Deep reinforcement learning-based incomplete industry chain supplementary industry recommendation method and system

Abstract

The invention discloses an incomplete industry chain supplementary industry recommendation method and system based on deep reinforcement learning, a deep reinforcement learning model is adopted to analyze industry chain missing links and recommend an optimal supplementary industry, and the method comprises the following steps: dividing existing industry chain network data according to administrative division; the network is disassembled into a plurality of incomplete regional industry chain sub-networks; and integrating the enterprise information in each sub-network to form complete industrial data, and calculating the importance of the industry by using the industrial data. And carrying out comprehensive evaluation on the industry importance and the industry chain toughness improvement effect on the industry chain missing industry by utilizing a deep reinforcement learning model based on generation of simulation network training, selecting the nodes with the highest score to increase, and repeating the evaluation and increase process until a preset supplementary industry quantity is reached. According to the method, industrial importance evaluation and industrial chain topological structure toughness analysis are creatively combined, a dual chain supplement target considering regional industrial requirements and industrial chain stability improvement is constructed, and identification of the chain supplement industry is realized by applying a deep reinforcement learning algorithm.

Description

Technical Field

[0001] The present invention belongs to the field of deep learning and data mining, and in particular relates to a method and system for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning. Background Art

[0002] Against the backdrop of intensifying global trade frictions, the ongoing impact of the pandemic, and rising geopolitical risks, building a resilient, flexible, and competitive industrial chain network has become a strategic imperative for safeguarding national economic security and enhancing international competitiveness. To effectively enhance the vitality and resilience of regional industrial chains, scientifically and rationally recommending complementary industries is crucial. However, existing methods for recommending complementary industries have significant shortcomings: they fail to fully consider the impact of industrial chain network topology on resilience, while over-reliance on manual analysis leads to inefficiencies.

[0003] In response to the above problems, this patent proposes an innovative solution: First, by integrating industry importance assessment and industry chain topology structure resilience analysis, a dual optimization goal is established that takes into account regional industry needs and industry chain resilience improvement; second, a deep reinforcement learning algorithm is introduced to identify and recommend chain-filling industries, significantly improving decision-making efficiency and accuracy.

[0004] Specifically, the current industry chain supplement industry recommendation faces two core challenges:

[0005] (1) Existing methods generally ignore the key factor of the industrial chain network topology structure, making it difficult to effectively evaluate and improve the overall resilience of the industrial chain;

[0006] (2) The recommendation process relies heavily on expert experience and judgment, which is not only inefficient but also difficult to cope with the complex and changing industrial environment.

[0007] The Chinese patent document with publication number CN119693026A discloses a method and system for recommending regional industrial chain supplementary enterprises based on communication big data. Starting from the perspective of the introduced enterprises, the patent analyzes regional policies and enterprise data to evaluate the support for the introduced enterprises, and uses enterprise communication data to measure their willingness to settle in, and finally calculates the enterprise recommendation index. The analysis perspective of this patent focuses on the introduced enterprises, analyzes the current region's support for the introduced enterprises, and does not consider the optimization goal of the resilience of the industrial chain network topology structure. Starting from the perspective of the introducing party, this patent not only considers the regional industrial needs, but also takes into account the improvement of the resilience of the industrial chain, constructs a dual chain supplementation goal, and more comprehensively assists regional industrial development. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for recommending supplementary industries for incomplete industrial chains based on deep reinforcement learning, which solves the dual defects of traditional solutions that ignore network resilience and are inefficient.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: a method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning, comprising the following steps:

[0010] S1. Integrate the information of enterprises in each industry in the existing industrial chain network data to form industrial data;

[0011] S2. Use an attribute importance analysis method to analyze the proportion of each attribute in the industrial data, such as principal component analysis, autoencoder sensitivity analysis, or hierarchical analysis method;

[0012] S3. Decompose the industrial chain network data by administrative region to generate multiple regional and incomplete sub-networks. Each sub-network only contains enterprise information in the corresponding region. The enterprise information is integrated into the regional industrial data, and the industry weight is calculated according to the proportion to form the industrial chain network dataset.

[0013] S4. Based on the Barabasi-Albert network model, networks of different sizes are generated, and nodes are deleted and weights are distributed in the networks to form a regional industrial chain simulation network dataset;

[0014] S5. Construct a scoring function for chain-filling behavior to evaluate the simulated network chain-filling behavior score, selectively add industries and update the status, repeat this process and record the scores to form a training set;

[0015] S6. Use the training set to train the deep reinforcement learning model to enable the model to learn the scoring evaluation mechanism of the chain-filling behavior;

[0016] S7. Extract relevant network data of the chain to be supplemented from the industrial chain network dataset, combine the weight of each node in the extracted network with the weight of the neighboring nodes to calculate its importance, and use the deep reinforcement learning model to recommend supplementary industries.

[0017] Furthermore, S1 includes:

[0018] S1.1. Integrate the information of enterprises in each industry in the existing industrial chain network data to form industrial data, where the enterprise data is D = {d1, d2, d3, ..., d n},n∈N + , each piece of data contains m attributes, and the enterprise data in each industry are integrated to form industry data Industry = {ind1, ind2, ind3,…, ind q},q∈N + , each piece of data contains l attributes;

[0019] Furthermore, S2 includes:

[0020] S2.1. Use an attribute importance analysis method to analyze the weight β of each attribute in the industrial data, such as principal component analysis, autoencoder sensitivity analysis, or hierarchical analysis.

[0021] Furthermore, S3 includes:

[0022] S3.1. Decompose the industrial chain network data by administrative division to generate multiple regional and incomplete sub-networks. Each sub-network only contains enterprise information in the corresponding region. The enterprise information is integrated into the regional industrial data, and the industry weight is calculated according to the proportion to form the industrial chain network dataset. The industry weight calculation formula is as follows:

[0023] w i =β1ind i1 +β2ind i2 +…+β j ind ij

[0024] Where w i is the weight of industry i, ind ij is the j-th attribute of industry i.

[0025] Furthermore, S2 includes:

[0026] S4.1. Use the Barabasi-Albert network model to generate networks of different sizes and obtain the adjacency matrix A of each network.

[0027] S4.2. Delete nodes from the generated network in different proportions and randomly assign weights to the remaining nodes in the network to obtain a new adjacency matrix A′ and node weight vector W = (w1, w2, …, w n ); The importance of the industry is calculated by combining the weight of each node with the weight of the neighboring nodes, and then the importance vector Imp=(I1,I2,…I n ), and use this to construct a regional industrial chain simulation network dataset. The formula for calculating industry importance is as follows:

[0028]

[0029] Where, d ij For node v i and v j The shortest path length, d j For node v j degree.

[0030] Furthermore, S5 includes:

[0031] S5.1. Construct a chain-filling behavior score function based on industry importance and network connectivity metrics. The behavior score function can be calculated based on the current simulated network state S = (s1, s2, ... sn ) evaluates the score of the simulated network when adding different nodes; where the node status s i Use 0 and 1 to indicate whether a node exists. The score function is as follows:

[0032]

[0033] Where σ(G) is the connectivity of the network, σ(G∪{ν1,ν2,...,ν k}) Add node sequence {ν1,ν2,...,ν k}After the network connectivity.

[0034] S5.2. Use the behavioral scoring function to score the missing nodes of each network in the simulated network dataset, select the nodes with the highest scores to add, and thus complete the network state transition. Repeat the above evaluation-selection-transition process until the predetermined number of supplementary industries is reached, and fully record the state vectors and corresponding scores during the network state transition process to construct a training dataset.

[0035] Furthermore, S6 includes:

[0036] S6.1. Divide the training data set into a training set and a test set in a ratio of 8:2, and train a deep reinforcement learning model. The model learns and evaluates the benefits brought by the network's chain-filling behavior under different states, and ultimately realizes the function of recommending supplementary industries to the regional industrial chain network.

[0037] Furthermore, S5 includes:

[0038] S7.1. Extract relevant network data of the chain to be supplemented from the industrial chain network dataset, and calculate the importance of each node by combining its own weight with the weight of its neighboring nodes. The formula is as follows:

[0039]

[0040] S7.2. Use deep reinforcement learning models to recommend additional industries.

[0041] The present invention also provides a non-complete industry chain supplementary industry recommendation system based on deep reinforcement learning, comprising:

[0042] The data preprocessing module integrates the information of enterprises under each industry in the existing industrial chain network data to form industrial data, and uses an attribute importance analysis method to analyze the proportion of each attribute in the industrial data; the industrial chain network data is decomposed according to administrative regions to generate multiple regional and incomplete sub-networks; each sub-network only contains the enterprise information of the corresponding region, and the enterprise information is integrated into the industrial data of the region, and the industry weight is calculated according to the proportion.

[0043] The simulated network generation module generates networks of different sizes based on the Barabasi-Albert network model. It generates a residual network by randomly deleting nodes from the generated network, assigning random weights to the remaining nodes, and calculating the node importance to form a simulated network dataset.

[0044] The model training module constructs a scoring function for link-filling behavior based on industry importance and network connectivity metrics. This scoring function is then used to score the missing nodes in each network in the simulated network dataset. Nodes are added selectively and their status updated. This process is repeated, and the scores are recorded to form a training dataset. The training dataset is split into a training set and a test set at an 8:2 ratio. A deep reinforcement learning model is trained to learn the scoring mechanism for link-filling behavior.

[0045] The supplementary industry recommendation module extracts relevant network data of the chain to be supplemented from the industrial chain network dataset, combines the weight of each node with the weight of the neighboring nodes to calculate its importance, and uses the deep reinforcement learning model to recommend supplementary industries for the incomplete regional industrial chain network.

[0046] Beneficial effects of the present invention:

[0047] 1) The present invention proposes a method for recommending supplementary industries for incomplete industrial chains based on deep reinforcement learning, which aims to help regional industrial chains discover current needs and enhance their network resilience. Specifically, the industrial chain network data is decomposed according to administrative regions to generate multiple regional and incomplete sub-networks, and the enterprise information of each sub-network is integrated and the industry weight is calculated to construct a regional industrial chain network dataset. The generated simulated industrial chain network dataset is used to train a deep reinforcement learning model so that the model learns the benefits brought by the behavior of newly added nodes in the network under different states. On this basis, the relevant network data of the chain to be supplemented is extracted from the industrial chain network dataset, and the weight of each node itself is combined with the weight of the neighboring nodes to calculate its importance. The deep reinforcement learning model is used to conduct a comprehensive assessment of the industry importance and the effect of improving the resilience of the industrial chain on the missing industries in the industrial chain. The industry with the highest benefit is selected as a recommendation, and the state transition is performed until the number of supplementary industries is met. This method recommends supplementary industries based on the dual goals of regional industrial demand and improving the resilience of the industrial chain, effectively enhancing the resilience and competitiveness of the industrial chain network, and providing scientific decision-making support for regional economic development. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning disclosed in the present invention;

[0049] Figure 2This is a structural diagram of a supplementary industry recommendation system for an incomplete industrial chain based on deep reinforcement learning disclosed by the present invention. DETAILED DESCRIPTION

[0050] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0051] Example 1

[0052] Figure 1 As shown, this embodiment provides a method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning. The specific steps are as follows:

[0053] S1. Integrate the information of enterprises in each industry in the existing industrial chain network data to form industrial data;

[0054] S1 specifically includes the following steps:

[0055] S1.1. Integrate the information of enterprises in each industry in the existing industrial chain network data to form industrial data, where the enterprise data is D = {d1, d2, d3, ..., d 9154}, each data contains 10 attributes, some attributes of enterprise information data are shown in Table 1; the enterprise data in each industry are integrated to form industry data Industry = {ind1, ind2, ind3, ..., ind 578}, each piece of data contains three attributes, as shown in Table 2; the industrial chain network data contains industry names and connection relationships, as shown in Table 3.

[0056] Table 1 Partial attribute table of enterprise information data

[0057] Serial number property 1 registered capital 2 Economic Industry 3 Enterprise Type 4 Number of employees 5 Business Scope …… ……

[0058] Table 2 Industry data attribute table

[0059] Serial number property 1 Industry Name 2 registered capital 3 Number of employees 4 Number of enterprises

[0060] Table 3 Industrial chain network attributes

[0061] Serial number property 1 Industry Name 1 2 Connection relationship 3 Industry Name 2 4 Name of the industry chain

[0062] S2. Use an attribute importance analysis method to analyze the proportion of each attribute in the industrial data, such as principal component analysis, autoencoder sensitivity analysis, or hierarchical analysis method;

[0063] S2 specifically includes the following steps:

[0064] S2.1. Use industrial data to train the autoencoder model;

[0065] S2.2. Use the small perturbation method to quantify the impact of each attribute on the output of the autoencoder model and construct the Jacobian matrix J:

[0066]

[0067] Calculate the 2-norm of the Jacobian matrix column to get the contribution of each attribute. The calculation formula is as follows:

[0068]

[0069]

[0070]

[0071] Normalize the contribution to get the attribute proportion, and the calculation formula is as follows:

[0072]

[0073]

[0074]

[0075] S3. Decompose the industrial chain network data by administrative region to generate multiple regional and incomplete sub-networks. Each sub-network only contains enterprise information in the corresponding region. The enterprise information is integrated into the regional industrial data, and the industry weight is calculated according to the proportion to form the industrial chain network dataset.

[0076] S3 specifically includes the following steps:

[0077] S3.1. Decompose the industrial chain network data by administrative division to generate multiple regional and incomplete sub-networks. Each sub-network only contains enterprise information in the corresponding region. The enterprise information is integrated into the regional industrial data, and the industry weight is calculated according to the proportion to form the industrial chain network dataset. The industry weight calculation formula is as follows:

[0078] w i =0.45ind i1 +0.17ind i2 +0.38ind i3 ,i∈[1,578]

[0079] S4. Based on the Barabasi-Albert network model, networks of different sizes are generated, and nodes are deleted and weights are distributed in the networks to form a regional industrial chain simulation network dataset;

[0080] S4 specifically includes the following steps:

[0081] S4.1. Use the Barabasi-Albert network model to generate 200 networks with nodes between 300 and 500, and obtain the adjacency matrix A of their complete networks.

[0082] S4.2. Randomly select a proportion of nodes in the range of 0-100% with an interval of 10% for the generated network, delete the corresponding proportion of nodes, and randomly assign weights to the remaining nodes to obtain a new adjacency matrix A′ and node weight vector W=(w1,w2,…,w n ); Taking a complete network with 455 nodes as an example, 81% of the nodes in the network are deleted, that is, 369 nodes are randomly deleted, leaving 86 nodes. For the remaining 86 nodes, node weights are randomly assigned, resulting in a weight vector of W = (0, 0.16, …, 0.02, …, 0); the industry importance is calculated by combining each node's own weight with the weights of its neighboring nodes, resulting in an importance vector of Imp = (0.08, 0.09, …, 0.05, …, 0), calculated as follows:

[0083]

[0084] Where, d ij For node v i and v j The shortest path length, d j For node v j The degree of is used to construct the network dataset X2.

[0085] S5. Construct a scoring function for chain-filling behavior to evaluate the simulated network chain-filling behavior score, selectively add industries and update the status, repeat this process and record the scores to form a training set;

[0086] S5 specifically includes the following steps:

[0087] S5.1. Take a complete network with 455 nodes, 81% of which are deleted, that is, 369 nodes are randomly deleted, leaving a residual network with 86 nodes, and the number of supplementary industries is set to 3. The initial state vector is S = (1, 0, ..., 1, ..., 0), the node importance vector is Imp = (0.08, 0.09, ..., 0.05, ..., 0), and the adjacency matrix is ​​A′.

[0088] The score of adding different nodes to the network is calculated based on the score function. The score function is as follows:

[0089]

[0090] Where σ(G) is the connectivity of the network, which is the number of nodes of the largest connected component of the industrial chain network calculated by the depth-first search algorithm, and its value is 12. k}) Add node sequence {ν1,ν2,...,ν k}After the network connectivity.

[0091] S5.2. Use the behavior score function to score the missing nodes in each network in the simulated network dataset, obtaining a behavior score vector R = {0.38, 0.66, ..., 0.28}. Select the highest-scoring node 4 for addition, with a score of 0.660320, and perform a network state transition. Repeat the evaluation-selection-transition process until the predetermined number of supplementary industries, 3, is reached. The final recommended node numbers and scores are shown in Table 3:

[0092] Table 3 Added node gain table

[0093] Increase the node number Increase node score 4 0.660320 12 0.560582 15 0.483986

[0094] All networks go through the above process to find the optimal sequence of adding nodes and record them to construct the training data set.

[0095] S6. Use the training set to train the deep reinforcement learning model to enable the model to learn the scoring evaluation mechanism of the chain-filling behavior;

[0096] S6 specifically includes the following steps:

[0097] S6.1. Split the training dataset into a training set and a test set in a ratio of 8:2, and train a deep reinforcement learning model. This model learns and evaluates the benefits of chain-filling behaviors in different network states, then selects the nodes with the largest gains to add, and then performs network state transitions. This evaluation-selection-transition process is repeated until the predetermined number of supplementary industries (3) is reached, ultimately achieving the function of recommending supplementary industries for the regional industrial chain network.

[0098] S7. Extract relevant network data of the chain to be supplemented from the industrial chain network dataset, combine the weight of each node in the extracted network with the weight of the neighboring nodes to calculate its importance, and use the deep reinforcement learning model to recommend supplementary industries.

[0099] S7 specifically includes the following steps:

[0100] S7.1. Extract relevant network data for the link to be supplemented from the industrial chain network dataset. Take the low-altitude economic industrial chain in ** City, Sichuan as an example. Calculate the importance of each node by combining its own weight with the weights of its neighboring nodes (the shortest path length does not exceed 2). Obtain the importance vector Imp = (0.06, 0, ..., 0.29, ..., 0). The calculation formula is as follows:

[0101]

[0102] Where d ij For node v i and v j The shortest path length, d j For node v j degree.

[0103] S7.2. Use the deep reinforcement learning model to recommend supplementary industries to the regional industrial chain network. The results are shown in Table 5.

[0104] Table 5 Supplementary industry score table

[0105] Recommended new industries Increase industry score Airport equipment 0.907634 micro switch 0.604954 plastic 0.486339

[0106] Example 2

[0107] like Figure 2 As shown, this embodiment provides an incomplete industry chain supplementation system based on deep reinforcement learning, including: a data preprocessing module, a simulation network generation module, a model training module, and a supplementary industry recommendation module;

[0108] Among them, the data preprocessing module integrates the information of enterprises under each industry in the existing industrial chain network data to form industrial data, and uses an attribute importance analysis method to analyze the proportion of each attribute in the industrial data; the industrial chain network data is decomposed according to administrative regions to generate multiple regional and incomplete sub-networks; each sub-network only contains the enterprise information of the corresponding region, and the enterprise information is integrated into the industrial data of the region, and the industry weight is calculated according to the proportion.

[0109] The simulated network generation module generates networks of different sizes based on the Barabasi-Albert network model; generates a residual network by randomly deleting nodes from the generated network, assigns random weights to the remaining nodes, and calculates the node importance to form a simulated network data set.

[0110] The model training module constructs a scoring function for link-filling behavior based on industry importance and network connectivity metrics. This scoring function is then used to score missing nodes in each network in the simulated network dataset. The module then selectively adds nodes and updates their states, repeating this process and recording the scores to form a training dataset. The training dataset is split into a training set and a test set at an 8:2 ratio, and a deep reinforcement learning model is trained to learn the scoring mechanism for link-filling behavior.

[0111] The supplementary industry recommendation module extracts relevant network data of the chain to be supplemented from the industrial chain network dataset, combines the weight of each node with the weight of the neighboring nodes to calculate its importance, and uses a deep reinforcement learning model to recommend supplementary industries for the incomplete regional industrial chain network.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning, characterized in that: The steps include: S1. Integrate the information of enterprises in each industry in the existing industrial chain network data to form industrial data; S2. Use an attribute importance analysis method to analyze the proportion of each attribute in the industrial data, such as principal component analysis, autoencoder sensitivity analysis, or hierarchical analysis method; S3. Decompose the industrial chain network data by administrative region to generate multiple regional and incomplete sub-networks. Each sub-network only contains enterprise information in the corresponding region. The enterprise information is integrated into the regional industrial data, and the industry weight is calculated according to the proportion to form the industrial chain network dataset. S4. Based on the Barabasi-Albert network model, networks of different sizes are generated, and nodes are deleted and weights are distributed in the networks to form a regional industrial chain simulation network dataset; S5. Construct a scoring function for chain-filling behavior to evaluate the simulated network chain-filling behavior score, selectively add industries and update the status, repeat this process and record the scores to form a training set; S6. Use the training set to train the deep reinforcement learning model to enable the model to learn the scoring evaluation mechanism of the chain-filling behavior; S7. Extract relevant network data of the chain to be supplemented from the industrial chain network dataset, combine the weight of each node in the extracted network with the weight of the neighboring nodes to calculate its importance, and use the deep reinforcement learning model to recommend supplementary industries.

2. The method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning according to claim 1, characterized in that: Said S1 comprises: S1.

1. Integrate the information on enterprises within each industry from existing industry chain network data, such as the number of enterprises within each industry, the number of insured persons, and the total amount of registered capital, and standardize this information to form industry data. Industry chain network data includes the network topology relationships between industries and detailed data on enterprises within each industry.

3. The method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning according to claim 1, characterized in that: The S2 includes: S2.

1. Use an attribute importance analysis method to analyze the proportion of each attribute in the industrial data, such as principal component analysis, autoencoder sensitivity analysis, or hierarchical analysis method.

4. The method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning according to claim 1, characterized in that: The S3 includes: S3.

1. Decompose the industrial chain network data by administrative division to generate multiple regional and incomplete sub-networks. Each sub-network only contains enterprise information in the corresponding region. Integrate the enterprise information into the industrial data of the region, and calculate the industry weight according to the proportion to form the industrial chain network dataset.

5. The method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning according to claim 1, characterized in that: The S4 includes: S4.

1. Generate networks of different sizes using the Barabasi-Albert network model. S4.

2. Delete nodes from the generated network in different proportions, and randomly assign weights to the remaining nodes in the network. Combine the node's own weight with the weight of the neighboring nodes to calculate the node importance, thereby constructing a regional industrial chain simulation network dataset.

6. The method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning according to claim 1, characterized in that: The S5 includes: S5.

1. Construct a chain-filling behavior scoring function based on industry importance and network connectivity metrics. The behavior scoring function can evaluate the score of the simulated network when adding different nodes based on the current simulated network state. S5.

2. Use the behavior scoring function to score the missing nodes of each network in the simulated network dataset, selectively add nodes and update the status, repeat the scoring-adding node process, and record the scores to form a training set.

7. The method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning according to claim 1, characterized in that: The S6 includes: S6.

1. Use the training set to train the deep reinforcement learning model so that the model can learn the scoring evaluation mechanism of the chain-filling behavior.

8. The method for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning according to claim 1, characterized in that: The S7 includes: S7.

1. Extract relevant network data of the chain to be supplemented from the industrial chain network dataset, and calculate the importance of each node by combining its own weight with the weights of its neighboring nodes; S7.

2. Use deep reinforcement learning models to recommend additional industries.

9. The system for recommending supplementary industries for an incomplete industrial chain based on deep reinforcement learning according to any one of claims 1 to 7, characterized in that: include: Data preprocessing module, simulation network generation module, model training module, and supplementary industry recommendation module; Among them, the data preprocessing module integrates the information of enterprises under each industry in the existing industrial chain network data to form industrial data, and uses an attribute importance analysis method to analyze the proportion of each attribute in the industrial data; the industrial chain network data is decomposed according to administrative regions to generate multiple regional and incomplete sub-networks; each sub-network only contains the enterprise information of the corresponding region, and the enterprise information is integrated into the industrial data of the region, and the industry weight is calculated according to the proportion. The simulated network generation module generates networks of different sizes based on the Barabasi-Albert network model; generates a residual network by randomly deleting nodes from the generated network, assigns random weights to the remaining nodes, and calculates the node importance to form a simulated network data set. The model training module constructs a scoring function for link-filling behavior based on industry importance and network connectivity metrics. This scoring function is then used to score missing nodes in each network in the simulated network dataset. The module then selectively adds nodes and updates their states, repeating this process and recording the scores to form a training dataset. The training dataset is split into a training set and a test set at an 8:2 ratio, and a deep reinforcement learning model is trained to learn the scoring mechanism for link-filling behavior. The supplementary industry recommendation module extracts relevant network data of the chain to be supplemented from the industrial chain network dataset, combines the weight of each node with the weight of the neighboring nodes to calculate its importance, and uses a deep reinforcement learning model to recommend supplementary industries for the incomplete regional industrial chain network.

Images