Enterprise supply chain risk assessment based on improved neural network algorithm and machine learning

Abstract

It is of practical significance to study the decision-making subject in the supply chain under the influence of risk aversion to make a decision and make the supply chain compete in an orderly market environment. In order to improve the effect of enterprise supply chain risk assessment, this paper improves the traditional neural network algorithm, combines machine learning methods and supply chain risk assessment time requirements to set system function modules, and builds the overall system structure. Considering the multiple relationship attributes of supply chain risk knowledge, this paper uses a multi-element semantic network to represent the network structure of supply chain risk knowledge, and proposes a multi-level inventory control modelThis is based on the inventory of the coordination center and other retailers’ procurement/relocation strategy models. After building the model, this paper designs a simulation test to verify and analyze the model performance. The research results show that the model proposed in this paper has a certain effect.

Keywords

Improved algorithm neural network machine learning enterprise supply chain risk assessment

1 Introduction

Since the 1990 s, the development of the IT industry has promoted the formation of global economic integration. Global economic integration has made the relationship between enterprises closer and closer, and the degree of interdependence has been strengthened, which has brought a broader market to enterprises. However, the uncertainty of the external environment and some policy guidance of the country will cause the uncertainty of customer demand to increase, thus bringing greater volatility to the operation of the market. Faced with market uncertainty, corporate decision makers and customers will exhibit risk aversion, and the supply chain where the enterprise and customers are located is a risk aversion supply chain. Supply chain risk is the possibility of supply chain outcome variables and changes in value distribution. In short, supply chain risk refers to the impact and possibility of supply and demand mismatch in the supply chain. The inconsistency between supply and demand causes the uncertainty of customer demand. The increase in the uncertainty of customer demand brings risk loss to the enterprise, and even makes the enterprise bankrupt. It will also increase the cost of the supply chain and weaken the competitiveness of the supply chain [1]. Therefore, the risk that market uncertainty brings to the supply chain is a hot topic in academic research. The previous research on the decision-makers in the supply chain under risk-neutral conditions cannot be used to guide the supply chain operation practice. According to the risk behavior of the decision makers in the supply chain, the dynamic game model of the risk-averse supply chain is established and its long-term evolution process is analyzed to grasp the evolution process of the supply chain under dynamic conditions, which is beneficial to the management of the supply chain. It is impossible for an enterprise to obtain all the information of its competitors during its operation. If an enterprise wants to obtain more information, it must pay more costs. Therefore, the main body of the real supply chain makes decisions under the asymmetric information. In addition, under the influence of the decision-maker’s own factors, decision-making often shows limited rational behavior. Uncertainty in the market makes the main decision makers of the supply chain exhibit risk avoidance behaviors in order to reduce losses, and in order to obtain maximum effectiveness, customers will also exhibit risk avoidance behaviors. Under the influence of risk-averse behavior, the question of how the main decision-makers of the supply chain make decisions should arouse great attention from academia. The methods and theories of supply chain management based on the risk attitudes of decision makers have realistic supply chain operating characteristics and can draw strategies that are conducive to the operation of supply chain enterprises. Therefore, the risk attitude of decision makers is a key factor that affects the operation of the supply chain and should arouse the attention of academia [2].

In theory, applying management theory, economics, and nonlinear dynamics theory to the real economic problems of supply chains with risk-averse behaviors for decision makers to study the evolutionary behavior of variable long-term games is conducive to understanding the characteristics of risk-averse supply chain operation mechanisms from multiple perspectives, which reflects the research characteristics of interdisciplinary penetration and mutual influence. The supply chain is a complex self-organizing system, and the impact of the external environment on the stability of the supply chain is also very obvious. The non-linear discrete dynamic game model of the risk-averse supply chain established based on the realistic competition model has studied the influence of behavioral factors on the complexity of the supply chain, made up for some of the deficiencies of previous research, and has significant theoretical significance [3].

Because the external environment in which the company is located is very complex, the company will be affected by many factors, such as the untimely supply of raw materials, the shortened life cycle of products, the personalization and diversification of customer needs, etc. These uncertain factors will bring uncertainty and risk to corporate decision-making. Therefore, it is of practical significance to study the decision-making subject in the supply chain under the influence of risk aversion to make a decision to make the supply chain compete in an orderly market environment.

2 Related work

The literature [4] believed that although the replacement of suppliers is a short-term solution to exchange rate changes, in the long run we should design risk sharing agreements between suppliers and client companies. The literature [5] believed that visualization, capacity management and dynamic optimization of real-time artificial intelligence agents are urgently needed for supply chain risk management. The literature [6] proposed a system framework, that is, using mobile technology to plan vehicle running routes in real time to reduce logistics risks. Literature [7] believed that contextual factors such as product technology level, security requirements, the relative importance of suppliers, and the previous experience of buyers in this context should be taken into account when determining the level of risk management in the supply chain. The literature [8] showed that 44% of the vulnerabilities of enterprise supply chains will increase in the next five years. The literature [9] believed that product design is not just a creative function of an enterprise. Based on the research of a British retailer, researchers believed that the new design does help reduce supply chain risk. The literature [10] evaluated the risk management strategies under different environmental conditions using grounded theory, A framework for responding to natural disaster response needs through interviews with logistics managers has been developed. The literature [11] proposed a multi-criteria scoring model and applied it to supplier risk assessment of automobile manufacturers. The literature [12] summarized previous research and proposed global procurement risk mitigation strategies: supply chain network reconstruction, cooperation between global procurement parties, agility of the supply chain, and creation of a global procurement risk management culture. From the perspective of supply chain system composition, the literature [13] proposed to establish organizational flexibility from the five aspects of supply and procurement, conversion process, distribution channels and customer-facing activities, control systems, and corporate culture to reduce the risk of disruption. The literature [14] used the decision tree method to consider the impact of supplier reliability on business operation and procurement cost savings and determined the optimal number of suppliers under the risk of supply chain disruption.

The literature [15] proposed a framework for capturing and potentially disseminating knowledge in the supply chain. This framework can enable supply chain partners to utilize and disseminate skills and knowledge as much as possible. The study found that SMEs in any given supply chain still have differences on how knowledge is acquired, managed and disseminated. Theliterature [16] studied the application of knowledge maps in the supply chain and proposed a method to construct the knowledge map of the supply chain. Theliterature [17] analyzed how organizational conditions, technology applications, supplier relationship management, and customer relationship management in the supply chain affect knowledge creation through the SECI model. Quantitative research has found that these key factors can be beneficial to different types of knowledge conversion processes, and then achieve the success of knowledge creation in the supply chain. Theliterature [18] believes that to achieve success in supply chain management, enterprise organizations must possess and share knowledge about different aspects of the supply chain. The lack of information sharing among members of the supply chain has had a negative impact on overall profitability, Integrate knowledge management initiatives into supply chain management projects, conduct an empirical analysis of the service department of a large automobile dealership, examine the impact of manufacturer promotion effects on decision coordination in the supply chain, and use knowledge management to improve the service department’s order processing process Was discussed. Theliterature [19] examined how the factors related to trust affect the knowledge sharing among members of the green supply chain organization when cooperation and competition coexist. The study found that trust is a key factor that affects knowledge sharing between organizations, and factors that are positively related to trust are also positively related to knowledge sharing, and factors that are not related to trust will not affect knowledge sharing. Theliterature [20] believes that learning and knowledge management are the factors driving the development of the supply chain, proposes a conceptual framework for applying knowledge management to the supply chain, and verifies the effectiveness of the framework by empirical analysis with French companies. Theliterature [21] studied how supply chain management practices and knowledge management capabilities affect business performance. Through research, it was found that the implementation of supply chain management interacts with knowledge management capabilities and affects business performance. Theliterature [22] put forward a new vision to improve supply chain visualization to facilitate cooperative decision making from the perspective of knowledge management. Moreover, it provided a framework for managing the knowledge required for cooperative decision-making in the supply chain process (mainly for two types of knowledge: quantitative knowledge and qualitative knowledge). Theliterature [23] established a social network model to improve knowledge management in a multi-layer supply chain composed of small and medium-sized enterprises. This model helps academics and practitioners to have a better understanding of the knowledge management process, especially for supply chains composed of SMEs. The article [24] centers on IoT and its major part in sophisticating the human practices and endeavors. This paper moreover managed withthe collection of different information from different assets that are associated to the web. The literature [25] discusses the problem of vast volumes of big data and introduces the SmartBuddy idea of an adaptive and smart world incorporating human activity and human dynamics. The literature [26] talks about the development in parallel reconfigurable computing systems of a directed acyclic graph for video coding algorithms for motion estimation. Partitioning algorithm also plays a major role in speeding up the production of images. The article [27] deals with leveraging IoT and BigData Analytics in real-time applications using the Hadoop platform. The above-mentioned processes enable the deployment of an IoT-based Smart City. The literature [28] addresses the various problems in the field of vehicle communication with the suggestion of a mutual unified and dispersed spectrum sensing model. The application of the mutual cognitive paradigm minimizes conflict and multiple unknown problems [29, 30].

3 Model and analysis

(Q, r) stock replenishment system for single products on an unlimited level is considered. The demand process is random and fixed in λ units per unit time. The order cost K is related to each replenishment time. Moreover, the fixed cost of unit inventory per unit time is h, and the shortfall of unit cost per unit time is π. Federgruen and Zheng (1992) proved that the ideal average cost per unit of time meets the following formula: $C (Q, r) = \frac{λ K + \sum_{y = r + 1}^{Q + r} H (y)}{Q}$ (1)

Among them, H (y) represents the ratio of the fixed cost of inventory and the cost of out of stock at time t plus the extension time, and the inventory level at t is y. In other words: $H (y) = h \sum_{ξ = 0}^{y} (y - ξ) f_{D} (ξ) + π \sum_{ξ = y}^{\infty} (ξ - y) f_{D} (ξ)$ (2)

Among them, f_D (□) is the demand probability function in the replenishment period. This expression is true. At the same time, the replenishment time provided is fixed, the inventory status is evenly distributed, and the inventory status is independent of the replenishment time requirements. Meanwhile, it should be noted that the precise expression of the corresponding out-of-stock model with costs also requires some restrictions, such as: Q = 1 or r < Q.

In the model we have established, agents play a one-time role in the process of system design and application, that is, $e ⩾ \underline{e}$ . Among them, $\underline{e}$ represents the minimum action level that agents can choose. The agent’s role leads to a fixed replenishment time L, which is expressed in the model through a probability distribution. At the same time, it uses μ_L (e) as the average and $σ_{L}^{2} (e)$ as the variance to influence the agent’s decision. We assume that $μ_{L}^{'} (e), σ_{L}^{'} (e) ⩽ 0$ and $μ_{L}^{″} (e), σ_{L}^{″} (e) ⩾ 0$ , that is to say, the role of the agent can reduce the average value and the standard deviation of the distribution time of the replenishment by a small amount, which means that when the role of the agent rises, the decrease value of the replenishment time will become less and less. The agent incurs the effort cost G (e), and gets a one-time payment W (L), which is based on the replenishment time L. We assume that in the case of $e ⩾ \underline{e}$ , there are G (e) , G′ (e) and G″ (e) > 0. In other words, the agent’s role cost is positive, and it increases at a faster and faster rate according to his effort level. For simplicity of discussion, we assume $G (\underline{e}) = G^{'} (\underline{e}) = 0$ .

After choosing his effort level e, the agent will try his best to maximize the expected utility of his compensation, that is to say: $U (e | W (□)) = E [u (W (L) - G (e)) | e]$ (3)

In the formula, u (□) represents the utility function of the agent, and it is expected to take e given to L. Fixing $\underline{w}$ in a large amount of remuneration corresponds to the agent’s acceptable level of utility, which means that the agent must perform the contract until he can get $u (\underline{w})$ from it.

After L is observed and determined, the value of Qr is determined to satisfy the principle of minimizing the average inventory-related cost per unit time in long-term operation: $C (Q, r | L) = \frac{λ K + \sum_{y = r + 1}^{Q + r} H (y | L)}{Q}$ (4)

Among them, H (y|L) and f_D (·) are replaced by f_D|L (·), and it is a conditional probability function about demand when the lead time is fixed. We assume that Q (L) and r (L) represent the values of Q and r when C (Q, r|L) is the minimum value given L.

Definition: $C (L) = C (Q (L), r | L)$ (5)

As a given case of L, the average inventory-related cost per unit time is minimized in long-term operation.

The main goal of the article is to design a specific strategy for the agent to determine the inventory principles Q (L), r (L), and compensation W (L). We convert the one-time compensation into the period value i_p · W (L), where i_p is the main cost of capital. Then, the main expected total cost per unit time is: $TC (W (\cdot), e) = E [C (L) + i_{p} W (L) | e]$ (6)

Therefore, the most important optimization problem is to select W (·) and satisfy all $\tilde{e} ⩾ \underline{e}$ at the same time: $U (e | W (\cdot)) ⩾ u (\underline{w})$ (7) $U (e | W (\cdot)) ⩾ U (\underline{\tilde{e} | W (\cdot)})$ (8)

The above condition (7) of participation constraint ensures that the agent will accept the contract because its expected utility is much greater than or equal to its reserve utility. Motivation-compatibility constraints (8) enable the agent to choose the effort level according to the client’s needs. When it is desired to maximize it, the agent will choose e. Meanwhile, we note that the first optimal case is obvious, so that the motivation-compatibility constraint (8) is not necessary. However, the second best case is not obvious, so the constraints (7) and (8) are needed.

If the agent’s efforts are significant, it is the best conclusion for the agent that the principal pays a fixed amount of money. Meanwhile, this famous conclusion will be used in the following proposition.

Proposition 1. If the agent’s efforts are significant, for any given $e ⩾ \underline{e}$ , a fixed remuneration $\underline{w} + G (e)$ will minimize the total amount the principal needs to pay per time unit.

The above-mentioned adjusted remuneration is substituted into (6), there is: ${TC}^{FB} (e) = E [C (L) | e] + ip (w + G (e))$ (9)

The following proposition states that if the agent is as risk-neutral as u (ζ) = ζ and the agent’s efforts are non-significant, then a simple linear or quadratic payment scheme is optimal. If this risk-neutral agent affects the mean value of the distribution at the preparation stage, the linear payment scheme is optimal. However, if the agent’s efforts only affect the variance of the lead time, the quadratic payment scheme is optimal. Since the expected payout given by (10), (11), (12), and (13) is $\underline{w} + G (e)$ , this is equivalent to the first optimal payout. Proposition 2 gives a linear or quadratic contract to achieve the first optimal under the agent’s risk-neutral situation.

Proposition 2. If it is assumed that the agent’s efforts are not significant in a risk-neutral situation, there is:

(a) If there are $μ_{L}^{'} (\tilde{e}) < 0$ for all $\tilde{e} ⩾ \underline{e}$ , then for any given $e ⩾ \underline{e}$ , a linear payment equation derived from such as W (L) = α - βL is optimal, and the relevant parameters are: $α = \underline{w} + G (e) + \frac{G^{'} (e) + μ_{L} (e)}{- μ_{L}' (e)}$ (10) $β = \frac{G' (e)}{- μ_{L}' (e)}$ (11)

(b) If for all $\tilde{e} ⩾ \underline{e}$ , there is $μ'_{L} (\tilde{e}) = 0$ , and $σ'_{L} (\tilde{e}) < 0$ , then for any given $e ⩾ \underline{e}$ , a quadratic payment equation derived from W (L) = α - β (L - μ_L) ² is optimal, and the relevant parameters are: $α = \underline{w} + G (e) + \frac{G' (e) + μ_{L} (e)}{- 2 σ_{L}' (e)}$ (12) $β = \frac{G' (e)}{- 2 σ_{L} (e) σ_{L}' (e)}$ (13)

Proof. If the agent is an index utility function u (ζ) = - exp(- γζ) to avoid risks, and γ represents the risk to avoid. (This exponential utility function is widely used in economics, such as Holmstrom and Milgrom). We also assume that the fixed lead time L comes from a normal distribution with mean μ_L (e) and variance $σ_{L}^{2} (e)$ . Because the form of optimal payment is unknown, we focus on the linear payment model. If the linear payment model is assumed to be W (L), then it can be expressed as:

$E [u (W (L) - G (e)) | e] = u (M (e))$ (14)

When $M (e) = E [W (L) | e] - \frac{γ}{2} Var (W (L) | e) - G (e)$ (15) more information can be obtained from Lal and Srinivasan (1993). Therefore, the agent’s goal is to maximize M (e) in (15).

Correspondingly, the principal’s optimization problem is to minimize TC (W (·) , e), and the constraints are: $M (e) ⩾ \underline{w}$ (16) $M (e) ⩾ M (\tilde{e}) \tilde{e} ⩾ \underline{e}$ (17)

Proposition 3. If the agent is assumed to have an exponential utility function, and its efforts are not significant. We assume that the fixed lead time comes from a normal distribution. If the agent’s payment has W (L) = α - βL for any given $e ⩾ \underline{e}$ , the optimal linear payment parameter is: $α = \underline{w} + G (e) + \frac{γ}{2} β^{2} σ_{L}^{2} (e) + β μ_{L} (e)$ (18) $β = {\begin{matrix} \frac{\begin{matrix} - μ'_{L} (e) - \\ \sqrt{{(μ_{L}^{'} (e))}^{2} - 4 γ σ_{L}^{'} (e) G_{L}^{'} (e)} \end{matrix}}{2 γ σ_{L} (e) σ_{L}^{'} (e)} & σ_{L}^{'} (e) \neq 0 \\ \frac{G' (e)}{- μ_{L}^{'} (e)} & \begin{matrix} σ_{L}^{'} (e) = 0, \\ μ_{L}^{'} (e) \neq 0 \end{matrix} \end{matrix}$ (19)

The above formula is substituted into formula (6) to obtain:

${TC}^{SB} (e) = E [C (L) | e] + i_{p} (\underline{w} + G (e) + \frac{γ}{2} β^{2} σ_{L}^{2} (e))$ (20)

Proposition 4 In the case of the first optimal and the second optimal, an upper bound of the optimal strategy value e^* can be obtained. At the same time, this upper bound is determined by the following formula: $G (\bar{e}) = \frac{1}{i_{p}} E [C (L) | e = \underline{e}] - \underline{w}$ (21)

When $E [C (L) | e = \underline{e}] > i_{p} \underline{w}$ , otherwise, there is $e^{*} = \underline{e}$ .

Although the optimal linear payments of parameters do not have a fixed form, they can be calculated by using propositions 3 and 4. In particular, proposition 4 limits the range of possible values e^*.

4 Decentralized decision model

In the decentralized decision-making model, the manufacturer first makes decision (w, b_m), and then the seller determines its own price strategy (p, b_r) based on the optimal decision provided by the manufacturer. At this time, the expected profit of the seller is: $E (\prod_{R}) = E (D (p) (p - w) + G (b_{r}) (b_{m} - b_{r}))$ (22)

The utility function is: $\begin{matrix} U_{R} = (α - β p) (p - w) + (k + {hb}_{r}) \\ (b_{m} - b_{r}) - λ {(p - w)}^{2} σ \end{matrix}$ (23)

Equation (23) is used to obtain partial derivatives of p and b_r and make them equal to 0, we obtain: ${\begin{matrix} p^{*} = \frac{α + (β + 2 λ σ) w}{2 (β + λ σ)} \\ b_{r}^{*} = \frac{{hb}_{m} - k}{2 h} \end{matrix}$ (24)

The manufacturer’s expected profit is:

$E (\prod_{M}) = E (D (p) (w - c_{m}) + G (b_{r}) (Δ - b_{m})) = (α - β p) (w - c_{m}) + (k + h b_{r}) (Δ - b_{m})$ (25)

By substituting formula (24) into formula (25) to find the first-order partial derivatives of w and b_m, the optimal pricing strategy for manufacturers and sellers in the decentralized decision model is: ${\begin{matrix} w^{*} = \frac{1}{2} (c_{m} + \frac{α}{β}) \\ b_{m}^{*} = \frac{1}{2 h} (h Δ - k) \\ p^{*} = \frac{2 α + (β + 2 λ σ) (c_{m} + \frac{α}{β})}{4 (β + 2 λ σ)} \\ b_{r}^{*} = \frac{1}{4 h} (h Δ - 3 k) \end{matrix}$ (26)

The above formula is the equilibrium solution of the Starkberg game of manufacturers and sellers under the decentralized decision model. After bringing the content into the formula, the optimal utility of the seller and the optimal profit of the manufacturer can be obtained.

5 Model building

From the traditional supply chain risk early warning management research, we can see that the current process of supply chain risk early warning management (see Fig. 1) is still at the traditional level. Moreover, the methods of early warning are mostly from the supply chain capital flow, information flow and logistics to carry out risk early warning management modeling, and there is no in-depth study of knowledge management in supply chain risk early warning management. Today, knowledge management is gaining more and more attention from domestic and foreign companies. Using knowledge management theory to carry out early warning management of supply chain risks has practical significance. Moreover, the knowledge spiral theory applied to the supply chain risk early warning management process can solve the problems of the single repeated closure and knowledge solidification of traditional supply chain risk early warning management.

Fig. 1

Traditional supply chain risk early warning management process.

Considering the dynamic and uncertain characteristics of enterprise supply chain risk, this study uses the knowledge spiral SECI process model to design the knowledge spiral model of supply chain risk early warning management. The specific model is shown in Fig. 2. Generally speaking, supply chain risk knowledge refers to the sum of the sources, characteristics, hazards and responses of risks in the supply chain operation of the enterprise, and it mainly includes explicit knowledge and tacit knowledge of supply chain risk. In addition, the knowledge spiral theory emphasizes the transfer and transformation of knowledge in enterprises. In the supply chain enterprise risk early warning management process, its performance is as follows: by sharing, collating, combining, and internalizing the enterprise supply chain risk knowledge, Supply chain risk early-warning management personnel and senior management of enterprises can grasp the real-time status and safety requirements of the enterprise’s supply chain risks, and carry out dynamic early warning management of the new round of supply chain risks. At the same time, the ultimate goal is to improve the enterprise supply chain risk early warning management process through this dynamic model, and to realize the control of risk knowledge during the operation of the enterprise supply chain during the dynamic spiral upward process, so as to minimize the risk loss.

Fig. 2

The knowledge spiral model of supply chain risk early warning management.

Combined with the interviewer’s statement, a semantic network representation of the risk knowledge of the supply chain is made on the sources, characteristics, hazards and response of the risk knowledge. Considering the multiple relationship attributes of supply chain risk knowledge, multiple semantic networks are used to represent the network structure of supply chain risk knowledge as shown in Fig. 3. Ai the same time, the supply chain risk knowledge file is finally formed to realize the tacit knowledge of supply chain risk and provide the basis for the risk portfolio.

Fig. 3

The multiple semantic network represented by the risk knowledge of Yingchain.

The single risk knowledge flow mechanism of the supply chain risk early warning management knowledge spiral model is shown in Fig. 4. The above four functional modules are the key factors for this model to spiral upward: in the specific implementation process of supply chain risk early warning management, the enterprise can repeatedly operate the above-mentioned four contents to gradually internalize the risk early warning management knowledge in the enterprise supply chain system into the tacit knowledge of the relevant personnel in the supply chain risk early warning management. In the rounds of early warning management process, with the flow of risk knowledge as the main line, various types of supply chain risk knowledge are gradually identified, thereby reducing the adverse impact of supply chain risks on the operation of the enterprise’s supply chain.

Fig. 4

Knowledge spiral model of supply chain risk early warning management.

To study the simulation optimization problem of the inventory system, a detailed analysis of the system is needed. At the same time, combined with the actual situation, it optimizes the system and then establishes a most scientific and effective strategic measure. Generally speaking, this article proposes a multi-level inventory control model. This model is based on the procurement / transfer strategy model of the coordination center and other retailers’ inventory. The biggest difference between this model and previous research is that this model not only considers the retailer’s deployment plan, but also adjusts it according to geographic location. For example, when the retailer is closer than the supplier, it will greatly shorten the order time. Generally speaking, when the coordination center formulates the ordering strategy and the retailer is out of stock, it can start from the perspective of the coordination center, and the supplier has not reached the total order point. Therefore, the transfer of goods between retailers is considered. One is to reduce the waiting time, and also to avoid excessive accumulation of goods at some retailers. The specific inventory structure diagram is shown in Fig. 5.

Fig. 5

Inventory system model.

Five important factors, such as relevant selection time and specific growth rate of related products, are directly used as input nodes of the current network and can complete data input. In addition, the current sales volume is selected as an important output node, and the output task is completed. Usually every network model will include a hidden layer. In the formula, c is a constant, the default is 4, and m is the number of input nodes. At the same time, n is the number of output nodes. Therefore, the number of hidden nodes is 9.

In fact, the use of genetic algorithm can make the initial weight of BP neural network self-adjust, and finally obtain an optimized BP neural network. In addition, the ownership value of BP neural network is used to compile, and a most scientific fitness function will be obtained. If the network error of the genetic algorithm satisfies a certain parameter, the algorithm will stop running, and at the same time, the optimal solution obtained is transmitted to the neural network, after which certain samples can be input for training. The steps of combining genetic algorithm with BP neural network are shown in Fig. 7.

Fig. 6

BP neural network structure.

Fig. 7

Algorithm flow chart of genetic BP neural network.

For now, this article mainly summarizes N important factors that affect inventory based on current industry information and related papers. Moreover, these factors are directly used as the relevant inputs of the BP neural network, and generally include actual commodity prices, specific growth rates of substitutes, and so on. Of course, the output is usually the total ideal inventory in the current period. The demand for a certain product of an enterprise in the past two years and its related influencing factors are shown in Table 1.

Table 1

Statistics table partial data of the demand for commodities and its influencing factors

Month	Beginning inventory/kg	Commodity price/yuan	Monthly average customers/person	Substitute growth rate/%	Related Product Growth Rate/%	Purchase volume in this period/kg	Sales/kg
1	5.05	30.3	10.1	20.2	10.1	70.7	71.71
2	4.04	30.3	10.1	25.25	20.2	72.72	70.7
3	6.06	30.3	11.11	30.3	24.24	70.7	69.69
4	7.07	30.3	15.15	30.3	25.25	73.73	83.83
5	7.07	30.3	16.16	31.31	29.29	75.75	79.79
6	3.03	30.3	15.15	39.39	40.4	83.83	84.84
7	2.02	30.3	15.15	40.4	45.45	85.85	86.86
8	1.01	30.3	16.16	42.42	46.46	90.9	88.88
9	3.03	30.3	13.13	43.43	44.44	90.9	87.87
10	6.06	30.3	12.12	45.45	46.46	87.87	86.86

According to the input requirements of the BP neural network, the data in Table 1 needs to be normalized. Table 2 shows the results after normalizing the data in Table 1. The corresponding statistical graph is shown in Fig. 9.

Table 2

Statistics table of partial data of the demand for commodities and its influencing factors after standardization

Time	Commodity price	Average customer volume	Alternative growth rate	Related Product Growth	Current sales Rate
0.042	0.797	0.000	0.202	0.101	0.505
0.084	0.797	0.000	0.253	0.202	0.015
0.126	0.797	0.101	0.303	0.242	0.000
0.157	0.797	0.505	0.303	0.263	0.101
0.210	0.797	0.606	0.313	0.293	0.253
0.253	0.797	0.505	0.394	0.404	0.379
0.295	0.797	0.505	0.404	0.455	0.429
0.337	0.797	0.606	0.424	0.465	0.480
0.379	0.797	0.303	0.434	0.444	0.455
0.421	0.797	0.202	0.455	0.465	0.429

Fig. 8

Statisticsdiagram of partial dataofthe demand for commodities and its influencing factors.

Fig. 9

Statistics diagram of partial data of the demand for commodities and its influencing factors after standardization.

After building the model, this paper analyzes the performance of the system and predicts the supply chain risk coefficient. Since the verification period is a period in the past, the actual risk factor can be calculated and compared with the risk factor constructed by the system. The results are shown in Table 3 and Fig. 10.

Table 3

Statistical table of risk coefficients predicted by the system

	Actual risk factor	System prediction risk coefficient
1	0.2588	0.2233
2	0.4186	0.5154
3	0.3630	0.3724
4	0.3380	0.3593
5	0.5476	0.5997
6	0.5126	0.5258
7	0.4613	0.5248
8	0.4173	0.4586
9	0.3908	0.4230
10	0.2648	0.2432

As shown in Fig. 10 above, the predicted risk coefficient of the system constructed in this paper is relatively close to the actual risk coefficient, which shows that the model constructed in this paper has good performance and can be applied to the risk assessment of the enterprise supply chain. We can see that it is used in daily network training, that is, to link the weights in the learning network; in fact, it is parameter optimization. In this entire process, all the information of the neural network will be involved. The traditional way is to constantly find the optimal solution. After continuous adjustment, the weights can be well distributed. By using the genetic algorithm continuously, different weights can be obtained until the state stops. Combining the data obtained at this time, the optimal solution can be obtained through the BP algorithm. Moreover, the network structure is constantly optimized and perfected, and the most important in the neural network is the connection network and the transfer function. Whether the structure is properly handled has a great influence on the network. If the structure is intact, the problem can be solved very well, and there will be no redundant connections. However, no systematic structure can be used to design neural networks. Therefore, the use of genetic algorithms can solve these problems well. Although genetic algorithms do not have valid data, it can solve mathematical problems well in actual use. Therefore, the genetic algorithm can calculate a variety of more complex problems more quickly.

Fig. 10

Statistical diagram of risk coefficients predicted by the system.

6 Conclusion

Supply chain management is a new management concept and new model that has gradually received attention at home and abroad in recent years. On the basis of the interrelationships between the various links, the supply chain management controls the various links of the supply chain reasonably to achieve maximum profits at the minimum cost. However, in reality, customer needs are affected by the environment and the herd mentality, and there is no clear rule at all. Therefore, a good inventory control strategy also requires good implementation to achieve results. Therefore, it is very meaningful to design an efficient inventory management model for supply chain multi-level inventory management problems with random demand. Moreover, this article applies BP neural network to inventory control. First of all, this article analyzes the training process of the BP neural network inventory control model using sales records over a period of time as sample data. At the same time, this article also carries out a systematic inspection of the basic situation of BP neural network, and finally formulates a commodity inventory control model based on the BP neural network algorithm according to market demand. The relevant test results show that the model constructed in this paper can effectively improve work efficiency, which is very scientific.

Footnotes

Acknowledgments

The research in this paper was supported by Jiangsu Association of Productivity 2019 Open Project (No JSSCL2019B001) and Jiangsu University of Philosophy and Social Science Outstanding Innovation Cultivating Team Construction Project (No 2017ZSTD035).

References

Darabi

, Choubin

, Rahmati

, et al., Urban flood risk mapping using the GARP and QUEST models: A comparative study of machine learning techniques, Journal of Hydrology 569(3) (2019), 142–154.

Rajkomar

, Dean

and Kohane

, Machine learning in medicine, New England Journal of Medicine 380(14) (2019), 1347–1358.

Xin

, Kong

, Liu

, et al., Machine learning and deep learning methods for cybersecurity, IEEE Access 6(2) (2018), 35365–35381.

Meredig

, Agrawal

, Kirklin

, et al., Combinatorial screening for new materials in unconstrained composition space with machine learning, Physical Review B 89(9) (2014), 094104.

Ward

, Agrawal

, Choudhary

, et al., A general-purpose machine learning framework for predicting properties of inorganic materials, Npj Computational Materials 2(1) (2016), 1–7.

Feng

, Wang

, Li Liu

, et al., Incorporating machine learning with biophysical model can improve the evaluation of climate extremes impacts on wheat yield in south-eastern Australia, Agricultural and Forest Meteorology 275(5) (2019), 100–113.

Kourou

, Exarchos

T.P.

, Exarchos

K.P.

, et al., Machine learning applications in cancer prognosis and prediction, Computational and Structural Biotechnology Journal 13(2) (2015), 8–17.

Amershi

, Cakmak

, Knox

W.B.

, et al., Power to the people: The role of humans in interactive machine learning, Ai Magazine 35(4) (2014), 105–120.

Rodriguez-Galiano

, Sanchez-Castillo

, Chica-Olmo

, et al., Machine learning predictive models for mineral prospectivity: An evaluation of neural networks, random forest, regression trees and support vector machines, Ore Geology Reviews 71(11) (2015), 804–818.

10.

Coley

C.W.

, Barzilay

, Jaakkola

T.S.

, et al., Prediction of organic reaction outcomes using machine learning, ACS Central Science 3(5) (2017), 434–443.

11.

Chowdhury

, Kautz

, Yener

, et al., Image driven machine learning methods for microstructure recognition,, Computational Materials Science 123(5) (2016), 176–187.

12.

Basith

, Manavalan

, Shin

T.H.

, et al., SDM6A: A web-based integrative machine-learning framework for predicting 6mA sites in the rice genome, Molecular Therapy-Nucleic Acids 18(7) (2019), 131–141.

13.

Trost

S.G.

, Zheng

and Wong

W.K.

, Machine learning for activity recognition: hip versus wrist data, Physiological Measurement 35(11) (2014), 2183.

14.

Voyant

, Notton

, Kalogirou

, et al., Machine learning methods for solar radiation forecasting: A review, Renewable Energy 105(20) (2017), 569–582.

15.

Folberth

, Baklanov

, Balkovič

, et al., Spatio-temporal downscaling of gridded crop model yield estimates based on machine learning, Agricultural and Forest Meteorology 264(12) (2019), 1–15.

16.

Sieg

, Flachsenberg

and Rarey

, In need of bias control: evaluating chemical data for machine learning in structure-based virtual screening, Journal of Chemical Information and Modeling 59(3) (2019), 947–961.

17.

Thabtah

and Peebles

, A new machine learning model based on induction of rules for autism detection, Health Informatics Journal 26(1) (2020), 264–286.

18.

Narudin

F.A.

, Feizollah

, Anuar

N.B.

, et al., Evaluation of machine learning classifiers for mobile malware detection, Soft Computing 20(1) (2016), 343–357.

19.

Yao

, Yang

, Zhu

, et al., Core, mode, and spectrum assignment based on machine learning in space division multiplexing elastic optical networks, IEEE Access 6(1) (2018), 15898–15907.

20.

Bzdok

and Meyer-Lindenberg

, Machine learning for precision psychiatry: opportunities and challenges, Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 3(3) (2018), 223–230.

21.

Chen

, Hao

, Hwang

, et al., Disease prediction by machine learning over big data from healthcare communities, Ieee Access 5(1) (2017), 8869–8879.

22.

Itu

, Rapaka

, Passerini

, et al., A machine-learning approach for computation of fractional flow reserve from coronary computed tomography, Journal of Applied Physiology 121(1) (2016), 42–52.

23.

Jayasinghe

, Lee

G.M.

, Um

T.W.

, et al., Machine learning based trust computational model for IoT services, IEEE Transactions on Sustainable Computing 4(1) (2018), 39–52.

24.

Paul

, Daniel

, Ahmad

and Rho

, Cooperative cognitive intelligence for internet of vehicles, IEEE Systems Journal 11(3) (2017), 1249–1258.

25.

Chan

W.Y.

, Quenching for Multi-Dimensional Semilinear Parabolic Problems on a Ball with a Localized Source, Dynamic Systems and Applications 29(1) (2020), 1–10.

26.

Paul

, Ahmad

, Mazhar Rathore

and Jabbar

, Smartbuddy: defining human behaviors using big data analytics in social internet of things, IEEE Wireless Communications 23(5) (2016), 68–74.

27.

Ahmad

, Agarwal

R.P.

, Alsaedi

, Ntouyas

S.K.

and Alruwaily

, Fractional Order Coupled Systems for Mixed Fractional Derivatives with Nonlocal Multi-Point and Riemann-stieltjes Integral-Multi-Strip Conditions, Dynamic Systems and Applications 29(1) (2020), 71–86.

28.

Paul

, Jiang

Y.C.

, Wang

J.F.

and Yang

J.F.

, Parallel reconfigurable computing-based mapping algorithm for motion estimation in advanced video coding, ACM Transactions on Embedded Computing Systems (TECS) 11(S2) (2012), 1–18.

29.

Rathore

M.M.

, Paul

, Hong

W.H.

, Seo

H.C.

, Awan

and Saeed

, Exploiting IoT and big data analytics: Defining smart digital city using real-time urban data, Sustainable Cities and Society 40 (2018), 600–610.

30.

Paul

, Internet of Things: A primer’ , R Jeyaraj Human Behavior and Emerging Technologies 1(1) (2019), 37–47.