A Grey Markov Prediction Method Optimized by GSA for Forecasting Agricultural Cold Chain Transportation Volume in China

In the vast, intricate web of modern agriculture, keeping food fresh from the farm to the fork is a race against time and temperature. This is the domain of cold chain logistics, a system that relies on refrigerated trucks and storage facilities to prevent spoilage. Yet, for all its technological sophistication, the industry faces a persistent challenge: knowing exactly how much frozen produce will need to be moved in the coming years. This is not a simple matter of counting crates; it is a complex puzzle where small amounts of data, unpredictable weather, shifting economic policies, and fluctuating consumer demand collide. Traditional forecasting tools often struggle here, requiring massive amounts of historical data or making rigid assumptions that break down when the real world behaves erratically. When planners guess wrong, the consequences are tangible: trucks sit idle with empty space, or worse, perishable goods are left without transport, leading to waste and financial loss.

To navigate this uncertainty, researchers have turned to a branch of mathematics known as grey system theory. Think of this approach as a way to find patterns in a foggy landscape. Unlike other methods that demand a clear, wide-open view of the past, grey theory is designed to work when the picture is incomplete, using the limited information available to make educated guesses about the future. It excels at handling "small samples," where only a few years of data exist, and "nonlinear trends," where growth does not follow a straight, predictable line. By focusing on the hidden rhythms within the data rather than forcing it into a rigid box, these models can offer a glimpse into what lies ahead. However, even these powerful tools have limitations. They can sometimes miss the subtle, random jitters in the data or fail to adapt quickly enough when the underlying conditions change.

A team of researchers from Donghua University in China has recently tackled these limitations by weaving together several advanced techniques into a single, more robust prediction engine. Their goal was to forecast the volume of frozen agricultural products moving through China's logistics network, a critical metric for ensuring food security and economic stability. They started with a multivariate grey model, which looks at the target—frozen goods transport—alongside five key drivers that influence it: the value added by the primary agricultural sector, the number of people working in the tertiary (service) industry, the total volume of agricultural market transactions, the national cost of logistics, and the price index of agricultural products. By feeding these five factors into their model, they hoped to capture the complex interplay that drives demand.

The researchers did not stop at the standard model. They recognized that the traditional way of processing this data treated all historical information equally, which often dilutes the importance of the most recent events. To fix this, they introduced a fractional-order accumulation method. In plain terms, this adjustment allows the model to pay closer attention to the newest data points, giving them more weight than older, potentially outdated figures. This shift ensures the prediction stays sensitive to current trends rather than being anchored too heavily to the past. They also improved how the model calculates its internal "background" values. Instead of using a fixed, unchanging average to smooth out the data, they created an adaptive system that shifts its focus based on how much the data is fluctuating at any given moment. If the numbers are swinging wildly, the model adjusts its sensitivity accordingly, making it more resilient to volatility.

Even with these improvements, the model still produced small errors, or residuals, which are the differences between what the model predicted and what actually happened. To address these, the team employed a Markov chain correction. This technique treats the errors not as random noise, but as a sequence of states that follow their own rules of probability. By analyzing how these errors tend to shift from one state to another, the model can anticipate the likely direction of the next mistake and correct for it before finalizing the prediction. Finally, to ensure all these moving parts worked together perfectly, the researchers used an improved gravitational search algorithm. This is a computational method that mimics the way objects attract each other through gravity to find the best possible solution. It automatically fine-tuned the model's settings, such as how much weight to give recent data and where to draw the lines between different error states, removing the need for human guesswork and ensuring the model was mathematically optimized for the specific dataset.

The results of this hybrid approach were striking. When the team tested their method against historical data from 2013 to 2020 and then used it to forecast the years 2021 through 2024, it significantly outperformed other common prediction techniques. The standard grey models, which lack these advanced corrections, produced errors that were more than five times larger than the new method. Even models that used fractional-order math but lacked the adaptive background values or the Markov correction fell short, with error rates roughly 30 to 50 percent higher. The new model, which combined all the improvements, achieved an average percentage error of just over 10 percent. In contrast, the simpler models without the full suite of corrections struggled with errors ranging from 13 percent to over 50 percent.

The forecast itself paints a picture of steady, robust growth for China's frozen agricultural logistics. The model predicts that the volume of frozen goods transported will rise from approximately 339 million tons in 2021 to nearly 629 million tons by 2024. This represents a consistent annual growth rate of roughly 22 to 23 percent. The researchers noted that while some alternative models predicted even steeper, almost unrealistic spikes in volume, or conversely, growth that was too flat to match the industry's momentum, their method found a middle ground that aligned with the actual economic and agricultural trends observed in the region. The model successfully captured the industry's upward trajectory without falling into the trap of overestimating or underestimating the scale of the expansion.

This study demonstrates that by layering multiple optimization strategies—focusing on recent data, adapting to volatility, correcting for random errors, and using intelligent algorithms to tune the settings—researchers can create a highly reliable tool for complex forecasting. The work provides a concrete example of how mathematical refinement can translate into practical value for supply chain managers and policymakers. By offering a more accurate view of future demand, the model helps ensure that resources are allocated efficiently, reducing waste and supporting the sustainable growth of the agricultural cold chain. The findings suggest that for industries dealing with small datasets and unpredictable trends, the key to better prediction lies not in choosing a single "best" method, but in intelligently combining several complementary techniques to build a system that is both flexible and precise.