From raw material storage through final product distribution, a cold supply chain is a technique in which all activities are managed by temperature. The expansion in the number of imported meat and other comparable commodities, as well as exported seafood has boosted the performance of cold chain logistics service providers. On the basis of the standard basic-pursuit (BP) neural network, a rough BP particle swarm optimization (PSO) neural network model is constructed by combining rough set and particle swarm algorithms to aid cold chain food production enterprises in quickly picking the best cold chain logistics service providers. To reduce duplicate information in the original data and make the input index more compact, the model employs rough set. Instead of using gradient descent to train the weights of the neural network, particle swarm optimization is utilized to ensure that the output results are not readily caught in local minima and that the network’s generalization capacity is improved. Finally, an example is presented to demonstrate the model’s validity and viability. The findings reveal that the model’s prediction error is 40.94 percent lower than the BP neural network model, and the prediction result is more accurate and dependable, providing a new technique for cold chain food production companies to swiftly pick the best cold chain logistics service provider.
The growth in the consumer food quality and demand has brought both possibilities and problems to the development of cold chain logistics for fresh agricultural goods, but the rise in demand has also resulted in an increase in waste [
Because a dependable service provider is critical to a company’s success, academics both at home and abroad have conducted extensive research on service provider selection [
In terms of evaluation methods, reference [
With the deepening of research and the popularization of computer technology, most scholars have begun to use intelligent algorithms to evaluate and select suppliers. Among them, the artificial neural network (ANN) has unique advantages in dealing with the complex relationship between the supplier evaluation indicators with its unique nonlinear adaptive information processing capability. The use of a BP neural network in supplier selection was introduced in reference [
After sorting through the previous research findings, this paper employs the particle swarm algorithm to optimize the BP neural network for selecting the cold chain logistics service provider, which not only compensates for the BP neural network algorithm’s tendency to fall into the local minimum value, but also eliminates the subjective influence of factors, making the evaluation selection results more objective and fair. The rough set theory, which is simple to apply and does not alter the information contained in the data, is used to minimize the amount of input indicators in order to overcome the problem of poor network generalization ability caused by redundant input indications. Finally, an example is used to demonstrate the algorithm’s viability.
The cold chain food production firms can accurately pick and assess cold chain logistics service providers is determined by whether the evaluation index method is scientific, thorough, and effective. Because various organizations have distinct demands for logistics service providers, the selection criteria are also variable. The companies whose products must be frozen and delivered should assess not just typical logistics service providers’ quality, cost, service, and other indicators, but also the particular skills and levels. As a result, this study uses dairy cold chain logistics service providers as an example, discusses [
The financial indicators are the embodiment of the cold chain logistics service provider’s business performance in
The model used by cold chain logistics service providers is upgraded based on the properties of the BP neural network. The PSO method is integrated with the BP neural network in this article, and the rough set theory difference moment is applied. The new matrix technique reduces the input index, and the proposed model enhances the original BP neural network model in two ways.
It is frequently important to simplify the input indications and erase redundant information in order to increase the accuracy of the BP neural network’s prediction outputs. The principal component analysis is now the most popular approach for simplifying the input indicators according to study findings on service provider selection. This approach uses the concept of dimensionality reduction to reduce a large number of linearly connected indicators to a small number of unrelated indications with the goal of preserving as much information as possible. That is, the principal component is a linear combination of the original indicators. The correlation between the indicators is, however, a condition of the principal component analysis approach. For starters, this approach is incapable of dealing with nonlinear situations. Second, the degree of correlation between the indicators is either too high or too low, affecting the accuracy of the results. The rough set theory can remove the unnecessary data from the original data without changing the information stated in the data, and it doesn’t require any prior knowledge beyond the data set to be processed. When it comes to examining index reduction, the set theory offers greater advantages. Because the BP neural network modifies the connection weights using the gradient descent approach, the network’s adaptability to new input is restricted, and the output result is a local minimum value. Researchers are progressively combining bionic intelligence algorithms with neural networks as research progresses. Because the intelligent algorithm has excellent global convergence and durability, combining the two can increase the neural network’s prediction accuracy and generalization mapping ability. The Genetic algorithm (GA), as the most extensively used evolutionary algorithm, may be used to solve a variety of complicated issues, but the optimization process is difficult to regulate because there are too many parameters to control in the connections of selection, crossover, and mutation. When dealing with optimization issues, the PSO algorithm’s operation is generally straightforward, does not require the function’s gradient information or other prior knowledge, and may be substantially parallelized. As a result, instead of using the genetic method to train the weights and thresholds of the neural network, this research employs the particle swarm approach. As a result, the neural network’s performance is improved.
The proposed approach works by first preprocessing the neural network’s input indications using rough sets. That is, normalize and discretize the evaluation index data to create a decision table, then minimize the evaluation index and eliminate the redundant indexes using the difference matrix improved technique. The reduced evaluation index is then used to learn and train the PSO-BP neural network. At this point, the PSO algorithm will optimize the BP neural network’s initial weights and thresholds as particles, and the BP network will be trained using the optimized weights and thresholds.
Among them,
The speed update formula is expressed as:
The location update formula is expressed as:
If
Among them,
Milk, yoghurt, and ice cream are among the items produced by a huge dairy company. To meet the increased market demand, the business plans to collaborate with one of five alternative cold chain logistics service providers. Following further investigation and analysis of these five cold chain logistics service providers, relevant data is obtained, based on the data of 13 cold chain logistics service providers who have previously collaborated (8 for training service providers, 5 for testing service providers quotient), using the rough PSO-BP neural network model proposed in this paper for selection. Because the units of the original data are varied, the original data of 13 cold chain logistics service providers is first adjusted to remove dimensional effect.
Service provider | |||||||
---|---|---|---|---|---|---|---|
1 | 0.5019 | 0.4590 | 0.4219 | 0.3718 | 0.6959 | 85.3 | |
2 | 0.4423 | 0.3960 | 0.3567 | 0.3091 | 0.6766 | 85.25 | |
3 | 0.4687 | 0.4126 | 0.3698 | 0.3146 | 0.6844 | 88.45 | |
4 | 0.4671 | 0.4224 | 0.3856 | 0.3453 | 0.6699 | 83.4 | |
5 | 0.5086 | 0.4515 | 0.4103 | 0.3636 | 0.7078 | 87.9 | |
13 | 0.4787 | 0.5552 | 0.4357 | 0.3923 | 0.7032 | 87.3 |
Because rough sets can only cope with discretized data, the data in
Service provider | |||||||
---|---|---|---|---|---|---|---|
1 | 0 | 0 | 0 | 0 | 1 | 1 | |
2 | 2 | 2 | 2 | 2 | 2 | 1 | |
3 | 2 | 2 | 2 | 2 | 2 | 2 | |
4 | 2 | 2 | 1 | 1 | 2 | 0 | |
5 | 0 | 0 | 1 | 0 | 0 | 2 | |
13 | 1 | 1 | 1 | 0 | 2 |
The 28 evaluation indexes are reduced by the improved difference matrix algorithm. For the convenience of calculation, the 28 evaluation indexes are divided into 4 groups, and the evaluation indexes are reduced respectively.
The universe of discourse in
Domain |
Condition property | Decision attribute D | ||||||
---|---|---|---|---|---|---|---|---|
1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 |
2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 1 |
3 | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 |
4 | 2 | 2 | 1 | 1 | 1 | 1 | 1 | 0 |
5 | 0 | 0 | 1 | 1 | 0 | 0 | 0 | 2 |
6 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
7 | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 2 |
8 | 1 | 1 | 1 | 1 | 1 | 2 | 2 | 2 |
9 | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 2 |
10 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 2 |
11 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 2 |
12 | 0 | 1 | 0 | 1 | 1 | 0 | 1 | 2 |
13 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 2 |
According to the two-dimensional decision table, calculate the difference matrix
According to the difference matrix
From the matrix
The attribute reduction results of evaluation indicators
The attribute reduction process of evaluation indicators
The results of attribute reduction
The proposed algorithm was simulated in MATLAB R2018a. The input node is 13 (the number of cold chain logistics service providers), the output node is 1 (the cold chain logistics service provider’s comprehensive evaluation value), and the hidden layer’s number of neurons is 9. The PSO algorithm has a population size of 40, an acceleration constant of
To predict the comprehensive evaluation results of cold chain logistics service providers, the normalized data of 13 cold chain logistics service providers was imported into the proposed neural network for training, and 8 cold chain logistics service providers were randomly selected as the training set and 5 cold chain logistics service providers as the test set. A comparison study is carried out in two parts to assess the efficiency of the model in this research. Comparing the BP neural network algorithm to the rough-BP neural network algorithm is the first step. Second, the results of the rough PSO-BP neural network method are compared to those of the rough-BP neural network algorithm.
The 28 assessment indicators are first input into the BP neural network without rough set reduction, and the network training curve is shown in
Algorithm | Number of attributes | Training steps | Training time/s | Training error |
---|---|---|---|---|
Before reduction attributes | 28 | 24 | 26.11 | 0.0181 |
After reduction attributes | 15 | 7 | 10.39 | 0.0171 |
The 15 evaluation indicators are input into the PSO-BP neural network and the BP neural network, respectively, after attribute reduction by rough set, and the data of 13 cold chain logistics service providers is trained, and the resultant prediction results are displayed in
The suggested approach has a better fitting impact on the comprehensive assessment findings of cold chain logistics service providers, as shown in
The comparison between the BP neural network algorithm and the rough-BP neural network algorithm, as well as the comparison between the rough PSO-BP neural network algorithm and the rough-BP neural network algorithm shows that after the rough set reduces the attributes of the evaluation indicators, the training samples and neural network structure are improved and the neural network training speed is improved, but the prediction accuracy is not improved. The use of a PSO and BP neural network together increases the neural network’s generalization ability and the algorithm’s accuracy. Finally, the proposed method can increase the prediction accuracy while lowering network running time, allowing the cold chain food production firms to correctly and rapidly pick the best cold chain logistics service provider.
Parameter | Chain service provider | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | ||||||
Rough BP NN | Proposed | Rough BP NN | Proposed | Rough BP NN | Proposed | Rough BP NN | Proposed | Rough BP NN | Proposed | |
Actual value | 88.55 | 85.50 | 88.25 | 88.40 | 85.50 | 88.70 | 86.60 | 88.85 | 88.50 | 87.00 |
Predictive value | 88.4692 | 85.7856 | 88.3131 | 88.3624 | 85.1839 | 88.6453 | 87.1781 | 86.9513 | 88.0517 | 86.8710 |
Prediction error | 0.0009 | 0.0033 | 0.0007 | 0.0004 | 0.0037 | 0.0006 | 0.0067 | 0.0012 | 0.0051 | 0.0015 |
Prediction error and | 0.0171 | 0.0070 |
Input the data of the five cold chain logistics service providers to be selected into the trained PSO-BP neural network model, and select the best partner according to the predicted comprehensive evaluation value.
According to
Service provider to be selected | Prediction score |
---|---|
1 | 88 |
2 | 88.1 |
3 | 86.6 |
4 | 88.75 |
5 | 86.1 |
This paper proposed a rough PSO-BP neural network algorithm that employs a rough set correlation theory to remove the redundant information from the original data in the selection of cold chain logistics service providers, resulting in a more streamlined input index, faster operation, and the use of particle swarm optimization. The swarm method improves the adaptability of the network to varied input and improved the algorithm’s convergence by training the weight of the neural network such that the output leaps out of the local minimum value. It is more accurate and dependable in the selection of cold chain logistics service providers than the classic BP neural network technique, as shown in the example study. When the proposed algorithm has been deployed for a length of time, it is capable of recalling information. As a result, evaluating new cold chain logistics service providers may be more precise and convenient, resulting in a high level of practicability.
How to choose the finest cold chain logistics service provider is critical for cold chain food production firms. Therefore, solving the problem of assessment and selection of cold chain logistics service providers is crucial. The focus of this study is on how the organization picks the service provider on its own, although recent research is focusing on the customer’s role in value co-creation. Because the downstream dealers are the clients of the manufacturing company, the next study will look at how the two-level supply chain of manufacturers and distributors collaborate to choose cold chain logistics service providers.
The authors would like to thank the Deanship of Scientific Research at Taif University for the grant received for this research. This research was supported by Taif University with research grant (TURSP-2020/77).