Industrial chain situation prediction analysis system and method based on AIGC

Abstract

The invention discloses an industrial chain situation prediction analysis system and method based on AIGC, and belongs to the field of data processing systems or methods specially suitable for the purposes of commerce, finance, management, supervision or prediction. An AIGC model based on the combination of a generative adversarial network (GAN) and a variational auto-encoder (VAE) is constructed; through adversarial training, KL divergence optimization and adaptive learning rate adjustment, a data rule is accurately captured, a moving average method, an autoregressive integral moving average model (ARIMA) and a random forest algorithm in situation prediction are cooperated, prediction stability and accuracy are enhanced, traditional defects are improved, in the aspect of analysis and management, polynomial fitting interpretation trend is deeper, and the prediction efficiency is improved. Risks and opportunity points are accurately identified through a comprehensive index system and a scientific method, decision making is assisted, on the whole, the industrial chain situation prediction analysis process is comprehensively innovated, scientific industrial development decision making is powerfully promoted, and enterprise competitiveness and industrial resource allocation efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing systems or methods specifically suitable for commercial, financial, management, supervision or forecasting purposes, and more specifically, to an AIGC-based industrial chain situation forecasting and analysis system and method thereof. Background Art

[0002] In today's economic development landscape, accurate prediction and analysis of the industrial chain situation is crucial for corporate strategic planning and industry resource allocation. However, traditional industrial chain research methods have exposed many limitations when facing a complex and changing market environment.

[0003] For example, in the construction of models for forecasting and analyzing the industrial chain situation, most of them are based on simple statistical regression or a single machine learning algorithm to simulate data and build models. The quality of the generated simulated data is low, and it is difficult to capture the potential laws and characteristics of the industrial chain data, which affects the accuracy of the forecasting model construction. At the same time, in the stage of situation forecasting, although the traditional forecasting method takes into account the volatility and uncertainty of the data, it lacks a mechanism to deal with data noise interference, which affects the accuracy of the multi-model fusion optimization mechanism, resulting in large deviations in the forecast results and poor stability. In the analysis and management links, the analysis of the development trend of the industrial chain only stays at the simple data description level, and it is impossible to accurately fit the trend curve and interpret the deep changes.

[0004] To sum up, in view of the shortcomings of existing technologies in the industrial chain situation prediction and analysis process, an innovative technical solution is urgently needed to break through these bottlenecks, realize efficient and accurate prediction and analysis of the industrial chain situation, and provide strong support for industrial development. Therefore, we proposed an industrial chain situation prediction and analysis system based on AIGC and its method to solve the above-mentioned problems. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the problems existing in the prior art, the purpose of the present invention is to provide an industrial chain situation prediction and analysis system based on AIGC and a method thereof. First, it constructs an AIGC model based on a combination of a generative adversarial network (GAN) and a variational autoencoder (VAE). Its generator uses a noise vector as input to generate simulated industrial chain data samples, and the discriminator discriminates between real samples and generated samples. The model performance is optimized through adversarial training, and the specific loss function is minimized to prompt the generator to generate data closer to the real data, effectively capture the potential laws and characteristics of the industrial chain data, overcome the problem of low quality of simulated data of traditional models, and improve the accuracy of prediction model construction. The KL divergence term is introduced in combination with the characteristics of the variational autoencoder (VAE) to optimize the overall loss function, so that the generated data has better semantic and structural characteristics, and the model's ability to express industrial chain data is further improved;

[0007] Secondly, after the situation prediction module generates data using the trained AIGC model, it uses the moving average method for smoothing. The moving average period is set according to the data volatility characteristics and historical laws of the industrial chain to reduce data volatility and uncertainty, provide stable data input for subsequent predictions, and improve the defect of traditional prediction methods that lack a data noise interference processing mechanism; combined with the autoregressive integrated moving average model (ARIMA) for prediction, the model order is determined through autocorrelation function and partial autocorrelation function analysis, and the maximum likelihood estimation method is used to estimate the parameters to accurately capture the inherent laws of the data. On this basis, the random forest algorithm is used to optimize the prediction results. By integrating multiple decision trees, the prediction accuracy is improved based on the randomness and diversity of the decision trees, which effectively improves the accuracy and stability of the prediction results, and overcomes the shortcomings of traditional prediction methods such as large deviations and poor stability;

[0008] Then, in the development trend analysis, the analysis and management module uses the polynomial fitting method to perform trend analysis on the forecast data, and interprets the development trend of each link in the industrial chain based on the fitting coefficient. It can accurately fit the trend curve and deeply interpret the deep-level changes, breaking through the limitation of traditional analysis that only stays at the simple data description level, and providing intuitive and in-depth trend information for industrial chain participants to assist in decision-making;

[0009] In terms of potential risk identification, a comprehensive risk assessment indicator system is constructed, covering multi-dimensional factors such as market environment, policies and regulations, and technological innovation. Each indicator is clearly defined and available. The analytic hierarchy process is used to determine the indicator weights and conduct consistency tests to ensure the effectiveness of the weights. Potential risks are then quantitatively assessed based on the weights and actual values ​​of the indicators, and high-risk areas and key risk factors are accurately identified, providing strong support for risk management.

[0010] In the opportunity point identification and evaluation stage, we comprehensively utilize text and visual data, use topic models and image recognition technology to mine potential opportunity information, and use fuzzy comprehensive evaluation methods from multiple dimensions to evaluate and rank opportunity points, effectively identify the most promising opportunity points, assist enterprises in strategic decision-making, and enhance their competitiveness and development advantages in the industrial chain.

[0011] 2. Technical solution

[0012] To solve the above problems, the present invention adopts the following technical solutions.

[0013] The AIGC-based industry chain situation forecasting and analysis system includes the following modules that work together:

[0014] Data collection module: used to collect industry chain related data from multiple data sources, including but not limited to industry reports, corporate financial reports, market research data and social media data.

[0015] Data preprocessing module: used to clean the collected data and remove duplicate, erroneous and incomplete data records;

[0016] AIGC model building module: building an AIGC model based on the combination of generative adversarial network (GAN) and variational autoencoder (VAE);

[0017] Situation prediction module: Use the trained AIGC model to predict the situation of the industrial chain;

[0018] Analysis and management module: Conduct in-depth analysis of forecast results, including development trends, potential risks and opportunity identification of each link in the industrial chain, and provide management decision support based on the analysis results.

[0019] The data preprocessing module processing method of the industrial chain situation prediction and analysis system based on AIGC, the data preprocessing module adopts a statistically based outlier detection method, for numerical data, calculates the mean of the data and standard deviation , will deviate from the mean by more than Data points with a standard deviation of 1 are considered outliers and processed accordingly. is the threshold value set according to the data characteristics. For text data, we use natural language processing technology to perform lexical analysis, syntactic analysis, and semantic analysis to convert text into structured data representation, and then use a Bayesian network-based method to fuse data from different types of data sources. By building a Bayesian network structure and learning the conditional probability distribution between nodes from different data sources, we can integrate multi-source data and enhance the integrity and accuracy of the data.

[0020] The AIGC model construction module construction method of the AIGC-based industrial chain situation prediction and analysis system includes the following steps:

[0021] S1. Model architecture construction: The AIGC model based on the combination of generative adversarial network (GAN) and variational autoencoder (VAE) includes a generator and the discriminator , the generator With noise vector As input, a simulated industrial chain data sample is generated through a multi-layer neural network ,in are the parameters of the generator, the discriminator Receive real industry chain data samples And the samples generated by the generator , and through a multi-layer neural network and To make a judgment, are the parameters of the discriminator;

[0022] S2. Model training and optimization:

[0023] Adversarial training: Optimize model performance through adversarial training and minimize the following loss function:

[0024]

[0025] in, represents the probability distribution of real data, represents the noise vector The probability distribution of the generator and the minimization of the loss function make the data generated by the generator closer to the real data, while the discriminator can more effectively judge the authenticity of the data;

[0026] Combined with the characteristics of variational autoencoder (VAE): In order to make the generated data have better semantic and structural characteristics, the KL divergence term is introduced, and the overall loss function is:

[0027]

[0028] in, The encoder is based on the input The hidden variables obtained The posterior distribution of is the prior distribution, is a hyperparameter used to balance the two losses;

[0029] S3, learning rate adjustment: Use adaptive learning rate adjustment strategy to improve the efficiency and stability of model training. Automatically adjust according to the gradient changes during training, and the adjustment formula is:

[0030]

[0031] in, is a hyperparameter, It is This strategy helps accelerate model convergence and improve training stability.

[0032] The prediction method of the situation prediction module of the industrial chain situation prediction and analysis system based on AIGC includes the following steps:

[0033] S1. Data preparation: Use the trained AIGC model to generate simulated industrial chain data. Since the generated data is volatile and uncertain, the moving average method is used to smooth the generated data. For time series data , calculate its Moving average , the formula is:

[0034] in, , It is a moving average period set according to the volatility characteristics of industrial chain data and the rules of historical data, which provides a more stable data basis for subsequent forecasts through smoothing.

[0035] S2. Autoregressive Integrated Moving Average Model (ARIMA) forecast:

[0036] S2-1. Determination of model parameters: Perform autocorrelation function (ACF) and partial autocorrelation function (PACF) analysis on the smoothed data to determine the order of the ARIMA model. , , Autocorrelation function The calculation formula is:

[0037]

[0038] in, is the mean of the data, yes The first-order autocorrelation coefficient is used to reveal the correlation of data at different lag periods, and the partial autocorrelation function is used to analyze the direct correlation after eliminating the influence of intermediate variables. Both together provide a basis for determining the order of the ARIMA model;

[0039] Based on the determined order, the maximum likelihood estimation method is used to estimate the autoregressive coefficient parameters of the ARIMA model. and the moving average coefficient , by maximizing the likelihood function, we find the parameters that best fit the data, so that the ARIMA model can accurately capture the inherent laws of the data;

[0040] S2-2. Forecasting operation: Use the estimated parameters to build an ARIMA model for forecasting. The predicted value of the period , the calculation formula is:

[0041]

[0042] in, It is predicted based on historical data The value of the period, is the prediction error, and the prediction results of the industrial chain situation at different time points in the future are obtained through iterative calculation;

[0043] S2-3. Optimization of prediction results: Based on the prediction of the ARIMA model, the prediction results are optimized by combining the random forest algorithm in machine learning. The random forest algorithm integrates multiple decision trees and uses the randomness and diversity of decision trees to improve the prediction accuracy. The specific process is as follows:

[0044] Randomly sample the training data and build multiple decision trees;

[0045] Each decision tree randomly selects some features from the feature set for splitting during training;

[0046] For classification problems, the prediction results of multiple decision trees are voted on; for regression problems, the prediction results of multiple decision trees are averaged to obtain the final optimized prediction result.

[0047] The analysis and management method of the analysis and management module of the AIGC-based industrial chain situation forecasting and analysis system includes the following steps:

[0048] S1. Development trend analysis:

[0049] S1-1. Trend fitting: Perform trend analysis on the predicted data of each link of the industrial chain, using the polynomial fitting method, that is, for time series data , assuming that the fitted polynomial is , solve the coefficients by the least squares method , so that the following sum of squared errors is minimized:

[0050]

[0051] in, is the number of data points, is the order of the polynomial. The appropriate order is selected according to the complexity of the data and the trend change to minimize the error between the fitting curve and the actual data, thereby obtaining the trend fitting curve of the data;

[0052] S1-2. Trend interpretation: Explain the development trend of each link in the industrial chain based on the coefficients of the fitted polynomial. When , it indicates that the link is in a growth trend; when the quadratic coefficient When the growth rate slows down gradually, it shows a nonlinear trend, which provides intuitive trend information for industry chain participants and assists decision-making.

[0053] S2. Potential risk identification:

[0054] S2-1. Construction of indicator system: A comprehensive risk assessment indicator system is constructed, taking into account the market environment, policies and regulations, and technological innovation of the industrial chain. The indicator system includes but is not limited to multiple dimensions of market demand fluctuations, raw material price changes, competitor dynamics, and policy adjustment impacts. Each indicator has a clear definition and data acquisition method to ensure its quantifiability and operability;

[0055] S2-2. Determination of indicator weights: The analytic hierarchy process (AHP) is used to determine the weights of each risk indicator. First, industry experts are invited to compare the relative importance of each indicator and construct a judgment matrix. ,in Indicator Relative to the indicator The importance of the index ranges from 1 to 9 and its reciprocal, where 1 means equal importance and 9 means Relative to the indicator Extremely important, then calculate the maximum eigenvalue of the judgment matrix and the corresponding eigenvector , the feature vector Normalization obtains the weight vector of each indicator. To ensure the consistency of the judgment matrix, the consistency index is calculated. , and compared with the random consistency index RI, where RI is based on the matrix order Obtain from the corresponding standard table and calculate the consistency ratio ,when When , the judgment matrix is ​​considered to have satisfactory consistency and the weight vector is valid;

[0056] S2-3. Quantitative risk assessment: Based on the determined weight vector and the actual data values ​​of each indicator, a quantitative assessment is conducted on the potential risks of each link in the industrial chain. Assuming that the risk assessment value is , the calculation formula is:

[0057]

[0058] in, It is The weight of the indicator, It is Through this quantitative assessment, potential risks can be ranked and analyzed to identify high-risk areas and key risk factors;

[0059] S3. Opportunity identification and assessment:

[0060] S3-1. Data mining and feature extraction: Comprehensively utilize various data to identify opportunities, including but not limited to industry news and research reports of text data, and also including but not limited to industry maps and market share distribution maps of visual data. For text data, use the topic model (LDA) in natural language processing technology to extract key themes and topics, and mine information related to industry chain innovation, emerging market demand, and policy support. For visual data, use deep learning-based image recognition technology to extract and classify images through convolutional neural networks (CNNs) to identify underdeveloped areas in industry maps or market segments with high growth potential in market share distribution maps;

[0061] S3-2. Opportunity evaluation and ranking: Evaluate the identified potential opportunities from multiple dimensions, including market size, growth rate, degree of competition, and technical feasibility. Use the fuzzy comprehensive evaluation method to first determine the evaluation factor set. and Commentary , construct the fuzzy relationship matrix ,in Indication factors Comments The membership degree of each factor is combined with the weight vector , calculate the fuzzy comprehensive evaluation results , get the membership distribution of each opportunity point at different comment levels, sort the opportunity points by analyzing the fuzzy comprehensive evaluation results, find out the most potential opportunity points, and provide a basis for the company's strategic decision-making.

[0062] 3. Beneficial effects

[0063] Compared with the prior art, the advantages of the present invention are:

[0064] (1) This scheme builds an AIGC model based on a combination of a generative adversarial network (GAN) and a variational autoencoder (VAE). The generator uses a noise vector as input to generate simulated industrial chain data samples, and the discriminator distinguishes between real samples and generated samples. The model performance is optimized through adversarial training, and the specific loss function is minimized to enable the generator to generate data closer to the real world, effectively capturing the potential laws and characteristics of industrial chain data, overcoming the problem of low quality of simulated data in traditional models, and improving the accuracy of predictive model construction. The KL divergence term is introduced in combination with the characteristics of the variational autoencoder (VAE) to optimize the overall loss function, so that the generated data has better semantic and structural characteristics, further improving the model's ability to express industrial chain data;

[0065] (2) In this solution, after the situation prediction module generates data using the trained AIGC model, it uses the moving average method to smooth it. The moving average period is set according to the data volatility characteristics and historical laws of the industrial chain to reduce data volatility and uncertainty, provide stable data input for subsequent predictions, and improve the defect of traditional prediction methods that lack a data noise interference processing mechanism; combined with the autoregressive integrated moving average model (ARIMA) for prediction, the model order is determined by analyzing the autocorrelation function and partial autocorrelation function, and the maximum likelihood estimation method is used to estimate the parameters to accurately capture the inherent laws of the data. On this basis, the random forest algorithm is used to optimize the prediction results. By integrating multiple decision trees, the prediction accuracy is improved based on the randomness and diversity of the decision trees, which effectively improves the accuracy and stability of the prediction results and overcomes the shortcomings of traditional prediction methods such as large deviation and poor stability;

[0066] (3) In this solution, the analysis and management module uses the polynomial fitting method to perform trend analysis on the forecast data in the development trend analysis, and interprets the development trend of each link in the industrial chain based on the fitting coefficient. It can accurately fit the trend curve and deeply interpret the deep-level changes, breaking through the limitation of traditional analysis that only stays at the simple data description level, and providing intuitive and in-depth trend information for industrial chain participants to assist decision-making;

[0067] (4) In terms of potential risk identification, this plan builds a comprehensive risk assessment indicator system that covers multiple dimensions such as market environment, policies and regulations, and technological innovation. Each indicator has a clear definition and a feasible way to obtain it. The analytic hierarchy process is used to determine the indicator weights and conduct consistency tests to ensure the effectiveness of the weights. Potential risks are then quantitatively assessed based on the weights and actual values ​​of the indicators, and high-risk areas and key risk factors are accurately identified, providing strong support for risk management.

[0068] (5) In the opportunity point identification and evaluation phase of this plan, text and visual data are comprehensively utilized, and topic models and image recognition technologies are used to mine potential opportunity information. Fuzzy comprehensive evaluation methods are used from multiple dimensions to evaluate and rank opportunity points, effectively identifying the most promising opportunity points, assisting enterprises in making strategic decisions, and enhancing their competitiveness and development advantages in the industrial chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a schematic diagram of the system architecture and data flow of the present invention;

[0070] Figure 2 It is a schematic diagram of the system modules of the present invention;

[0071] Figure 3 The system module operation method of the present invention is shown in FIG. Figure 1 ;

[0072] Figure 4The system module operation method of the present invention is shown in FIG. Figure 2 ;

[0073] Figure 5 The system module operation method of the present invention is shown in FIG. Figure 3 ;

[0074] Figure 6 The system module operation method of the present invention is shown in FIG. Figure 4 . DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the specification of the present invention; it is obvious that the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments, and all other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making creative work are within the scope of protection of the present invention.

[0076] Example:

[0077] The following is based on the drawings in the specification Figure 1-Figure 6 , an embodiment of an industry chain situation prediction and analysis system and method based on AIGC is listed below:

[0078] Step 1: Data collection by the data collection module of the AIGC industry chain situation forecasting and analysis system:

[0079] 1. Industry report collection:

[0080] More than 30 reports on the new energy vehicle industry in the past five years were obtained from well-known market research institutions such as McKinsey and Boston Consulting Group. These reports cover information on the global new energy vehicle market size, sales growth trends, changes in the market share of different types of new energy vehicles (pure electric, hybrid, etc.), the strategic layout of major automakers, and the development path of battery technology. For example, the "McKinsey China Automotive Consumer Insights 2024" report pointed out that the global new energy vehicle sales have achieved an annual compound growth rate of 30% in the past five years, and it is expected to maintain a high growth rate in the next three years.

[0081] 2. Obtaining corporate financial reports:

[0082] We collected the financial reports of 20 new energy vehicle and related parts companies in the past three years, including Tesla, BYD, and NIO. The financial report data includes the company's operating income, net profit, R&D investment, and revenue details of each business segment.

[0083] For example, the main contents of Tesla's third quarter 2024 financial report are:

[0084] Revenue: $25.182 billion

[0085] Net profit: $2.167 billion

[0086] R&D investment: R&D expenses in the third quarter were $1.04 billion;

[0087] Revenue breakdown by business segment:

[0088] Automotive business: revenue of $20.016 billion;

[0089] Model 3 and Model Y production was 443,700 units, and deliveries were 440,000 units;

[0090] The production of Model S, Model X and other models was 26,100, and the delivery volume was 22,900;

[0091] Power generation and energy storage business: revenue of $2.376 billion;

[0092] The installed capacity of energy storage products reached 6.9GWh, a year-on-year increase of 73%. In the first three quarters of this year, the cumulative installed capacity soared to 20.4GWh;

[0093] Energy gross margin reached a record high of 30.5% in the third quarter, with gross margin increasing 596 basis points sequentially despite lower Megapack volumes;

[0094] Services and other businesses: revenue of $2.79 billion.

[0095] 3. Market research data integration:

[0096] Through a combination of online and offline methods, market research was conducted in major global automobile sales markets (such as China, the United States, Europe, etc.), and more than 1,500 consumer questionnaires and dealer feedback were collected. The survey content involved consumers' willingness to buy new energy vehicles, expectations for driving range, price sensitivity, evaluation of the convenience of charging facilities, and awareness and preference for new energy vehicles of different brands. For example, the survey results show that 60% of consumers believe that driving range is a key factor affecting purchase decisions, and the willingness to buy is significantly enhanced when the driving range reaches more than 500 kilometers.

[0097] 4. Social Media Data Mining:

[0098] We used professional data collection tools to capture more than 8,000 posts and articles related to new energy vehicles from social media platforms such as Weibo, Autohome Forum, and LinkedIn. These data include users' sharing of their experience with new energy vehicles, their expectations and evaluations of new models, opinions and analysis by industry experts, and discussions on new energy vehicle policies. For example, on a popular forum on Autohome, users had a heated discussion on the automatic driving function of a new new energy vehicle, and provided feedback on its advantages and problems in actual use.

[0099] Step 2: Data processing by the data preprocessing module of the AIGC-based industrial chain situation forecasting and analysis system:

[0100] 1. Numerical data cleaning:

[0101] Taking the mileage data of new energy vehicles as an example, the mean is calculated to be 450 kilometers, the standard deviation is 50 kilometers, and k is set to 3. After screening, it is found that 15 data points deviate from the mean by more than 3 times the standard deviation. These outliers are due to test errors or data under special working conditions. They are treated as outliers and processed. For battery energy density data, this method is also used to process outliers to ensure the accuracy and reliability of the data.

[0102] 2. Text data processing:

[0103] Perform natural language processing on industry reports and social media texts. For example, in an industry report, it is mentioned that "with the breakthrough of solid-state battery technology, the range of new energy vehicles is expected to increase significantly, and the market competition pattern will change." Through lexical analysis, keywords such as "solid-state battery", "range", and "market competition pattern" are extracted, and semantic analysis is performed to convert them into structured data, such as "technological breakthrough, solid-state battery, impact on range and market competition pattern"; for user experience sharing on social media, such as "I drive a new energy vehicle of a certain brand, the charging speed is very fast, but the noise inside the car is a bit loud", key information such as "brand name", "charging speed", and "in-car noise" are extracted, and sentiment analysis is performed to determine whether the user's satisfaction with the brand of new energy vehicles in these aspects is positive or negative.

[0104] 3. Data Fusion:

[0105] The Bayesian network-based method is used to fuse data from different types of data sources. The Bayesian network structure is constructed, and the enterprise R&D investment, battery technical indicators, market demand, etc. are used as nodes to learn the conditional probability distribution between them. For example, it is found that there is a strong positive correlation between enterprise R&D investment and battery energy density improvement. When the enterprise R&D investment increases, the battery energy density is more likely to increase, thereby integrating multi-source data together to enhance the integrity and accuracy of the data.

[0106] Step 3: Modeling and processing of data based on the AIGC model building module of the AIGC industry chain situation forecasting and analysis system:

[0107] 1. Model architecture construction:

[0108] Construct an AIGC model based on a combination of a generative adversarial network (GAN) and a variational autoencoder (VAE). The generator uses a 150-dimensional noise vector as input and generates simulated new energy vehicle industry chain data samples through a 5-layer neural network, including new energy vehicle sales forecast data, battery cost trend data, parts supply data, etc. The discriminator receives real industry chain data samples and samples generated by the generator, and discriminates through a 6-layer neural network.

[0109] 2. Model training and optimization

[0110] Adversarial Training:

[0111] During training, the loss function is minimized:

[0112]

[0113] After 800 rounds of training, the similarity between the data generated by the generator and the real data has been significantly improved, and the discrimination accuracy of the discriminator has stabilized at more than 90%. For example, for new energy vehicle sales data, the generated data has a high degree of fit with the real data in terms of sales trends and seasonal fluctuations.

[0114] Combined with the characteristics of variational autoencoder (VAE):

[0115] Introduce KL divergence term and set hyperparameters , the overall loss function is:

[0116] In this way, the semantic and structural characteristics of the generated data are significantly improved. For example, the distribution of the generated battery raw material price fluctuation data at different times and regions is more in line with actual market laws.

[0117] 3. Learning rate adjustment:

[0118] Use adaptive learning rate adjustment strategy to set hyperparameters At the beginning of training, the learning rate starts from 0.001. As training progresses, when the gradient changes greatly, the learning rate gradually decreases. For example, when training to 400 rounds, due to the large gradient change, the learning rate is reduced to 0.0005, which effectively accelerates the convergence of the model and improves the stability of training. The model basically reaches a convergence state after 1000 rounds of training.

[0119] Step 4: The situation prediction module of the AIGC-based industrial chain situation prediction and analysis system predicts and processes the data:

[0120] 1. Data preparation:

[0121] The trained AIGC model is used to generate simulated new energy vehicle industry chain data, such as monthly sales data of new energy vehicles. Since the generated data has a certain volatility, the moving average method is used for smoothing. According to the volatility characteristics of the industry chain data and the rules of historical data, the moving average period is set. For time series data , calculate its 6-period moving average ,After smoothing, the fluctuation of the data is significantly reduced, providing a more stable data basis for subsequent ,prediction.

[0122] 2. Autoregressive Integrated Moving Average Model (ARIMA) Forecast

[0123] Model parameter determination:

[0124] Perform autocorrelation function (ACF) and partial autocorrelation function (PACF) analysis on the smoothed data to determine the order of the ARIMA model , , . Estimate the autoregressive coefficient parameters of the ARIMA model by using the maximum likelihood estimation method , , and the moving average coefficient , .

[0125] Prediction Operation:

[0126] The estimated parameters are used to construct an ARIMA model for forecasting the sales volume of new energy vehicles for the next 12 periods. , according to the formula

[0127]

[0128] Perform iterative calculations to obtain the forecast results of the industrial chain situation at different time points in the future.

[0129] Prediction results optimization:

[0130] Based on the prediction of the ARIMA model, the prediction results are optimized by combining the random forest algorithm. The training data is randomly sampled to build 70 decision trees. Each decision tree randomly selects 5 features from the feature set for splitting during training. The prediction results of multiple decision trees are averaged to obtain the final optimized prediction results, which reduces the mean absolute error (MAE) of the prediction from 5,000 vehicles predicted by the ARIMA model alone to 3,000 vehicles, effectively improving the prediction accuracy.

[0131] Step 5: Analysis and management of data based on the analysis and management module of AIGC's industry chain situation forecasting and analysis system:

[0132] 1. Development trend analysis

[0133] Trend fitting:

[0134] Perform trend analysis on the predicted data of each link in the new energy vehicle industry chain, such as using polynomial fitting method for new energy vehicle battery energy density data. Assume that the fitting polynomial is , solve the coefficients by the least squares method, select the appropriate order according to the complexity of the data and the trend change, and get , , , , so that the sum of squared errors is minimized, thus obtaining the trend fitting curve of the data.

[0135] Trend interpretation:

[0136] According to the coefficients of the fitted polynomial, the coefficient of the first term is , indicating that the battery energy density is on an upward trend in the short term, but the quadratic coefficient , which means that the growth rate is gradually slowing down, and the cubic coefficient , which indicates that in the long run there will be new technological breakthroughs that will lead to further increases in energy density, providing intuitive trend information for industry chain participants to assist in their decision-making. For example, battery manufacturers can adjust their R&D investment and production plans accordingly.

[0137] 2. Identification of potential risks

[0138] Construction of indicator system:

[0139] A risk assessment indicator system for the new energy vehicle industry chain is constructed, covering multiple dimensions, including market demand fluctuations (such as the standard deviation of the quarterly growth rate of new energy vehicle market demand), raw material price changes (such as the annual increase in the prices of battery raw materials such as lithium and cobalt), competitor dynamics (such as the market share change rate of major competitors and the speed of new model launches), and the impact of policy adjustments (such as the decline in new energy vehicle subsidy policies and the intensity of implementation of the double-point policy). Each indicator has a clear definition and data acquisition method.

[0140] Determination of indicator weights:

[0141] The analytic hierarchy process (AHP) was used to determine the weight of each risk indicator. 15 industry experts were invited to compare the relative importance of each indicator and construct a judgment matrix. For example, experts believe that the importance of market demand fluctuations relative to raw material price changes is 4. After calculating the maximum eigenvalue of the judgment matrix, , the corresponding eigenvector , normalize the feature vector to get the weight vector of each index. Calculate the consistency index , and the random consistency index RI (according to the matrix order Get from standard table ) comparison, calculate the consistency ratio , it is believed that the judgment matrix has satisfactory consistency and the weight vector is valid.

[0142] Quantitative risk assessment:

[0143] According to the determined weight vector and the actual data value of each indicator, the potential risks of each link in the new energy vehicle industry chain are quantitatively evaluated. Assuming that the risk assessment value of a certain link is , according to the formula

[0144]

[0145] Calculations such as standardized data values ​​for market demand volatility indicators , Standardized data value of raw material price change index Etc., calculate the risk assessment value of this link By sorting and analyzing the risk assessment values ​​of each link, high-risk areas and key risk factors can be identified. For example, the risk assessment value of the battery raw material supply link is high due to price fluctuations and supply stability issues. Enterprises can strengthen supply chain risk management based on this, sign long-term contracts with suppliers or look for alternative materials.

[0146] 3. Opportunity identification and evaluation

[0147] Data mining and feature extraction:

[0148] Comprehensively utilize multiple data to identify opportunities. For text data, such as industry news and research reports, we use the topic model (LDA) to extract key themes and topics, and dig out information such as "the application potential of hydrogen fuel cell vehicles in the commercial vehicle field" and "the development opportunities of intelligent network technology for new energy vehicles". For visual data, such as industry maps and market share distribution maps, we use deep learning-based image recognition technology to extract and classify images through convolutional neural networks (CNNs) to find out that the construction and operation of charging piles in a certain emerging region in the new energy vehicle industry map is not fully developed and has high growth potential.

[0149] Opportunity Assessment Ranking:

[0150] Evaluate the identified potential opportunities from multiple dimensions, including market size, growth rate, degree of competition, and technical feasibility, using the fuzzy comprehensive evaluation method. Determine the evaluation factor set and Commentary , construct the fuzzy relationship matrix , combined with the weight vector of each factor , calculate the fuzzy comprehensive evaluation results For example, for the opportunity point of "application potential of hydrogen fuel cell vehicles in the field of commercial vehicles", its membership distribution at different comment levels was calculated, and the opportunity points were sorted by analyzing the fuzzy comprehensive evaluation results. It was found that this opportunity point has outstanding performance in market size and growth rate, and has high potential, which provides a basis for the company's strategic decision-making. The company can increase its investment in hydrogen fuel cell technology research and development and commercial vehicle market expansion.

[0151] It can be seen from the above embodiments that the industry chain situation prediction and analysis system based on AIGC can effectively realize functions such as data collection, preprocessing, model building, situation prediction, analysis and management in the new energy vehicle industry chain, provide valuable decision-making support information for industry chain participants, and enhance the competitiveness and risk response capabilities of enterprises in the industry chain.

[0152] The above is only a preferred specific implementation of the present invention; however, the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solution and its improved conception within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. The AIGC-based industry chain situation prediction and analysis system is characterized by: Includes the following modules that work together: Data collection module: used to collect industry chain related data from multiple data sources, including but not limited to industry reports, corporate financial reports, market research data and social media data; Data preprocessing module: used to clean the collected data and remove duplicate, erroneous and incomplete data records; AIGC model building module: building an AIGC model based on the combination of generative adversarial network (GAN) and variational autoencoder (VAE); Situation prediction module: Use the trained AIGC model to predict the situation of the industrial chain; Analysis and management module: Conduct in-depth analysis of forecast results, including development trends, potential risks and opportunity identification of each link in the industrial chain, and provide management decision support based on the analysis results.

2. A data preprocessing module processing method of an AIGC-based industrial chain situation prediction and analysis system, comprising the AIGC-based industrial chain situation prediction and analysis system according to claim 1, characterized in that: The data preprocessing module uses a statistically based outlier detection method to calculate the mean of the data for numerical data. and standard deviation , will deviate from the mean by more than Data points with a standard deviation of 1 are considered outliers and processed accordingly. is the threshold value set according to the data characteristics. ,For text data, natural language processing technology is used to perform lexical analysis, syntactic analysis and semantic analysis, and the text is converted into a structured data representation. Then, a Bayesian network-based method is used to fuse data from different types of data sources. By building a Bayesian network structure, the conditional probability distribution between nodes of different data sources is learned, and multi-source data is integrated together to enhance the integrity and accuracy of the data.

3. A method for constructing an AIGC model construction module of an AIGC-based industrial chain situation prediction and analysis system, comprising the AIGC-based industrial chain situation prediction and analysis system according to claim 1, characterized in that: The following steps are involved: S1. Model architecture construction: The AIGC model based on the combination of generative adversarial network (GAN) and variational autoencoder (VAE) includes a generator and the discriminator , the generator With noise vector As input, a simulated industrial chain data sample is generated through a multi-layer neural network ,in are the parameters of the generator, the discriminator Receive real industry chain data samples And the samples generated by the generator , and through a multi-layer neural network and To make a judgment, are the parameters of the discriminator; S2. Model training and optimization: Adversarial training: Optimize model performance through adversarial training and minimize the following loss function: ； in, represents the probability distribution of real data, represents the noise vector The probability distribution of the generator and the minimization of the loss function make the data generated by the generator closer to the real data, while the discriminator can more effectively judge the authenticity of the data; Combined with the characteristics of variational autoencoder (VAE): In order to make the generated data have better semantic and structural characteristics, the KL divergence term is introduced, and the overall loss function is: ； in, The encoder is based on the input The hidden variables obtained The posterior distribution of is the prior distribution, is a hyperparameter used to balance the two losses; S3, learning rate adjustment: Use adaptive learning rate adjustment strategy to improve the efficiency and stability of model training. Automatically adjust according to the gradient changes during training, and the adjustment formula is: ； in, is a hyperparameter, It is This strategy helps accelerate model convergence and improve training stability.

4. A prediction method for a situation prediction module of an AIGC-based industrial chain situation prediction and analysis system, comprising the AIGC-based industrial chain situation prediction and analysis system according to claim 1, characterized in that: The following steps are involved: S1. Data preparation: Use the trained AIGC model to generate simulated industrial chain data. Since the generated data is volatile and uncertain, the moving average method is used to smooth the generated data. For time series data , calculate its Moving average , the formula is: ; in, , It is a moving average period set according to the volatility characteristics of industrial chain data and the rules of historical data, which provides a more stable data basis for subsequent forecasts through smoothing. S2. Autoregressive Integrated Moving Average Model (ARIMA) forecast: S2-1. Determination of model parameters: Perform autocorrelation function (ACF) and partial autocorrelation function (PACF) analysis on the smoothed data to determine the order of the ARIMA model. , , Autocorrelation function The calculation formula is: ； in, is the mean of the data, yes The first-order autocorrelation coefficient is used to reveal the correlation of data at different lag periods, and the partial autocorrelation function is used to analyze the direct correlation after eliminating the influence of intermediate variables. Both together provide a basis for determining the order of the ARIMA model; Based on the determined order, the maximum likelihood estimation method is used to estimate the autoregressive coefficient parameters of the ARIMA model. and the moving average coefficient , by maximizing the likelihood function, we find the parameters that best fit the data, so that the ARIMA model can accurately capture the inherent laws of the data; S2-2. Forecasting operation: Use the estimated parameters to build an ARIMA model for forecasting. The predicted value of the period , the calculation formula is: ； in, It is predicted based on historical data The value of the period, is the prediction error, and the prediction results of the industrial chain situation at different time points in the future are obtained through iterative calculation; S2-3. Optimization of prediction results: Based on the prediction of the ARIMA model, the prediction results are optimized by combining the random forest algorithm in machine learning. The random forest algorithm integrates multiple decision trees and uses the randomness and diversity of decision trees to improve the prediction accuracy. The specific process is as follows: Randomly sample the training data and build multiple decision trees; Each decision tree randomly selects some features from the feature set for splitting during training; For classification problems, the prediction results of multiple decision trees are voted on; for regression problems, the prediction results of multiple decision trees are averaged to obtain the final optimized prediction result.

5. The analysis and management method of the analysis and management module of the AIGC-based industrial chain situation forecasting and analysis system comprises the AIGC-based industrial chain situation forecasting and analysis system according to claim 1, characterized in that: The following steps are involved: S1. Development trend analysis: S1-1. Trend fitting: Perform trend analysis on the predicted data of each link of the industrial chain, using the polynomial fitting method, that is, for time series data , assuming the fitted polynomial is , solve the coefficients by the least squares method , so that the following sum of squared errors is minimized: ； in, is the number of data points, is the order of the polynomial. The appropriate order is selected according to the complexity of the data and the trend change to minimize the error between the fitting curve and the actual data, thereby obtaining the trend fitting curve of the data; S1-2. Trend interpretation: Explain the development trend of each link in the industrial chain based on the coefficients of the fitted polynomial. When , it indicates that the link is in a growth trend; when the quadratic coefficient When the growth rate slows down gradually, it shows a nonlinear trend, which provides intuitive trend information for industry chain participants and assists decision-making. S2. Potential risk identification: S2-1. Construction of indicator system: A comprehensive risk assessment indicator system is constructed, taking into account the market environment, policies and regulations, and technological innovation of the industrial chain. The indicator system includes but is not limited to multiple dimensions of market demand fluctuations, raw material price changes, competitor dynamics, and policy adjustment impacts. Each indicator has a clear definition and data acquisition method to ensure its quantifiability and operability; S2-2. Determination of indicator weights: The analytic hierarchy process (AHP) is used to determine the weights of each risk indicator. First, industry experts are invited to compare the relative importance of each indicator and construct a judgment matrix. ,in Indicators Relative to the indicator The importance of the index ranges from 1 to 9 and its reciprocal, where 1 means equal importance and 9 means Relative to the indicator Extremely important, then calculate the maximum eigenvalue of the judgment matrix and the corresponding eigenvector , the feature vector Normalization obtains the weight vector of each indicator. To ensure the consistency of the judgment matrix, the consistency index is calculated. , and compared with the random consistency index RI, where RI is based on the matrix order Obtain from the corresponding standard table and calculate the consistency ratio ,when When , the judgment matrix is ​​considered to have satisfactory consistency and the weight vector is valid; S2-3. Quantitative risk assessment: Based on the determined weight vector and the actual data values ​​of each indicator, a quantitative assessment is conducted on the potential risks of each link in the industrial chain. Assuming that the risk assessment value is , the calculation formula is: ； in, It is The weight of the indicator, It is Through this quantitative assessment, potential risks can be ranked and analyzed to identify high-risk areas and key risk factors; S3. Opportunity identification and assessment: S3-1. Data mining and feature extraction: Comprehensively utilize various data to identify opportunities, including but not limited to industry news and research reports of text data, and also including but not limited to industry maps and market share distribution maps of visual data. For text data, use the topic model (LDA) in natural language processing technology to extract key themes and topics, and mine information related to industry chain innovation, emerging market demand, and policy support. For visual data, use deep learning-based image recognition technology to extract and classify images through convolutional neural networks (CNNs) to identify underdeveloped areas in industry maps or market segments with high growth potential in market share distribution maps; S3-2. Opportunity evaluation and ranking: Evaluate the identified potential opportunities from multiple dimensions, including market size, growth rate, degree of competition, and technical feasibility. Use the fuzzy comprehensive evaluation method to first determine the evaluation factor set. and Commentary , construct the fuzzy relationship matrix ,in Indication factors Comments The membership degree of each factor is combined with the weight vector , calculate the fuzzy comprehensive evaluation results , get the membership distribution of each opportunity point at different comment levels, sort the opportunity points by analyzing the fuzzy comprehensive evaluation results, find out the most potential opportunity points, and provide a basis for the company's strategic decision-making.

Images