Optimization and Strong Chain Extension Strategy of Hainan Free Trade Port Industry Chain Based on Multi-layer Network Dynamic Programming
Publicado en línea: 17 mar 2025
Recibido: 24 oct 2024
Aceptado: 30 ene 2025
DOI: https://doi.org/10.2478/amns-2025-0839
Palabras clave
© 2025 Shi Wang et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.
The construction of HFTP is an important strategic measure taken by the Chinese government to promote economic openness and transformation and upgrading in the context of economic globalization [1]. As the largest special economic zone in China, the organization of HFTP is aimed at building a more open and free economic system [2]. Hainan Island has a superior geographical location, abundant natural resources, and a unique ecological environment. These advantages provide a firm basis for the structure of HFTP. The Chinese government fully utilizes Hainan's unique geographical advantages, accelerates the upgrading of industrial structure, attracts international investment, promotes technological innovation, and further improves the level of opening up to the outside world through the construction of a free trade port [3]. As the core of economic activities, the industrial chain connects various links of production, circulation, and consumption, determining the comprehensive strength and competitiveness of an industry [4]. It is the key cornerstone of building a modern industrial system and an important support for achieving high-quality growth. In the era of intelligence and informatization, traditional manufacturing enterprises are facing unprecedented challenges.
The inefficient labor and resource intensive development model constantly compresses the profit margins of enterprises, making it difficult to adapt to rapidly changing market demands [5]. Therefore, promoting the extension of manufacturing enterprises from low-end links to high-end links in the industrial chain has become an important strategic direction for enterprise transformation and upgrading [6]. Through technological innovation, brand building, and market expansion, enterprises can enhance their added value, strengthen their market competitiveness, and achieve sustainable development [7]. The modern industrial system is the material and technical basis for HFTP to take the lead in realizing Chinese path to modernization. In this system, strong chain and extended chain strategies have become the key paths to building a modern industrial system. The strong chain strategy aims to enhance the core competitiveness of the entire industry chain by strengthening technological innovation, talent cultivation, and supply chain management in the upstream links of the industry chain. The chain extension strategy focuses on expanding downstream links in the industrial chain, by enhancing the added value of products or services, promoting service-oriented transformation, and developing e-commerce, to achieve the extension and expansion of the industrial chain.
In the industrial chain structure of HFTP, modern service industry, high-end manufacturing industry, and digital economy are the three core areas. The modern service industry plays a core role in the entire industry chain, covering multiple fields such as tourism, finance, logistics, and education. High end manufacturing and digital economy are gradually emerging as important forces driving industrial upgrading. Especially in the context of regional integration, Guangdong Province has fully leveraged its advanced experience in high-end manufacturing and technological innovation, injecting new vitality into the development of HFTP. However, the industrial chain growth of HFTP still faces many challenges. Based on multi-layer network dynamic programming and combined with the actual situation of HFTP, this article systematically studies the optimization of the industrial chain and the strategy of strengthening and extending the chain. In addition, this article proposes a market demand forecasting model that combines DL technology. This model can more accurately predict future market demand trends and provide reliable data support for optimizing industrial chain decisions by analyzing multidimensional information such as historical market data, consumer behavior, and policy changes. The innovation points are as follows:
This article make a breakthrough that the limitations of traditional industry chain research that only focuses on a single industry perspective or static analysis, and dynamically studies the HFTP industry chain in a complex system with multiple layers of networks. This article introduces the advanced method of multi-layer network dynamic programming to quantitatively analyze and simulate the industrial chain of HFTP. By constructing a multi-layer network dynamic programming model, the optimal allocation of industrial chain resources and overall performance improvement are achieved, which makes up for the shortcomings of traditional research methods in dealing with complex dynamic systems. This article innovatively proposes a market demand forecasting model that combines DL technology. This model can fully utilize multidimensional information such as historical market data, consumer behavior, and policy changes, and achieve accurate prediction of future market demand trends through the efficient processing capabilities of DL algorithms.
This article begins by analyzing the research background and highlighting the importance of optimizing and strengthening the industrial chain of HFTP. Next, from the perspective of multi-layer network dynamic programming, we will explore its optimization strategies, including measures such as technological innovation for strong chains, as well as directions such as product value-added for extended chains. The core part introduces a market demand forecasting model based on DL technology, which integrates multi-dimensional information for accurate forecasting and helps optimize the industrial chain. After verifying the effectiveness of the model through multi scenario experiments, summarize the results and reflect on limitations in the conclusion to clarify the direction for future research.
In the field of industrial chain optimization, numerous scholars have conducted in-depth research and achieved certain results. Huang et al. [8] constructed a grey GM (1,1) model to predict and analyze the demand for cold chain logistics of agricultural products, providing a basic prediction approach for this field. Yang et al. [9] used multiple linear regression to screen indicators that affect logistics demand, and combined them with grey prediction to estimate the demand for cold chain logistics, making an attempt at method fusion. Chu et al. [10] constructed an indicator system from three dimensions: regional economic development level, supply level, and cold chain logistics service level, and proposed a combined prediction model based on BP neural network (BPNN), which expanded the construction dimensions of the prediction model. Yu et al. [11] used a grey multiple regression combination model to calculate the demand for cold chain logistics of aquatic products in Yantai City, and conducted research on specific regions and product types.
In terms of resource allocation and decision-making algorithms, Ansari et al. [12] used genetic algorithm (GA) to achieve industrial chain resource allocation, which improved the efficiency of the algorithm partly, but had problems such as poor fault tolerance and slow convergence speed. Sun et al. [13] improved the shortcomings of traditional static group decision-making that cannot reflect the dynamic changes of the decision-making object, and proposed a fuzzy dynamic group decision-making algorithm based on AHP, which promoted the development of decision-making algorithms. In other studies on logistics requirement forecasting, Park et al. [14] used the grey Markov chain combination model to forecast the cold chain logistics demand of agricultural products in Beijing Tianjin Hebei region. The results show that the combined model is superior to the single model in accuracy. Abidi et al. [15] proposed a simulation based short-term energy demand load curve prediction method for calculating the reliable load of terminals and battery swapping stations in response to the demand for electric vehicles in port logistics container terminals.
Dirza et al. [16] established an MLP neural network model with three hidden layers to verify the industrial logistics demand in Shanxi Province and achieved good prediction results. Wu et al. [17] used multiple linear regression analysis to calculate the needs for cold chain logistics of agricultural products. However, previous research has certain limitations. This article takes a unique approach and proposes an optimization strategy for the HFTP industry chain based on multi-layer network dynamic programming, comprehensively considering the dynamic relationships and resource allocation of each link in the industry chain. Meanwhile, innovatively constructing a market demand forecasting model based on DL, integrating multidimensional information to accurately predict market demand trends. These two achievements are expected to further promote the upgrading of the industrial chain of HFTP and inject new vitality into its development.
In the strategy of strengthening the industrial chain, technological innovation and R&D investment are key driving forces. Through technological innovation, not only can product quality be improved and production costs reduced, but it can also promote the industrial chain to climb towards high value-added links and enhance the core competitiveness of the industry. Talents are the cornerstone of industrial development. In terms of talent cultivation and introduction, on the one hand, increase investment in training local talents in Hainan; On the other hand, introducing advanced management experience and cutting-edge technological knowledge can promote the integration of local industrial chains with the international community and enhance the level of internationalization. Efficient supply chain management is the guarantee for the stable operation of the industry. Establish deep strategic partnerships with suppliers, build efficient supply chain networks, and ensure timely and stable supply of raw materials. By utilizing advanced technologies such as the Internet of Things and big data, we aim to achieve intelligent management of the supply chain, improve its transparency, traceability, and operational efficiency, and reduce operational risks. Brand is the intangible asset of a company. In terms of brand building and market promotion, we carefully create a unique brand image, enhance product awareness and reputation, expand the market through various marketing channels, and make our products stand out in fierce market competition.
In the strategy of extending the industrial chain, increasing product added value is an important direction. By extending diversified product lines, continuously launching new products, expanding product categories, meeting different customer needs, expanding market coverage, and improving product added value and corporate profitability. Market expansion and internationalization are the necessary path for the sustainable development of the industry. The extended chain strategy helps enterprises enter new market areas, not only by deeply cultivating the domestic market, but also actively exploring international markets, integrating into the global value chain, and enhancing the competitiveness and influence of enterprises in the international market. The service-oriented transformation has opened up new paths for extending the industrial chain. By providing high-quality after-sales service, professional training services, and other value-added services, we aim to enhance customer stickiness, build a service-oriented business model, shift from simple product supply to "product+service", and effectively extend the industrial chain. The chain extension strategy can expand online sales channels through the use of e-commerce platforms, form complementary relationships with offline channels, and improve product sales efficiency. Meanwhile, the extended chain strategy utilizes big data technology to analyze market trends, consumer demand, and other information, providing data support for product upgrades and service optimization.
In the process of optimizing the industrial chain, DL plays an increasingly crucial role, among which BPNN performs outstandingly in the field of market demand forecasting. Accurately predicting market demand is of great significance for optimizing the industrial chain and can help enterprises allocate production capacity reasonably. When enterprises use DL models to accurately grasp market demand trends, they can scientifically arrange production plans based on predicted results, avoid warehouse backlog caused by overproduction, reduce inventory costs, and prevent market opportunities from being missed due to insufficient production, thereby effectively improving enterprise profits. BPNN is an artificial neural network model trained based on the error backpropagation algorithm. Its structure consists of multiple layers of neurons, including an input layer, one or more hidden layers, and an output layer.
In practical operation, the input layer receives external data such as historical sales data, market dynamics information, consumer preferences, and other multidimensional data. These data are processed through complex nonlinear transformations in the hidden layer, and the output layer provides prediction results. The essence of BPNN's continuous optimization lies in comparing each output result with the expected output in detail, using the error backpropagation mechanism to continuously update and adjust the weights and thresholds, and using gradient descent to make the model's predicted results approach the true values. With the increase of training data and training frequency, BPNN's prediction of market demand becomes more accurate, providing powerful data support for decision-making in various links of the industry chain, further promoting the majorization and upgrading of the industry chain, and making the industry more adaptable and competitive in market competition. Figure 1 shows the topology of BPNN.
BPNN topology structure
Convolutional Neural Network (CNN) is a unique model based on local convolution and backpropagation algorithms, which differs from fully connected neural network structures. It was initially widely used as a handwritten digit image recognition model. Its classic structure includes one input layer for receiving raw data; Two convolutional layers are used to extract local attributes from the input data using different convolution kernels, and to mine key information in the data; Two downsampling layers are used to reduce feature dimensionality, decrease computational complexity, and preserve important features; Two fully connected layers integrate the processed features; Finally, there is an output layer that provides the predicted results. In handwritten font recognition tasks, the CNN recognition rate of this structure exceeds 95%, reaching commercial level.
The core idea is to use convolutional kernels to slide on data, extract local features, downsample to further refine features, and then use fully connected neural networks for comprehensive analysis to achieve accurate recognition of handwritten digits. With the development of technology, improved CNNs can not only process images, but also process and model complex data such as speech and text. This article fully combines the powerful feature extraction ability of CNN and the optimization advantages of BPNN based on error backpropagation algorithm to construct a market demand forecasting model based on CNN and BPNN (as shown in Figure 2). This model can deeply explore the complex features and potential patterns in market data, effectively integrate multidimensional market information, provide more accurate results for market demand forecasting, provide strong support for decision-making in various links of the industrial chain, and help optimize and upgrade the industrial chain.
Model structure
In the data preprocessing stage, we adopted the commonly used method of neural network pre prediction analysis - data normalization method. This method aims to convert all data into the range of 0 to 1, thereby accelerating the convergence process of the neural network and reducing its prediction error. The normalization process used in this article follows the following formula:
Among them,
In order to accurately predict the monthly market need for a product, our first step is to calculate the coefficient of variation for its monthly sales data. In this process, we adopted the
In the formula:
BPNN consists of an input layer, an intermediate hidden layer, and an output layer. In this network, vertical full connections are achieved between layers, while neurons in the same layer remain unconnected. It is a feedforward network trained based on the error backpropagation algorithm. This network adjusts the weights and thresholds in multi-layer connections layer by layer from back to front by minimizing the error between the actual output and the expected output. As this error backpropagation and correction process continues, the precision of the neural network's response to input patterns also continues to improve. When building a BPNN model, the initial step is to assign a random value within the (-1,1) interval to each connection weight and threshold. Subsequently, a set of samples is randomly selected from the training sample set, including the input vector
In the formula,
In the formula:
Dynamic programming method is an optimization approach that balances current and future benefits, cleverly distinguishing and combining the current stage with subsequent stages. In this process, the optimal decision for each segment is selected based on a global perspective, often different from the local optimal choice for that segment. The basic equations of dynamic programming form the cornerstone of recursive piecewise solving, which can usually be expressed in the following general form:
Among them,
When constructing BPNN, the determination of the number of hidden layer nodes usually follows empirical formulas and is optimized using a step-by-step experimental method. Specifically, first set an initial number of hidden layer nodes, and then gradually increase the number of nodes based on this. After each increase, the predictive performance of the network is evaluated. Finally, the number of nodes corresponding to the network with the best predictive performance is selected as the number of neurons in the hidden layer. The empirical formula is as follows:
In this formula,
In order to comprehensively verify the validity of our model, a comparative experiment was carefully designed to compare the model combining CNN and BPNN with the traditional prediction model based on the grey GM (1,1) model. During the experiment, a large amount of historical market demand data was selected within the same time period, covering market fluctuations in different industries, regions, and seasons. By analyzing and processing these data, two models are used to predict market demand. Figure 3 visually compares the accuracy of market demand forecasting between the two. It is evident from the figure that the prediction accuracy of our model is significantly higher than that of traditional prediction models based on the grey GM (1,1) model. This is because the model in this article fully combines the powerful feature extraction ability of CNN and the optimization advantage of BPNN based on error backpropagation algorithm, which can deeply explore the complex features and potential patterns in market data. Whether it is subtle changes in consumer behavior or fluctuations in the macroeconomic environment, they can be effectively captured. At the same time, the model can effectively integrate multidimensional market information, providing more accurate and reliable results for market demand forecasting.
Comparison of prediction accuracy
In addition to its significant advantage in prediction accuracy, this model also performs well in terms of efficiency in market demand forecasting tasks. Figure 4 presents a visual comparison of the prediction time between the model combining CNN and BPNN in this article and the traditional grey GM (1,1) model. As can be clearly seen from the figure, the prediction time of the model in this article is significantly less. Thanks to the ingenious design of the model structure, CNN's convolution operation can quickly extract preliminary features from a large amount of market data, greatly improving processing speed through parallel computing. BPNN, based on error backpropagation optimization algorithm, is efficient and accurate in parameter adjustment process. Compared with traditional models, it can quickly mine complex features and potential patterns in market data, effectively integrate multi-dimensional information such as historical sales data, consumer preference data, and market dynamic data, and complete high-quality market demand forecasting tasks in a short period of time.
Comparison of predicted time consumption
Recall rate refers to the proportion of data that are correctly predicted as positive samples among all data that are actually positive samples. In market demand forecasting tasks, the level of recall directly affects the degree of capturing the true market demand. Figure 5 clearly presents the comparison between the model combining CNN and BPNN in this article and the traditional grey GM (1,1) model in terms of recall rate. It is not difficult to see that the recall rate of the model in this article is superior. This is because CNN has strong capabilities in processing and modeling various complex data, and can quickly extract key features from market data. BPNN utilizes error backpropagation mechanism and gradient descent method to continuously update and adjust weights and thresholds, allowing the model's prediction results to continuously approach the true values. The combination of the two can more quickly and comprehensively explore complex features and potential patterns in market data compared to traditional models, thereby effectively improving recall rates and providing strong support for accurate decision-making in various links of the industry chain.
Comparison of recall rates
Figure 6 clearly shows the comparison of F1 values between the model combining CNN and BPNN in this paper and the traditional grey GM (1,1) model in market demand forecasting tasks. It can be intuitively observed from the figure that the F1 value of the model in this article far exceeds that of traditional models. This is because CNN has powerful feature extraction capabilities, which can quickly process and model various complex market data. Whether it is massive historical sales data or constantly changing market dynamic data, it can accurately analyze them. BPNN utilizes error backpropagation mechanism and gradient descent method to continuously update and adjust weights and thresholds, promoting the model's prediction results to be more in line with the true values. The strong combination of the two enables our model to more deeply explore the complex features and potential patterns in market data compared to traditional models, thereby obtaining higher F1 values and providing better results for market demand forecasting.
Comparison of F1 values
Mean Square Error (MSE) is one of the key indicators for assessing the precision of model predictions. The lower its value, the smaller the deflection between the predicted value and the true value of the model. Figure 7 visually presents the MSE comparison between the model combining CNN and BPNN in this article and the traditional grey GM (1,1) model in market demand forecasting tasks. It can be clearly seen from the figure that the MSE of the model in this article is significantly lower. This is mainly due to CNN's excellent complex data processing and modeling capabilities, which can quickly and accurately extract key features from massive market data. BPNN relies on the error backpropagation mechanism and uses gradient descent to constantly optimize and adjust the weights and thresholds, allowing the model predictions to continuously approach the true values. The synergistic effect of the two enables the model in this article to more deeply explore the complex features and potential patterns in market data compared to traditional models, thereby effectively reducing prediction errors and outputting results that are closer to real market demand in market demand forecasting.
MSE comparison
Figure 8 clearly shows the comparison between the model combining CNN and BPNN in this article and the traditional grey GM (1,1) model in terms of user satisfaction. It is not difficult to see from the graph that the model in this article achieved higher satisfaction. This is thanks to the ingenious design of the model, CNN's powerful processing and modeling capabilities, which can quickly analyze various complex market data and accurately extract key features. BPNN utilizes error backpropagation mechanism and gradient descent method to continuously update weights and thresholds, promoting the prediction results to approach the true values infinitely. Compared to traditional models, the model proposed in this article can delve deeper into the complex features and potential patterns in market data, thereby providing users with prediction results that are more in line with their actual needs. Accurate predictions help users seize the initiative in market decision-making, avoid resource waste, and significantly improve their experience in using the model, resulting in higher satisfaction.
Satisfaction comparison
This article is based on the current growth status of HFTP, using multi-layer network dynamic programming as the theoretical foundation, and deeply and systematically studies the optimization of the industrial chain and the strategy of strengthening and extending the chain. Meanwhile, this article innovatively proposes a market demand forecasting model that combines CNN and BPNN. This model breaks the limitations of traditional forecasting by comprehensively analyzing multidimensional information such as historical market data, and more accurately predicts future market demand trends. The experimental results show that the prediction model performs well in key indicators such as accuracy, recall rate, and F1 score, with shorter prediction time and higher user satisfaction.
However, the model presented in this article also has certain limitations. For example, the comprehensiveness and accuracy of data acquisition still need to be improved, and there may be missing data in some niche markets or emerging fields, which affects the generalization ability of the model; The model has room for improvement in response speed and adaptability to complex and changing external environments, such as sudden global events, significant policy adjustments, etc. Subsequent research can focus on improving the data collection system, optimizing model structure, and other aspects to further enhance model performance, providing more solid support for the continuous optimization of the HFTP industry chain.