data, re-evaluating technical solutions, and adjusting optimization algorithm parameters.

[0044] Obtain the latest patent information and extract key technical features, which are used to characterize the patent content; calculate the cosine similarity with the existing technology in the knowledge base based on the key technical features to obtain the newly added and changed technical content; recalculate the evaluation indicators of the technical solution for the newly added and changed technical content, and the evaluation indicators include innovation, market potential and technical maturity; use the hierarchical analysis method to build a multi-dimensional scoring model to quantitatively score the technical solution to obtain an updated evaluation result; if the updated evaluation result exceeds the preset threshold, adjust the parameter settings of the optimization algorithm, and the parameter settings include weight coefficients, number of iterations and convergence thresholds; optimize the values ​​of the parameter settings through genetic algorithms to obtain optimized parameters; update the knowledge graph structure, and the update includes adding new nodes, deleting obsolete nodes and adjusting edge weights; recalculate the association strength between nodes in the knowledge graph to obtain an updated knowledge graph; perform community discovery on the updated knowledge graph to achieve local coordination.

[0045] For example, a technology update monitoring mechanism is established, the latest patent information is collected from the patent database regularly, key technical features are extracted through the TF-IDF algorithm, and the cosine similarity is used to calculate the similarity with the existing technology in the knowledge base to identify the newly added and changed technical content. According to the newly added and changed technical content, the evaluation indicators of the technical solution are recalculated, including innovation, market potential and technical maturity. The hierarchical analysis method is used to build a multi-dimensional scoring model to quantitatively score the technical solution. Based on the updated evaluation results, the parameter settings of the optimization algorithm are adjusted, including the weight coefficient, number of iterations and convergence threshold. The parameter values ​​are optimized through the genetic algorithm, and the population size is set to 100, the crossover rate is 0.8, and the mutation rate is 0.1 to improve the adaptability of the algorithm. The Neo4j graph database is used to update the knowledge graph structure, add new nodes, delete outdated nodes, adjust the weights of edges, and recalculate the association strength between nodes. The Louva in algorithm is used for community discovery to implement the local coordination strategy, and the consistent hashing algorithm is used to implement the global diversion strategy to optimize the structure and performance of the knowledge graph. In the process of tracking technology updates, we first collect 5,000 pieces of the latest patent information from the patent database every week, and use the TF-IDF algorithm to extract the top 20 keywords of each patent as technical features. Through cosine similarity calculation, the new patents are compared with 100,000 existing technologies in the knowledge base, and the similarity threshold is set to 0.8, identifying 300 new and 50 changed technical contents. For these 350 technical contents, the hierarchical analysis method is used to construct a scoring model, with weights of 0.4 for innovation, 0.3 for market potential, and 0.3 for technical maturity, and each technology is quantitatively scored from 1 to 10. Based on the new scoring results, the genetic algorithm is used to optimize the parameters, setting the population size to 100, the crossover rate to 0.8, the mutation rate to 0.1, and iterating 500 times to obtain the optimized weight coefficient and convergence threshold. The optimization results are imported into the Neo4j graph database, and the knowledge graph containing 500,000 nodes and 2 million edges is updated, 300 new nodes are added, 100 obsolete nodes are deleted, and the weights of 5,000 edges are adjusted. The Louvain algorithm was used for community discovery, 100 technical communities were identified, and local coordination was achieved. The consistent hashing algorithm was used to map 100,000 technical nodes to 1,000 virtual nodes, achieving global load balancing and optimizing query performance by 30%.

[0046] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) to form a technical solution.

Claims

1. A technical avoidance design method based on a large language model, characterized in that: The method comprises: Use natural language processing methods to perform semantic analysis on existing technical texts, combine the decentralized characteristics of blockchain, use hierarchical clustering algorithms to divide technical data into multiple dimensions and hierarchies, transform technical data into ordered knowledge modules with different functions, and construct a knowledge base from several knowledge modules; Preset the knowledge base technology similarity calculation method, use the DBSCAN algorithm to perform technical analysis on the patent group, identify the core characteristics of the technology, split the technology into principle set and key set, use the Bloom filter algorithm to build a potential conflict set, map each technology key parameter to the conflict set, obtain the conflict set, use the ant colony algorithm to locally explore, improve the technical parameter value through multiple iterations, and obtain a local coordination temporary solution; Based on the obtained temporary solution, the particle swarm optimization algorithm is used to generate an optimized adjustment plan by locally adjusting the value range of technical parameters, find the best parameter combination to solve the technical conflict, build a local harmonious coexistence technical solution, add it to the knowledge base, and mark the resolved technical conflicts in the conflict set; Determine whether the conflict set is empty. If not, use the support vector machine algorithm to analyze the remaining technical characteristics, establish a multi-dimensional weighted evaluation function to calculate and predict the optimal application scenario and combination method, and obtain an evaluation score. According to the evaluation score, the conflicting technologies are introduced to different application fields to achieve complementary coexistence at the global level, obtain the optimized coexistence technology solution, and add the optimized coexistence technology solution to the knowledge base; Using the random forest algorithm, we construct an evaluation index system from several key perspectives, including technological innovation and market acceptance. We quantitatively score the coexistence technology solutions in the knowledge base, simulate the application scenarios of the coexistence technology solutions, build a virtual environment model, simulate the performance of the coexistence technology solutions under different conditions, calculate the evaluation score, measure the deviation between the evaluation score and the actual score, and adjust the evaluation model parameters. Adopting multi-objective optimization genetic algorithm, by constructing a fitness function including technical reliability, cost control, and market competitiveness, the optimal set of coexisting technical solutions is optimized, and the optimal technical combination solution that takes into account local coordination and global complementarity is searched and identified. According to specific application requirements, the most suitable solution is selected from the optimal patent technology combination solution, and the technical knowledge base is updated; We continue to track technological updates, dynamically manage and optimize the knowledge base, and constantly enrich the optimization strategies of local coordination and global diversion by regularly updating patent data, re-evaluating technical solutions, and adjusting optimization algorithm parameters.

2. The method according to claim 1, characterized in that The natural language processing method is used to perform semantic analysis on existing technical texts, and combined with the decentralized characteristics of blockchain, the hierarchical clustering algorithm is used to divide the technical data into multiple dimensions and levels, and the technical data is converted into orderly functional knowledge modules, and several knowledge modules are constructed to form a knowledge base, including: Perform part-of-speech tagging and dependency syntax analysis on technical texts, extract keywords and core concepts, and generate text semantic vector representations; According to the text semantic vector, a hierarchical clustering algorithm is used to perform multi-level clustering on the technical data to obtain a hierarchical technical knowledge tree structure; According to the hierarchical technical knowledge tree structure, the technical data is divided into independent knowledge modules, a unique identifier is assigned to each knowledge module, the relationship and execution order between the knowledge modules are defined, and a knowledge module dependency graph is constructed; Serialize the knowledge module into serialized data; Distributed hash tables are used to store serialized knowledge modules, and the knowledge base status is updated synchronously in the blockchain network.

3. The method according to claim 1, characterized in that The preset knowledge base technology similarity calculation method uses the DBSCAN algorithm to perform technical analysis on the patent group, identify the core characteristics of the technology, split the technology into principle set and key set, use the Bloom filter algorithm to construct a potential conflict set, map each technical key parameter to the conflict set, obtain the conflict set, use the ant colony algorithm to locally explore, and improve the technical parameter value through multiple iterations to obtain a local coordination temporary solution, including: Calculate the cosine similarity matrix between each patent in the patent group based on the patent text feature vector to generate the patent similarity value; Clustering the patent groups using the DBSCAN density clustering algorithm, wherein clustering parameters include a neighborhood radius adaptively determined according to data distribution characteristics and a preset minimum sample number threshold; If there is a high-density area in the clustering result, the high-density area is identified to form a technology cluster, and core patents are extracted from the technology cluster as the core technology features; Deconstruct the core features of the technology to construct a principle set and a key set, wherein the key set includes technical parameters; A Bloom filter algorithm is used to establish a fixed-size bit array, where the size of the bit array is a preset multiple of the number of key set elements, and a plurality of independent hash functions are used to map the technical parameters in the key set to the bit array to generate a potential conflict set; Constructing a conflict relationship graph, wherein the technical parameters in the potential conflict set serve as nodes of the graph, and the conflict relationships between the parameters serve as edges of the graph; Identify the strongly connected components in the conflict relationship graph through depth-first search to obtain a specific conflict set; Applying an ant colony algorithm to perform local exploration on the conflict set, wherein the initial concentration and volatility of the pheromone are set as preset values, and heuristic information is designed according to the value range of the technical parameters; The technical parameter values ​​are updated through multiple iterations. If the number of iterations reaches the preset value or the parameters converge, a local coordination temporary solution is obtained.

4. The method according to claim 1, characterized in that: The temporary solution based on the acquisition adopts the particle swarm optimization algorithm to generate an optimized adjustment solution by locally adjusting the value range of the technical parameters, find the best parameter combination to solve the technical conflict, build a local harmonious coexistence technical solution, add it to the knowledge base, and mark the resolved technical conflicts in the conflict set, including: Initializing the particle swarm algorithm, the initialization includes: setting the number of particles, the number of iterations, the inertia weight and the learning factor, and using the technical parameters in the temporary solution as the initial position of the particle swarm; Iteratively updating particle positions and velocities through the particle swarm algorithm, wherein the iterative update includes: calculating a fitness function value, wherein the fitness function is designed based on the conflict degree of technical parameters and performance indicators; Update the individual optimal position and the global optimal position according to the fitness function value; Determining whether an iteration termination condition is satisfied, and if so, extracting an optimal parameter combination from the global optimal position; Generate a specific technical implementation plan according to the optimal parameter combination, wherein the technical implementation plan includes parameter explanation and application description; Adding the technical implementation scheme to a preset knowledge base; For the conflict set, marking the resolved technical conflicts and updating the conflict set status; Calculate the number and severity of remaining conflicts; It is determined whether the remaining number of conflicts is greater than a preset threshold or whether the maximum conflict severity exceeds a preset value. If so, the next round of optimization is performed; if not, the optimization process is terminated.

5. The method according to claim 1, characterized in that The judgment of whether the conflict set is empty or not, if not empty, uses the support vector machine algorithm to analyze the remaining technical characteristics, establishes a multi-dimensional weighted evaluation function to calculate and predict the optimal application scenario and combination method, obtains an evaluation score, and according to the evaluation score, introduces the conflicting technologies to different application fields, realizes complementary coexistence at the global level, obtains an optimized coexistence technical solution, and adds the optimized coexistence technical solution to the knowledge base, including: Performing a non-empty judgment on the conflict set, and if it is not empty, extracting the remaining technical feature vector from the conflict set; According to the remaining technical feature vectors, a support vector machine algorithm is used to perform classification and clustering to obtain a technical feature distribution map; Based on the technical characteristics distribution map, a multi-dimensional weighted evaluation function is established, wherein the multi-dimensional weighted evaluation function includes innovation, market potential and cost-effectiveness, and an evaluation score is calculated for each technical solution; According to the evaluation scores, the technical solutions are divided into different application fields using a clustering algorithm, and the optimal number of clusters is determined by the elbow rule; For the divided technical solutions, the complementarity index between the technical solutions is calculated, where the complementarity index is defined as the sum of the complementarity degrees of the two technical solutions in different dimensions; By adjusting the combination of the technical solutions, a maximized overall complementarity index is obtained; Add the obtained coexisting technical solutions to the knowledge base, update the knowledge graph structure, add new nodes to the knowledge graph to represent the technical solutions, and the edges represent the association relationship between technologies. The weight of the edge reflects the strength of the association. It also includes: assigning weights to each evaluation dimension, collecting relevant data for each conflict technology, and quantifying scores according to each evaluation dimension; Design a multi-dimensional weighted sum evaluation function to calculate the comprehensive evaluation score of each technology. Based on the comprehensive evaluation score, predict the application scenario of each technology and its combination with other technologies; combine technologies with high scores and complementary application scenarios to form a complementary coexistence technology solution; optimize the combination solution to obtain the target combination solution as the final global complementary coexistence technology solution and add it to the knowledge base.

6. The method according to claim 5, characterized in that The method assigns a weight to each evaluation dimension, collects relevant data for each conflicting technology, and performs quantitative scoring according to each evaluation dimension; designs a multi-dimensional weighted sum evaluation function, calculates the comprehensive evaluation score of each technology, and predicts the application scenario of each technology and the combination method with other technologies based on the comprehensive evaluation score; combines technologies with high scores and complementary application scenarios to form a complementary coexistence technology solution; optimizes the combination solution to obtain a target combination solution as the final global complementary coexistence technology solution, and adds it to the knowledge base, including: An evaluation matrix is ​​constructed based on the three dimensions of technology maturity, application difficulty, and market potential, and the weight vector of each dimension is calculated using the hierarchical analysis method; After obtaining the weight vector, relevant data is extracted from patent documents and market reports, and a fuzzy comprehensive evaluation method is used to quantitatively score each conflicting technology; After obtaining the quantitative scores, a multi-dimensional weighted sum evaluation function f(x)=w1*x1+w2*x2+w3*x3 is designed, where w1, w2, w3 are the weight vectors, x1, x2, x3 are the quantitative scores, and the comprehensive evaluation score of each technology is calculated; According to the comprehensive evaluation score and the technical feature vector, the best application scenario of each technology is predicted using the K-nearest neighbor algorithm, wherein the K value is determined by cross-validation; After obtaining the optimal application scenario, optimizing the technology combination method through a genetic algorithm to generate a complementary and coexisting technology combination solution; For the technology combination scheme, calculating the complementarity index of the technology combination, wherein the complementarity index is defined as the sum of the complementarity degrees of the technologies in the combination in different dimensions; If the complementarity index meets a preset threshold, the technology combination scheme is subjected to multi-objective optimization, and a non-dominated solution set is screened out using the Pareto optimal principle.

7. The method according to claim 1, characterized in that The random forest algorithm is used to construct an evaluation index system from several key perspectives, including technological innovation and market acceptance, to quantitatively score the coexistence technology solutions in the knowledge base, to simulate the application scenarios of the coexistence technology solutions, to build a virtual environment model, to simulate the performance of the coexistence technology solutions under different conditions, to calculate the evaluation score, to measure the deviation between the evaluation score and the actual score, and to adjust the evaluation model parameters, including: Acquire characteristic variables from technological innovation and market acceptance, wherein the characteristic variables include patent citation number, technological complexity, market size and user feedback; Constructing an evaluation index system of a random forest algorithm according to the characteristic variables; Acquire characteristic parameters of the coexistence technology solution, wherein the characteristic parameters include market capacity, competition intensity, and technology maturity; Constructing a virtual environment model according to the characteristic parameters, and setting a simulation scene in the virtual environment model; Running a simulation experiment for the simulation scenario to obtain performance data; Calculate the evaluation score of the coexistence technology solution in each simulation scenario according to the performance data; Determine the actual score yi and the predicted score ; Get the actual score yi and predicted score After that, according to the mean absolute error loss function Calculate the deviation between the evaluation score and the actual score, where n is the number of samples; The gradient descent method is used to iteratively optimize the parameters of the random forest model until the iterative optimization is stopped, and the feature weights and decision tree structure are adjusted.

8. The method according to claim 1, characterized in that The multi-objective optimization genetic algorithm is used to optimize the optimal coexistence technology solution set by constructing a fitness function that includes technology reliability, cost control, and market competitiveness, search and identify the optimal technology combination solution that takes into account local coordination and global complementarity, select the most suitable solution from the optimal patent technology combination solution according to specific application requirements, and update the technology knowledge base, including: Construct a multi-objective fitness function, including three sub-objectives: technical reliability, cost control, and market competitiveness, adopt maximum and minimum value normalization processing, and set weight coefficients to preset values ​​of the technical reliability, cost control, and market competitiveness respectively; Calculate the comprehensive score of the technical solution according to the multi-objective fitness function; Initialize the population, encode the technical solution into chromosomes, set the crossover probability and mutation probability to preset values, and select the parent individuals through the roulette wheel selection algorithm; According to the crossover probability and the mutation probability, performing crossover and mutation operations on the parent individuals to generate a descendant population; Calculating the fitness value of the offspring population, and using a non-dominated sorting method to determine the dominance relationship between individuals; According to the non-dominated sorting result, the crowding degree of each individual in the offspring population is calculated to maintain population diversity; Select elite individuals from the offspring population and keep them in the next generation, and repeat the iteration until the termination condition is reached; If the termination condition is met, the Pareto optimal solution set is extracted from the final population; According to the preset screening conditions, the most suitable technical combination solution is selected from the Pareto optimal solution set; The most suitable technology combination solution is added to the technology knowledge base, and the association relationship in the technology knowledge base is updated.

9. The method according to claim 1, characterized in that: The above mentioned continuous tracking of technology upgrades, dynamic management and optimization of the knowledge base, and the enrichment of local coordination and global diversion optimization strategies by regularly updating patent data, re-evaluating technical solutions, and adjusting optimization algorithm parameters include: Obtain the latest patent information and extract key technical features, which are used to characterize the patent content; Calculate the cosine similarity with the existing technology in the knowledge base based on the key technical features to obtain the newly added and changed technical content; Recalculate the evaluation indicators of the technical solution for the newly added and changed technical contents, including innovation, market potential and technical maturity; A multi-dimensional scoring model is constructed using the analytic hierarchy process to quantitatively score the technical solution and obtain updated evaluation results; If the updated evaluation result exceeds a preset threshold, adjusting the parameter settings of the optimization algorithm, the parameter settings including weight coefficient, number of iterations and convergence threshold; Optimizing the values ​​of the parameter settings by a genetic algorithm to obtain optimized parameters; Updating the knowledge graph structure, wherein the update includes adding new nodes, deleting outdated nodes, and adjusting edge weights; Recalculating the association strength between nodes in the knowledge graph to obtain an updated knowledge graph; Perform community discovery on the updated knowledge graph to achieve local coordination.

Images