CN113793091A

CN113793091A - Credible monitoring and cold-chain storage intelligent decision-making method for cold-chain supply chain

Info

Publication number: CN113793091A
Application number: CN202110951734.8A
Authority: CN
Inventors: 张平; 唐艳艳
Original assignee: Hunan University of Technology
Current assignee: Hunan University of Technology
Priority date: 2021-08-10
Filing date: 2021-08-10
Publication date: 2021-12-14

Abstract

The cold chain is a special branch of the logistics industry, and has more rigorous requirements on links such as storage, transportation and the like on a supply chain. The safety of the medicine cold chain products represented by the vaccine is directly related to the stability of the people and the society. According to the invention, the credible monitoring method of the cold chain supply chain based on the block chain, the cloud storage and the Internet of things is constructed, so that the credible tracing of the whole life cycle of the cold chain product is realized, the product safety is ensured, and the credibility of the quality of the cold chain product is improved. The intelligent management decision of the cold-chain storage is an important way for reducing the cold-chain storage cost and improving the market competitiveness of cold-chain enterprises. The method is based on high-quality and large-scale data generated in the cold chain supervision process, and a cold chain product demand prediction scheme based on deep learning is constructed and used for assisting cold chain storage management decisions so as to reduce the storage cost of cold chain products.

Description

Credible monitoring and cold-chain storage intelligent decision-making method for cold-chain supply chain

Technical Field

The invention relates to the field of cold chain supply chains, in particular to a credible monitoring and cold chain storage inventory intelligent decision-making method for a cold chain supply chain.

Background

The cold chain is a special branch of the logistics industry, and has more rigorous requirements on links such as storage, transportation and the like on a supply chain. The safety of the medicine cold chain products represented by the vaccine is directly related to the stability of the people and the society. Pharmaceutical cold stores circulating in the pharmaceutical cold chain include prophylactic biologics, therapeutic biologics and diagnostic biologics. Representative biologicals include various vaccines, antisera (immune sera), antitoxins, toxoids, immunological agents (e.g., thymosin peptides, immunological nucleic acids, etc.), diagnostic reagents, and the like. The complete medical cold chain comprises a series of links of production, transportation, storage, use and the like. Due to the particularity of the refrigerated medical products, the medical cold chain has more severe requirements in the links of storage, transportation and the like. Cold chains require physical means to ensure proper temperature conditions along the supply chain. Special storage, handling facilities and refrigeration unit temperature control needs to be implemented.

The pharmaceutical cold chain industry is still in its early stages of development on a global scale. The drug cold chain coverage capacity is very low, most cold chain drug quality problems are related to cold chain logistics, and the problem of cold chain breakage still occurs at some time. The current medicine cold-chain logistics system is not sound. The cold-chain logistics technology and the informatization degree are not high, and the overall planning and coordination of the upstream and downstream are lacked. And each entity on the cold chain is independent from each other and maintains own service data. There is a possibility of tampering with the data, which necessarily reduces the trustworthiness of the information. In order to guarantee the life safety of people, the credibility tracing of the whole life cycle of the medicine product is necessary to be realized, and the credibility of the medicine product is improved.

The cold chain storage and transportation cost is high, and how to reduce the cost of cold chain logistics is an important research topic. Drugs and biological products on the cold chain are sensitive to the ambient temperature and require special materials and temperature control facilities. The cost of cold chain storage and transportation is significantly increased compared to the cost of a common supply chain. According to the supply chain theory, the cold chain storage management capacity is improved, and the method is an important way for reducing the cost of cold chain enterprises and avoiding product overstock. How to improve the cold-chain storage management capacity through technical means becomes practical and urgent.

Disclosure of Invention

The invention aims to solve the technical problem of providing an intelligent decision-making method for credible monitoring of a cold chain supply chain and cold chain storage inventory management aiming at the defects of the prior art.

1. A credible monitoring method of a cold chain supply chain based on a block chain, cloud storage and the Internet of things (if no special description is given, the meanings of the Internet of things and the IoT are the same) is constructed, credible tracing of the whole life cycle of a cold chain product is realized, the product safety is ensured, and the credibility of the quality of the cold chain product is improved.

2. Based on high-quality and large-scale data generated in the cold chain supervision process, a cold chain product demand prediction scheme based on deep learning is constructed and used for assisting cold chain storage inventory management decisions so as to reduce the storage cost of cold chain products.

The technical scheme of the invention is as follows:

1. a credible monitoring method of a cold chain supply chain is characterized in that the structure of a credible monitoring system adopted by the method mainly comprises a business layer, a physical layer and a data persistence layer; the business layer mainly comprises business bodies such as manufacturers, wholesalers, distributors, consumers, government regulatory agencies and the like; the service layer is used for abstracting and mapping with the real service logic; the cold chain product entity and the related information realize the physical circulation and the information circulation in the service interaction process among all the main bodies of the service layer; the entity layer mainly relates to a plurality of different entity process objects such as a production process, multi-stage cold chain storage, multi-stage cold chain transportation and the like; in order to ensure the safety of the whole life cycle of a cold chain product, an internet of things monitoring module is introduced to monitor different entity process objects; the data persistence layer relates to block chains and cloud storage; the service data of the service layer and the monitoring data of the Internet of things of the entity layer are stored in a cloud storage platform; the data are uploaded to a block chain platform at the storage address of the cloud storage platform, the hash value of the data and the like so as to enhance the credibility of the data; mainly comprises the following processes:

(1) cold chain product registration: a manufacturer submits cold-chain product registration related support data to a cloud storage platform; a manufacturer submits a cold chain product registration request to a block chain platform;

(2) cold chain supply chain business process: the cold chain supply chain process is formed by mutually connecting a series of cold chain supply chain sub-processes in series; each cold chain supply chain subprocess consists of three service entities, namely a supplier, a cold chain logistics transporter and a receiver; a supplier, a cold-chain logistics transporter and a receiver respectively initiate data storage requests to the cloud storage platform; a supplier, a cold-chain logistics transporter and a receiver respectively submit the storage address of the service data in the cloud storage platform and the hash value of the data to the block chain platform;

(3) cold chain supply chain monitoring process: the monitoring of the cold chain supply chain is implemented by the module of the Internet of things; the cold chain supply chain monitoring is mainly divided into three types of monitoring, namely production process monitoring, cold chain storage monitoring and cold chain logistics transportation monitoring; the Internet of things monitoring module is responsible for monitoring data acquisition and storing the monitoring data in the cloud storage platform; monitoring a storage address of data in the cloud storage platform and submitting a hash value of the data to a block chain platform;

(4) integration of a cold chain supply chain business process and a cold chain supply chain monitoring process: when the cold chain product is registered, establishing a mapping relation between a cold chain product number (if the product ID. has no special description, the meaning of the number is the same as that of the ID) and production process monitoring data; when the cold chain storage module carries out cold chain product storage registration, establishing a mapping relation between the cold chain product number and cold chain storage monitoring; and when the cold chain transportation module starts to carry out cold chain product loading registration, establishing a mapping relation between the cold chain product number and the cold chain logistics monitoring.

Further, the step (1) of cold chain product registration comprises the following specific steps:

(1.1) a manufacturer carries out digital signature on the cold chain product registration related service support data by using a private key of the manufacturer; then storing the data together with the digital signature into a cloud storage platform; the cloud storage platform feeds back a storage address of the service support data to a manufacturer;

(1.2) a manufacturer initiates a product registration request to the block chain platform and sends the number, the service support data storage address and the data hash value of the manufacturer; after the block chain platform checks the data, a cold chain product number is generated for the product, and the received data such as the data storage address, the data hash value and the like are written into the block chain together with the cold chain product number; and the block chain platform feeds back the serial number of the cold chain product to a manufacturer.

Further, the step (2) of the cold chain supply chain business process comprises the following specific steps:

(2.1) cold chain supply chain business process: the complete cold chain supply chain is formed by mutually connecting a series of cold chain business subprocesses in series; each cold-chain business subprocess consists of three business entities, namely a supplier, a cold-chain logistics transporter and a receiver; the supplier of the current cold chain supply chain subprocess is the receiver of the previous cold chain supply chain subprocess; the supplier of the current cold-chain supply chain subprocess is the supplier of the next cold-chain supply chain subprocess;

and (2.2) a supplier service data storage process: a supplier prepares supplier numbers and service data, and digitally signs the prepared data content by using a private key of the supplier; then, a data storage request is initiated to the cloud storage platform; after the cloud storage platform finishes the data storage process, feeding back a data storage address to a supplier; the cloud storage platform initiates a service recording request to the block chain platform, and writes information such as a supplier number, a data storage address, a data hash value and the like into the block chain platform; the block chain platform feeds back the record address on the block chain to a supplier;

and (2.3) a cold-chain logistics transport party service data storage process: the transporter prepares transporter numbers and service data and digitally signs the prepared data contents by using a private key of the transporter; then, a data storage request is initiated to the cloud storage platform; after the cloud storage platform finishes the data storage process, feeding back a data storage address to a transport party; the cloud storage platform initiates a service recording request to the block chain platform, and writes information such as a transporter number, a data storage address and a data hash value into the block chain platform; the block chain platform feeds back the record address on the block chain to the transport party;

(2.4) the storage process of the business data of the consignee: the consignee prepares the consignee number and the service data and carries out digital signature on the prepared data content by using a private key of the consignee; then, a data storage request is initiated to the cloud storage platform; after the cloud storage platform finishes the data storage process, feeding back a data storage address to a receiver; the cloud storage platform initiates a service recording request to the block chain platform, and writes information such as a cargo receiver number, a data storage address, a data hash value and the like into the block chain platform; and the block chain platform feeds back the record address on the block chain to the goods receiver.

Further, the specific steps of the cold chain supply chain monitoring process in the step (3) are as follows:

(3.1) monitoring process of a cold chain supply chain: the monitoring of the cold chain supply chain is implemented by the module of the Internet of things; the cold chain supply chain monitoring is mainly divided into three types of monitoring, namely production process monitoring, cold chain storage monitoring and cold chain logistics transportation monitoring;

(3.2) the monitoring of the Internet of things in the production process is mainly organized according to the modes of production flow and the like; after the monitoring data in the production process are regularly collected, the monitoring module of the internet of things sends a data storage request to the cloud storage platform, and sends the production flow program number, time, data and related digital signatures corresponding to the monitoring data of the internet of things to the cloud storage platform; the cloud storage platform receives the request, performs data storage operation, stores related data in the cloud storage platform, and feeds back a monitoring data storage address to the Internet of things monitoring module;

(3.3) monitoring of the Internet of things in cold-chain storage is organized mainly according to storage partition numbers or storage shelf numbers and other modes; after the internet of things monitoring module finishes the regular collection of cold-chain storage monitoring data, a data storage request is sent to the cloud storage platform, and a storage position number (storage position ID), time, data and a related digital signature corresponding to the internet of things monitoring data are sent to the cloud storage platform together; the cloud storage platform receives the request, performs data storage operation, stores related data in the cloud storage platform, and feeds back a monitoring data storage address to the Internet of things monitoring module;

(3.4) monitoring the Internet of things in cold-chain logistics transportation is mainly organized according to container numbers or vehicle numbers and other modes; after the monitoring data of the cold-chain logistics are regularly collected, the monitoring module of the internet of things sends a data storage request to the cloud storage platform, and sends the vehicle number (vehicle ID), time, data and related digital signatures corresponding to the monitoring data of the internet of things to the cloud storage platform; the cloud storage platform receives the request, performs data storage operation, stores related data in the cloud storage platform, and feeds back a monitoring data storage address to the monitoring layer of the internet of things.

Further, the step (4) of integrating the cold-chain supply chain business process and the cold-chain supply chain monitoring process specifically comprises the following steps:

(4) integration of a cold chain supply chain business process and a cold chain supply chain monitoring process:

(4.1) integrating production process monitoring data with a business process: when a manufacturer registers the cold chain product, monitoring data of the Internet of things generated in the production process of the cold chain product is used as a component of business support data; the data is submitted to a cloud storage platform; the block chain record comprises a storage address of the data in the cloud storage; when the block chain platform generates a cold chain product number for a product, establishing a mapping relation between the cold chain product number and production process monitoring data; each interest related party can obtain the block chain platform record corresponding to the record through the cold chain product number, further obtain the monitoring data in the cloud storage platform through the cloud storage address in the record, and verify the authenticity of the monitoring data through the data hash value in the record;

(4.2) integration of cold-chain storage monitoring and business process: when the cold chain storage module performs cold chain product storage registration, a specific storage position is allocated to a cold chain product in storage, so that mapping of a cold chain product number and a storage position number is established; the data belong to business support data of a medicine cold chain supply chain and need to be uploaded to a cloud storage module according to a business data storage mode; the cold chain storage module takes the storage position number and the in-storage time interval of the cold chain product as input data to request an access address of the monitoring data of the Internet of things from the monitoring module of the Internet of things; the Internet of things monitoring module feeds back a corresponding monitoring data storage address according to the information provided by the cold chain storage module; the cold chain storage module sends the cold chain product number, the Internet of things monitoring data storage address and the data hash value to the block chain platform, and requests to write the cold chain product number, the Internet of things monitoring data storage address and the data hash value into the block chain platform; after the block chain checks the data, writing the data into the block chain, and feeding back a record address on the chain to the cold chain storage module; the on-chain recorded address monitored by the Internet of things is sent to a stakeholder as a part of service data; each interest-related party can find out records according to the block chain record address so as to obtain an access address and a hash value of the monitoring data;

(4.3) integration of cold chain transportation monitoring and business process: when the cold chain transport module starts to carry out cold chain product loading registration, vehicle number information and cold chain product number information are registered in the cold chain logistics transport business record, and therefore mapping of the cold chain product number and the vehicle number is established; the data belong to business support data of a medicine cold chain supply chain and need to be uploaded to a cloud storage module according to a business data storage mode; the cold chain transportation module takes the vehicle number and the transportation time interval of the cold chain product as input data to request an access address of the monitoring data of the Internet of things from the monitoring module of the Internet of things; the Internet of things monitoring module feeds back a corresponding monitoring data storage address according to the information provided by the cold chain transportation module; the cold chain transportation module sends the cold chain product number, the Internet of things monitoring data storage address and the data hash value to the block chain platform, and requests to write the cold chain product number, the Internet of things monitoring data storage address and the data hash value into the block chain platform; after the block chain checks the data, writing the data into the block chain, and feeding back an on-chain recording address to a cold chain supply chain module; the on-chain recorded address monitored by the Internet of things is sent to a stakeholder as a part of service data; and each interest-related party can find the record according to the blockchain record address so as to obtain the access address and the hash value of the monitoring data.

2. A cold chain storage inventory management method comprises the following steps:

(1) data preprocessing: the data preprocessing module acquires storage addresses of cold chain service data and cold chain product consumption data in the cloud computing platform through the storage addresses in the block chain records; verifying the authenticity of data stored in the cloud computing platform through the hash value in the block chain;

(2) cold chain storage demand prediction: the cold chain storage demand prediction model is constructed based on a deep neural network technology and mainly comprises three modules: the system comprises a convolution operation module, a long-term and short-term memory network operation module and a graph embedding operation module; after the original data are preprocessed, the obtained data are input into a convolution operation module according to a time sequence; meanwhile, the original data is used for constructing a similar graph after being subjected to data preprocessing and is input to a graph embedding operation module; the data processed by the convolution layer is gathered through the full connection layer and input to the long-term and short-term memory network operation module; after the output of the graph embedding operation module is processed by a full connection layer, the output of the graph embedding operation module and the output of the long-term and short-term memory network operation module are integrated and spliced, and then the final output is obtained after the full connection layer processing;

(3) measurement of uncertainty of cold chain storage requirement: modeling storage demand uncertainty distribution by using a Gaussian model; based on the deep neural network prediction model, estimating the uncertainty of the cold chain storage requirement by using a dropout method, and obtaining the mean value and the variance of a Gaussian model, wherein the mean value and the variance are respectively as follows:

where T is the number of samples and τ is the model accuracy.

(4) And (3) cold-chain storage management decision: modeling the optimal cold chain warehouse inventory decision into an expected cost minimization problem; the optimal decision result is obtained as

Phi represents a cumulative distribution function of uncertain demand y, alpha is the purchasing cost of unit inventory, beta is various loss values caused by insufficient inventory of each unit, and gamma is various cost increase values caused by excessive inventory of each unit;

the cumulative distribution function F (x; mu, sigma) of the Gaussian random model is used instead of the aforementioned cumulative distribution function φ:

the height obtained in the step (3) isMean and variance of the Gaussian model are assigned to the values of μ and σ, respectively, in the cumulative distribution function as uncertainty measures²。

Has the advantages that:

according to the scheme adopted by the invention, the credible monitoring method of the cold chain supply chain based on the block chain, the cloud storage and the Internet of things is constructed, so that the credible tracing of the whole life cycle of the cold chain product is realized, the product safety is ensured, and the credibility of the quality of the cold chain product is improved. Based on high-quality and large-scale data generated in the cold chain supervision process, a cold chain product demand prediction scheme based on deep learning is constructed for assisting cold chain storage management decisions, and the storage cost of cold chain products is reduced.

Drawings

FIG. 1 System architecture diagram

FIG. 2 medical product registration

FIG. 3 Cold chain Logistics and Cold chain Logistics sub-Process

FIG. 4 trusted storage of cold-chain business data

FIG. 5 trusted storage of cold chain warehouse monitoring data

FIG. 6 trusted storage of cold chain transportation monitoring data

FIG. 7 flow of data preprocessing and inventory management

FIG. 8 Cold-chain warehouse demand prediction based on deep learning

FIG. 9 comparison of performance at different parameters: (a) different convolution sizes and different sequence lengths of the number of layers (b)

The specific implementation mode is as follows:

the specific implementation process of the invention is as follows:

1. overview of the system architecture.

Typical application scenarios of the cold chain supply chain include a medicine cold chain, an agricultural product cold chain, and the like, and for convenience of description, the description is mainly based on the medicine (e.g., vaccine) cold chain application scenario to explain the implementation process.

The trusted monitoring system architecture of the cold chain supply chain trusted monitoring method is shown in fig. 1. The system architecture mainly comprises a business layer, a physical layer and a data persistence layer. The business layer is used for abstracting and mapping with the real business logic. Representative business layer entities include primarily manufacturers, wholesalers, distributors, consumers, government regulators, and the like. Taking the vaccine cold chain scenario as an example, the main business in the pharmaceutical cold chain mainly includes manufacturers (e.g., pharmaceutical factories), distributors, terminal channels (e.g., hospitals or clinics), consumers (e.g., vaccination subjects, patients, etc.). In addition, there are government regulatory agencies responsible for the regulatory control of the safety of production and consumption of pharmaceutical products. And the cold chain product entity and the related information realize the physical circulation and the information circulation in the service interaction process among all the main bodies of the service layer.

The various entities in the supply chain are briefly described below.

Pharmaceutical factory (pharmaceutical manufacturing enterprise): enterprises engaged in pharmaceutical production activities need to obtain pharmaceutical production licenses. The manufacturer should audit and verify the overall process of drug production and drug quality as specified. Manufacturers should establish a complete production quality management system, continuously strengthen deviation management, and adopt informatization means to faithfully record all data formed in the production and inspection processes, so as to ensure that the whole production process continuously meets the legal requirements.

A distributor: the distributor is typically an intermediary who transfers the goods from the manufacturer to the next link entity. There are typically multiple levels of distributors in the market. The existence of distributors is beneficial to promoting the circulation of medical products and expanding the sales range of the products.

Hospital or clinic: hospitals or clinics provide medical diagnosis services for patients, and prescribe and sell medical products for patients according to the diagnosis results. Some types of drugs may also be sold at drug stores. The scheme of the invention can be popularized to the situations.

The consumer: the patient is the individual who ultimately purchases and uses the product. For infants in a special group, related matters are usually processed by legal guardians instead.

Government regulatory department: government regulators include two categories, one of which is the pharmaceutical product quality regulator. Mainly comprises a drug supervision and management department of a state department and an authorized product batch issuing organization. The other is the health administration department of the state department and the subordinate institutions.

The entity layer mainly relates to a plurality of different entity links (entity process objects) such as a production process, multi-stage cold chain storage, multi-stage cold chain transportation and the like. The process of distributing vaccine products from manufacturing enterprises to vaccine consumers usually goes through the transfer of multiple business entities such as multi-level wholesalers and multi-level disease prevention and control centers. At the corresponding entity level, multi-stage cold chain storage and multi-stage cold chain transportation also exist. Cold chain storage is typically used to meet the needs of vaccine storage at various levels and classes of business entities. Cold chain transportation is commonly used to meet the transportation needs of vaccine products between two adjacent business entities (e.g., between upper and lower centers for disease prevention and control). Most vaccines are very sensitive to ambient temperature and require lower temperatures throughout. In order to ensure the safety of the whole life cycle of a cold chain product, an internet of things monitoring module is introduced to monitor different entity process objects. Many vaccine safety issues are associated with cold chain scission. In order to ensure the safety of the whole life cycle of a vaccine, an internet of things monitoring module is introduced into a physical layer to monitor different physical layer objects.

The monitoring module of the internet of things is an abstract interface, and the entity objects with different entity layers construct respective object examples of the monitoring module of the internet of things according to specific requirements so as to complete corresponding monitoring tasks. The monitoring module of the Internet of things consists of a sensor layer, a communication layer and an interface layer of the Internet of things. The internet of things monitoring module collects various vaccine environments and vaccine state monitoring data through a sensor layer of the internet of things monitoring module. The sensor type and the number of the sensor layers can be configured in different types according to the needs. For example, in a cold chain transportation monitoring scenario, a GPS module is generally required to be equipped in the internet of things monitoring module to monitor the vehicle track, and in a cold chain storage monitoring scenario, the GPS module is not required to be equipped. The communication layer is responsible for communication among all modules of the Internet of things layer. The interface layer of the Internet of things module abstracts the functions of the Internet of things module and provides an interaction interface for the outside. For example, the internet of things module interacts with the cloud storage module to realize data storage.

The data persistence layer relates to block chains and cloud storage; the former ensures the credibility of data, and the latter is used for cloud storage of data. The service data of the service layer and the monitoring data of the Internet of things of the entity layer are stored in a cloud storage platform; information such as the storage address of the data in the cloud storage platform and the hash value of the data is uploaded to the block chain platform, so that the reliability of the data is enhanced. The block chain has the characteristics of decentralization, no tampering and the like, and can be used for solving the problem of data credibility in the medicine cold chain monitoring and tracing process. The data maintenance on the blockchain is too costly. The data volume of the monitoring data of the internet of things and the business support data in the medicine product circulation process is very large. And thus are not suitable for direct storage into a blockchain. For this reason, a cloud storage module is introduced to be responsible for storing data. The private key signature is carried out on the data before the data are stored, so that the data can be prevented from being forged. By recording the addresses of these data and the hash values of the data in the blockchain, the relevant data can be accessed and its integrity verified.

The system distributes public and private key pairs, block chain platform account numbers and cloud storage platform account numbers for the business layer objects, the physical layer objects and the internet of things monitoring modules associated with the business layer objects and the physical layer objects as required. When various main bodies communicate and transmit data, a public key mechanism is used for identity verification and digital signature of data so as to ensure the communication safety of the system. For convenience of description, such security-domain consensus techniques will not be further noted.

The credible monitoring method of the cold chain supply chain mainly comprises the following processes:

(4) integration of a cold chain supply chain business process and a cold chain supply chain monitoring process: establishing a mapping relation between a cold chain product number and production process monitoring data during cold chain product registration; when the cold chain storage module carries out cold chain product storage registration, establishing a mapping relation between the cold chain product number and cold chain storage monitoring; and when the cold chain transportation module starts to carry out cold chain product loading registration, establishing a mapping relation between the cold chain product number and the cold chain logistics monitoring.

2. And (5) registering a cold chain product.

A manufacturer carries out digital signature on the cold chain product registration related business support data by using a private key of the manufacturer; then storing the data together with the digital signature into a cloud storage platform; the cloud storage platform feeds back the storage address of the service support data to the manufacturer. A manufacturer initiates a product registration request to a block chain platform and sends the number, the service support data storage address and the data hash value of the manufacturer; after the block chain platform checks the data, a cold chain product number is generated for the product, and the received data such as the data storage address, the data hash value and the like are written into the block chain together with the cold chain product number; and the block chain platform feeds back the serial number of the cold chain product to a manufacturer.

As shown in fig. 2, the medical product registration is initiated by a medical manufacturing enterprise. The medical production enterprise signs the medical product registration related business support data (such as approval data of government authorities and support data of production processes) by using a private key of the medical production enterprise. The data is then stored in the cloud storage module along with the signature. And the cloud storage module stores the feedback data into an address. The private key signature can prevent others from forging data.

The method comprises the steps that a medicine production enterprise sends a medicine product registration request to a block chain platform and sends the number, the support data storage address and the data hash value of the medicine production enterprise. And after the block chain platform checks the data, generating a serial number for the medical product. And writing the received data such as the data storage address, the data hash value and the like into the block chain together with the product number. By writing the data hash value into the blockchain record, the pharmaceutical manufacturing enterprise can be prevented from tampering with the business support data.

3. A cold chain supply chain business process.

Cold chain supply chain business process: the complete cold chain supply chain is formed by mutually connecting a series of cold chain business subprocesses (also called cold chain logistics subprocesses) in series; each cold-chain business subprocess consists of three business entities, namely a supplier, a cold-chain logistics transporter and a receiver; the supplier of the current cold chain supply chain subprocess is the receiver of the previous cold chain supply chain subprocess; the supplier of the current cold-chain supply-chain subprocess is the supplier of its next cold-chain supply-chain subprocess.

The product is transferred from the producer to the consumer by a cold chain logistics process. In general, cold-chain logistics processes typically involve many different business entities. In order to achieve seamless supervision of a product cold chain circulation process, a complete cold chain supply chain is abstracted into a plurality of cold chain logistics sub-processes which are connected in series. As shown in fig. 3, the cold-chain logistics sub-process is composed of three types of business entities, namely a supplier, a cold-chain logistics transporter and a receiver. The consignee of the previous sub-process will become the supplier of the next process. All three types of business entities generate different types and amounts of business data. The process of storing business data by different business entities is basically similar. They need to provide their own numbers, service data, and sign the contents of the service data using their own private keys. The storage address of the service data and the hash value thereof are written into the block chain platform. And after the cloud storage platform finishes data storage, the storage address of the service data is fed back. The block chain platform is responsible for feeding back the recording address on the block chain.

In the cold-chain logistics process, a large number of business processes are involved, and a large number of business data are generated. If these data are stored directly in the blockchain, the cost is prohibitive. Therefore, the business data are stored in the cloud storage module. The memory address of the service data is written in the block chain, which can significantly reduce the data volume of the block chain.

The supplier service data storage process is shown in fig. 4. The supplier (the first column in fig. 4) prepares supplier numbers, service data, and digitally signs the prepared data content using its own private key; then, a data storage request is initiated to the cloud storage platform; after the cloud storage platform finishes the data storage process, feeding back a data storage address to a supplier; the cloud storage platform initiates a service recording request to the block chain platform, and writes information such as a supplier number, a data storage address, a data hash value and the like into the block chain platform; and the block chain platform feeds back the record address on the block chain to the supplier.

The cold-chain logistics transportation party business data storage process is shown in figure 4. The cold-chain logistics transportation party (the second column in fig. 4) prepares the transportation party number, the service data and digitally signs the prepared data content by using the own private key; then, a data storage request is initiated to the cloud storage platform; after the cloud storage platform finishes the data storage process, feeding back a data storage address to a transport party; the cloud storage platform initiates a service recording request to the block chain platform, and writes information such as a transporter number, a data storage address and a data hash value into the block chain platform; and the block chain platform feeds back the record address on the block chain to the transporter.

The consignee business data storage process is shown in figure 4. The consignee (the third column in fig. 4) prepares the consignee number, the service data, and digitally signs the prepared data content using its own private key; then, a data storage request is initiated to the cloud storage platform; after the cloud storage platform finishes the data storage process, feeding back a data storage address to a receiver; the cloud storage platform initiates a service recording request to the block chain platform, and writes information such as a cargo receiver number, a data storage address, a data hash value and the like into the block chain platform; and the block chain platform feeds back the record address on the block chain to the goods receiver.

4. Cold chain supply chain monitoring process

Cold chain supply chain monitoring process: the monitoring of the cold chain supply chain is implemented by the module of the Internet of things; the cold chain supply chain monitoring is mainly divided into three types of monitoring, namely production process monitoring, cold chain storage monitoring and cold chain logistics transportation monitoring. The main difference of the internet of things module in the three types of monitoring lies in the monitoring granularity and the data storage organization mode. The monitoring of the Internet of things in the production process is mainly organized according to the production flow and other modes. The monitoring of the Internet of things in cold-chain storage is mainly organized according to storage subareas or goods shelves and the like. The monitoring of the internet of things in cold-chain logistics transportation is mainly organized in the mode of containers or vehicles. The sensor modules sharing certain data with the same storage subarea or shelf are beneficial to reducing the cost. For example, taking video monitoring as an example, all products placed in the same camera monitoring area can share the video monitoring data of the camera. For example, in cold chain warehouse monitoring (column 2 of fig. 5), the data is stored by using the warehouse location number (warehouse location ID) as the key field of data index. And for monitoring cold-chain logistics transportation, the vehicle number is used as a key field of data index when data is stored.

The monitoring of the Internet of things in the production process is mainly organized according to the modes of production flow and the like; after the monitoring data in the production process are regularly collected, the monitoring module of the internet of things sends a data storage request to the cloud storage platform, and sends the production flow program number, time, data and related digital signatures corresponding to the monitoring data of the internet of things to the cloud storage platform; the cloud storage platform receives the request, performs data storage operation, stores related data in the cloud storage platform, and feeds back a monitoring data storage address to the internet of things monitoring module.

The monitoring of the Internet of things in the cold-chain storage is mainly organized according to storage partition numbers or storage shelf numbers and other modes. After the internet of things monitoring module (column 2 in fig. 5, an "IoT monitoring" module) finishes regular acquisition of cold-chain warehousing monitoring data, a data storage request is sent to the cloud storage platform, and a warehousing position number (warehousing position ID), time, data and a related digital signature corresponding to the internet of things monitoring data are sent to the cloud storage platform together; the cloud storage platform receives the request, performs data storage operation, stores related data in the cloud storage platform, and feeds back a monitoring data storage address to the internet of things monitoring module.

The monitoring of the Internet of things in cold-chain logistics transportation is mainly organized according to container numbers or vehicle numbers and other modes. After the internet of things monitoring module (column 2 in fig. 6, an "IoT monitoring" module) finishes regular acquisition of cold-chain logistics monitoring data, a data storage request is sent to the cloud storage platform, and a vehicle number, time and data corresponding to the internet of things monitoring data and a related digital signature are sent to the cloud storage platform together; the cloud storage platform receives the request, performs data storage operation, stores related data in the cloud storage platform, and feeds back a monitoring data storage address to the monitoring layer of the internet of things.

The internet of things module is responsible for storing data into the cloud storage module. When the internet of things module stores data, the data needs to be signed by using a private key so as to confirm a data source. And the cloud storage module feeds back the storage address of the data.

5. And integrating the cold-chain supply chain business process and the cold-chain supply chain monitoring process.

Integration of production process monitoring data and business process: when a manufacturer registers the cold chain product, using the monitoring data of the internet of things generated in the production process of the cold chain product as a component of service support data (for example, "registration related service data" in fig. 2); the data is submitted to a cloud storage platform; the block chain record comprises a storage address of the data in the cloud storage; when the block chain platform generates a cold chain product number for a product, establishing a mapping relation between the cold chain product number and production process monitoring data; each interest relevant party can acquire the block chain platform record corresponding to the record through the cold chain product number, further acquire the monitoring data in the cloud storage platform through the cloud storage address in the record, and verify the authenticity of the monitoring data through the data hash value in the record.

Integration of cold-chain storage monitoring and business process: when the cold chain warehousing module performs cold chain product warehousing registration (warehousing registration in fig. 5), a specific storage position is allocated to a cold chain product in the warehousing, so that a mapping between a cold chain product number and a warehousing position number is established (warehousing registration (product ID, warehousing position ID) in fig. 5); the data belong to business support data of a medicine cold chain supply chain, and are uploaded to a cloud storage module according to the business data storage mode introduced in the foregoing; the cold chain storage module takes the storage position number and the in-storage time interval of the cold chain product as input data to request an access address of the monitoring data of the Internet of things from the monitoring module of the Internet of things; the Internet of things monitoring module feeds back a corresponding monitoring data storage address according to the information provided by the cold chain storage module; the cold chain storage module sends the cold chain product number, the Internet of things monitoring data storage address and the data hash value to the block chain platform, and requests to write the cold chain product number, the Internet of things monitoring data storage address and the data hash value into the block chain platform; after the block chain checks the data, writing the data into the block chain, and feeding back a record address on the chain to the cold chain storage module; the on-chain recorded address monitored by the Internet of things is sent to a stakeholder as a part of service data; each interest-related party can find out records according to the block chain record address so as to obtain an access address and a hash value of the monitoring data; when the cold chain storage is used for the storage registration of the medical products, specific storage positions are distributed for the medical products in the storage, so that the mapping of the medical product numbers and the storage position numbers is established. The data belongs to business data of medicine cold-chain logistics and needs to be uploaded to a cloud storage module. The storage of service data has been described in the context of "cold chain supply chain service procedures".

Integration of cold chain transport monitoring and business processes: when the cold chain transportation module starts to carry out cold chain product loading registration, vehicle number information and cold chain product number information are registered in the cold chain logistics transportation business record, so that a mapping of the cold chain product number and the vehicle number is established (the transportation loading registration (product ID, vehicle ID) in the figure 6); the data belong to business support data of a medicine cold chain supply chain and need to be uploaded to a cloud storage module according to a business data storage mode; the cold chain transportation module takes the vehicle number and the transportation time interval of the cold chain product as input data to request an access address of the monitoring data of the Internet of things from the monitoring module of the Internet of things; the Internet of things monitoring module feeds back a corresponding monitoring data storage address according to the information provided by the cold chain transportation module; the cold chain transportation module sends the cold chain product number, the Internet of things monitoring data storage address and the data hash value to the block chain platform, and requests to write the cold chain product number, the Internet of things monitoring data storage address and the data hash value into the block chain platform; after the block chain checks the data, writing the data into the block chain, and feeding back an on-chain recording address to a cold chain supply chain module; the on-chain recorded address monitored by the Internet of things is sent to a stakeholder as a part of service data; and each interest-related party can find the record according to the blockchain record address so as to obtain the access address and the hash value of the monitoring data.

In the cold-chain storage, the storage position number and the storage time interval of the medical product are used as input data to request an access address of the monitoring data of the Internet of things from the Internet of things monitoring module. The internet of things monitoring module feeds back a corresponding monitoring data storage address (interaction between the 1 st column and the 2 nd column in fig. 5) according to information provided by cold chain storage. The cold chain storage sends the medical product number, the Internet of things monitoring data storage address and the data hash value to the block chain platform, and requests to write the medical product number, the Internet of things monitoring data storage address and the data hash value into the block chain platform. And after the block chain checks the data, writing the data into the block chain and feeding back the record address on the chain. And the chain record address monitored by the Internet of things is sent to the interest relevant party as part of the service data. And each interest-related party can find the record according to the blockchain record address so as to obtain the access address and the hash value of the monitoring data. The cold chain transportation process is handled in a similar manner.

6. And managing cold-chain storage inventory.

Cold chain storage management is an important link on a medicine cold chain. According to modern supply chain theory, warehousing is the core of the supply chain. Warehousing is an inventory control center in the supply chain. Warehousing is a dispatch center in logistics and supply chains. Warehousing is directly related to the efficiency and reaction rate of the supply chain.

Cold chain storage costs are very high. On the one hand, if the inventory quantity is insufficient, the cold chain product allocation or temporary urgent purchasing across the region is triggered, extra purchasing cost and time delay are brought, and the cold chain transportation cost is also increased sharply. Particularly, when the upper-level supplier also has the problem of insufficient inventory, the purchasing cost and period are greatly prolonged. The quality of service is also reduced, losing existing customers or potential customers, as consumer demands are not met in a timely manner. In severe cases, the health of the planted person can be even negatively affected. On the other hand, if the purchase is excessive, the cold chain storage operation cost is increased. In severe cases, this can even lead to cold chain product backlogs, resulting in large amounts of expired products.

Through a feasible monitoring link based on a cold chain supply chain, each entity node on the supply chain uploads various service data on the supply chain to a cloud storage. Data is signed by each entity node before being uploaded. And uploading the hash value and the cloud storage address of the data to the blockchain platform. The signature mechanism of the data prevents other nodes from forging the data, and the data hash and block chain technology prevents entity nodes from tampering the data. Therefore, the credibility of the cloud storage data is realized. And seamless butt joint of the full-flow service data is realized by cold chain monitoring based on the block chain. The method not only ensures the credible tracing of the whole life cycle of the product, but also brings convenience for the extraction and pretreatment of the training data. High quality, large volumes of data stored in cloud storage platforms and blockchains provide the potential for intelligent applications.

The data preprocessing module respectively acquires data on the cloud storage and the block chain. The data on the block chain is mainly used for data verification. The data is used as training data for training various prediction models after being cleaned and arranged.

The cold chain storage inventory management method comprises the following steps:

(1) cold chain storage demand prediction: the cold chain storage demand prediction model is constructed based on a deep neural network technology and mainly comprises three modules: the system comprises a convolution operation module, a long-term and short-term memory network operation module and a graph embedding operation module; after the original data are preprocessed, the obtained data are input into a convolution operation module according to a time sequence; meanwhile, the original data is used for constructing a similar graph after being subjected to data preprocessing and is input to a graph embedding operation module; the data processed by the convolution layer is gathered through the full connection layer and input to the long-term and short-term memory network operation module; after the output of the graph embedding operation module is processed by a full connection layer, the output of the graph embedding operation module and the output of the long-term and short-term memory network operation module are integrated and spliced, and then the final output is obtained after the full connection layer processing;

(2) measurement of uncertainty of cold chain storage requirement: modeling storage demand uncertainty distribution by using a Gaussian model; based on the deep neural network prediction model, estimating the uncertainty of the cold chain storage requirement by using a dropout method, and obtaining the mean value and the variance of a Gaussian model, wherein the mean value and the variance are respectively as follows:

(3) and (3) cold-chain storage management decision: modeling the optimal cold chain warehouse inventory decision into an expected cost minimization problem; the optimal decision result is obtained as

Phi represents a cumulative distribution function of uncertain demand y, alpha is the purchasing cost of unit inventory, beta is various loss values caused by insufficient inventory of each unit, and gamma is various cost increase values caused by excessive inventory of each unit; the cumulative distribution function F (x; mu, sigma) of the Gaussian random model is used instead of the aforementioned cumulative distribution function φ:

taking the mean value and the variance of the Gaussian model obtained in the last step as uncertainty measurement values, and respectively assigning the uncertainty measurement values to mu and sigma in the cumulative distribution function². Each link is specifically described below.

7. Data pre-processing

The data preprocessing process is shown in fig. 7. The data preprocessing module acquires storage addresses of cold chain service data and cold chain product consumption data in the cloud computing platform through the storage addresses in the block chain records; and verifying the authenticity of the data stored in the cloud computing platform through the hash value in the block chain. Dividing the whole city into a plurality of mutually disjoint rectangular areas, and integrating data in each area to obtain a data set of spatial dimensions. The nodes in the spatial dimension are the service data aggregations of the region positions. It should be noted that, since the service data are all high-dimensional data, the spatial dimension data is actually formed by stacking a plurality of spatial dimension data pieces. Next, the spatial dimension data is organized in a time series, typically with a time slice size of 1 day.

8. Cold chain warehouse demand prediction

As shown in fig. 8, the cold-chain storage demand prediction model is composed of three modules: a convolution operation module, a long-short term memory network operation module and a graph embedding operation module. And the convolution operation module is used for learning the cold chain product service data from the perspective of spatial correlation. And the long-term and short-term memory network operation module is used for learning the cold-chain product service data from the perspective of time relevance. And the graph embedding operation module is used for breaking through the constraint of spatial distance and establishing contact for nodes which are far away but have similar business modes.

A convolution operation module: and inputting the data output by the data preprocessing module into the convolution operation module according to the time sequence. And the convolution operation module extracts the spatial correlation characteristics in the data through the convolution layer. The data processed by the convolution layer is gathered through the full connection layer, and the butt joint with the long-term and short-term memory network operation module is realized. After the spatial data at different moments are subjected to feature extraction through the convolution operation module, a convolution operation feature sequence organized according to time sequence can be obtained.

A long-short term memory network operation module: the long-term and short-term memory network operation module has better long-term memory capability. The extraction of time-related features is realized through a long-term and short-term memory network operation module. In the previous step, the convolution operation module extracts the spatial correlation characteristics of the data on different time nodes. The convolutional layer characteristic sequences are sent to a long-term and short-term memory network operation module according to the time sequence for extracting time-related characteristics contained in the sequences.

A graph embedding operation module: based on the common sense, we can judge that the spatial correlation between nodes is not completely determined by the distance of the geographical position. For example, the reader can compare the following two situations: first, two areas which are far apart from each other but belong to areas with concentrated residential houses; two areas are close to each other, but one area is a residential area and the other area is an industrial area. Clearly, we have reason to believe that the demand pattern similarity between the two regions is higher in the first case. In the present invention, we will capture this distant similarity by the graph embedding algorithm. The method specifically comprises the following steps:

first, a similar diagram is constructed. All the regions are used as vertexes, and the similarity between the regions is used as weight to construct an undirected graph. A demand sequence can be obtained according to the demand of each region at each moment, and the similarity between different regions can be obtained through a dynamic time reduction algorithm.

Next, a graph embedding operation is performed. Based on the similarity graph obtained in the last step, the graph embedding operation is used for converting the similarity graph, and the method can construct a vector for each region, wherein the vector describes the spatial similarity between nodes.

And finally, the vector is transformed through a full connection layer to realize integration with the output of the long-term and short-term memory network operation module.

9. Uncertainty measurement of cold chain storage requirements

The invention uses deep learning technology to predict the cold-chain storage demand. However, the mainstream deep learning techniques do not express model uncertainty. These conventional neural networks usually use maximum likelihood estimation or maximum a posteriori to train, optimize parameters, neglect uncertainty, and thus produce a point estimation result, which does not give a probability distribution of uncertainty values. In this type of scheme, the most common method of quantifying uncertainty is to apply a softmax function to obtain probabilities, which however tends to produce large training biases. The Bayesian model has a mathematically complete framework, and model uncertainty can be deduced. But the uncertainty estimation algorithm based on the Bayesian neural network has very complex calculation.

To address this challenge, the present invention employs a dropout-based method to estimate the uncertainty of cold-chain warehouse demand. dropout is a regularization technique which is common in the field of deep learning, and is often used for solving the problem of overfitting often encountered when training a neural network. The basic idea of Dropout is to let neurons in the hidden layer stop working with a certain probability during the forward propagation. This makes the model less dependent on some local features, thereby improving the generalization ability of the model. Dropout training in a deep neural network can be viewed as an approximate Bayesian inference in a deep Gaussian process, so uncertainty can be modeled by dropout. The method can reduce the calculation complexity and does not influence the accuracy of the estimation result.

The prediction probability based on the depth gaussian model is described by the following expression:

p(y_t|x_t，X，Y)＝∫p(y_t|ω)p(ω|x_t，X，Y)dω

wherein

Is a random variable associated with the L-layer depth model. X and Y are input and output data sets, respectively.

Based on the predicted probability distribution. The mean and variance can be calculated.

Where T is the number of samples and τ is the model accuracy.

10. Cold chain inventory management decisions

By making optimal decisions on cold chain warehouse inventory, in the context of uncertain cold chain product demand, it can be ensured that the expected cost is minimized within a single ordering cycle. We model the optimal cold chain warehouse inventory decision as an expected cost minimization problem as follows:

s^*＝argminC

where C is the desired cost. The expected cost C is mainly composed of a fixed cost C₁Variable cost c₂Various losses c caused by insufficient stock resources₃Various types of cost increase due to excessive stock c₄The four components can be described by the following formula:

C＝c₁+c₂+c₃+c₄

fixed cost c₁Mainly the basic cost of maintaining the normal operation of the business, the variable cost can be expressed as c₂＝α(s-s₀) Wherein s is₀Is the inventory at the initial stage of the current decision period, s is the inventory of the current decision period, and alpha is the purchase cost of unit inventory. Various losses caused by insufficient stock resources c₃Can be expressed as c₃β (max (y-s, 0)), where β is the value of each type of loss per unit of inventory deficiency, and y is a random variable used to describe uncertain demand. The various costs resulting from an excess inventory can be expressed as c₄γ (max (s-y, 0)), where γ is the various types of cost increases per unit of excess inventory.

The cold chain warehouse inventory decision model is a classic child reporting problem in nature. It can therefore be solved by the standard first order conditions of the classical newborn problem. The optimal result is

Where phi denotes the cumulative distribution function of the uncertain demand y. According to the inventory management decision problem model and the solution, in order to realize the optimal decision of inventory management, the cold-chain storage demand and the uncertainty of the demand need to be predicted and quantitatively measured so as to obtain the cumulative probability distribution function of the cold-chain storage demand and the uncertainty of the demand. The key to solving the cold chain warehouse inventory decision problem is how to obtain the cumulative distribution function of the uncertain demand y. More specifically, under the hypothesis of a gaussian model, we need to quantitatively calculate the mean and variance of uncertain demand.

In the last step, based on the deep neural network prediction model, estimating uncertainty of cold-chain storage demand by using a dropout method, and obtaining a mean value and a variance of a gaussian model, wherein the mean value and the variance are respectively as follows:

the storage demand uncertainty distribution is modeled by using a Gaussian model. Thus, the cumulative distribution function F (x; μ, σ) of the Gaussian random model can be used instead of the aforementioned cumulative distribution function φ:

wherein the mean is and the variance is. Taking the mean value and the variance of the Gaussian model obtained in the last step as uncertainty measurement values, and respectively assigning the mean value and the variance (mu and sigma) in the cumulative distribution function²)。

11. And (4) analyzing credible tracing.

And (4) credibility of the data. First, the problem of forgery and repudiation of data. All supporting data are stored in the cloud computing platform, and private key signature is needed before data storage. This aspect prevents a malicious third party from forging the data. On the other hand, the data uploading party can be avoided, and the data is denied. Second, tampering of the data. And a malicious third party cannot tamper the data and cannot be discovered because the malicious third party does not have the private key. While once there is a drug security event, legitimate users have an incentive to tamper with the data to escape liability. However, since the hash value of the data is stored on the blockchain. The hash value will change once the data is tampered with. Thus data tampering is easily detected by the user.

The problem can be traced. Through cold chain product registration on the blockchain, the cold chain products are assigned unique numbers. All the monitoring data and the service data of the internet of things generated in the product circulation process are associated with the serial number, so that the traceability of the product is realized. Meanwhile, the cold chain business subprocess is overlapped in series before and after, so that the whole cold chain monitoring is in seamless connection.

12. Feasibility of introducing deep learning model.

Challenges for deep learning models. Deep learning has the ability to solve complex non-linear problems. There are many factors that affect cold chain storage, both demand-side and supply-side, as well as logistics-side, market share changes, and so on. The impact of these factors is highly nonlinear in nature, while deep learning enables nonlinear transformation from input to output, which is one of the important reasons why deep learning makes a breakthrough in many complex problems. Deep learning emphasizes end-to-end learning, namely: instead of going to an artificial sub-step or partition sub-problem, it is left entirely to the neural network to directly learn the mapping from the original input to the desired output. Compared with a divide-and-conquer strategy, the end-to-end learning has the advantage of synergy and has higher possibility of obtaining a better solution on the whole. However, using deep learning for demand prediction presents challenges from model complexity. According to machine learning theory, complex models require more training data. The increased number of model variables places higher demands on the amount of data used for training data. The increase of the required amount of training data caused by the complexity of deep learning cannot be the reason why the deep learning model is abandoned. The training process of machine learning utilizes models to fit the data. However, the purpose of machine learning is not to fit the training data set correctly, but to be able to predict correctly samples that have not occurred in the training set. The factors influencing cold chain storage are numerous and highly nonlinear, which in turn means that the model used should have a certain complexity, and only then can these factors be expressed. The adoption of complex models necessarily requires an increase in the amount of data used for training. Thus, the increase in data demand is caused by the complexity of the problem, not the model complexity due to deep learning.

The idea of solving the challenge is as follows: historical data accumulated in daily business processes of a node to be researched is most closely related to the node, so that the effect of directly using the data to predict the requirement of the node is the best. However, the traffic data volume of a single node is not large, and the short-time historical data accumulation cannot meet the requirement of a complex model on the training data volume. Theoretically, there is a certain direct or indirect relationship between services in different areas on the same market. For example, similar consumption patterns exist between users in different areas. At the same time, the same brand may have similar market impact in different areas. Furthermore, the link between entities downstream and upstream in the same supply chain is inherently very tight. It is therefore feasible, at least theoretically, to increase the amount of training data by exploiting the fact that data implications between different entities are correlated.

Through a credible monitoring method of a cold chain supply chain, the requirements of the deep learning training process on the quantity of data and the quality of the data are met. On the one hand, in terms of the amount of data. And a traceability system constructed based on the block chain establishes the relevance between the data of the full supply chain. The data of each node on the supply chain is not isolated any more, and the data acquisition is more convenient. We can increase the amount of data trained by introducing data from other nodes in the supply chain. This will greatly increase the effectiveness for training. On the other hand, in terms of the quality of the data. Due to the introduction of the block chain technology, the credibility problem of the data is solved, and the quality of the training data is guaranteed.

13. And (4) experimental analysis.

Experimental setup: to verify the performance of the cold chain storage demand prediction scheme, we performed experimental evaluations based on the real data. The data set used is business data of T corporation. The T corporation is primarily engaged in cold-chain logistics services related to medical care. The time span of the data set is 2016, 1/2019, 12/31/2019 for three years. The minimum time granularity of the data is daily. The benchmark schemes include ARIMA, PSO-ELM and XGboost. ARIMA is a time series prediction model that combines moving average and autoregressive. PSO-ELM is a hybrid machine learning model based on Particle Swarm Optimization (PSO) and Extreme Learning Machines (ELM). XGBoost is a lifting tree based scheme. The experimental evaluation indices included mean percent error (MAPE) and Root Mean Square Error (RMSE), which are defined below.

Where N is the total number of samples.

And y_iRespectively predicted and actual values.

The experimental result shows that MAPE of ARIMA, PSO-ELM and XGboost are 0.2531, 0.2103 and 0.1986 respectively, and MAPE of the invention is 0.1752; the RMSE of the ARIMA, PSO-ELM and XGboost schemes is 12.352, 11.212 and 10.926 respectively, and the RMSE of the invention is 9.782. Based on the comparison of performance between the proposed scheme and the reference scheme. In this experiment, the proposed solution has the best performance. ARIMA performs the worst of these approaches. ARIMA relies entirely on historical demand values, not considering non-linearity and spatial correlation. Compared with ARIMA, the performance of PSO-ELM and XGboost is obviously improved. PSO-ELM employs heuristic and population-based optimization techniques. XGBoost takes into account contextual characteristics. The performance of XGBoost is relatively better than the other two reference schemes. However, the performance of XGBoost is still worse than the solution proposed in this experiment.

Performance under different parameters: we investigated the effect of different parameter settings on the performance of the scheme by adjusting the convolution kernel size, the number of convolution layers and the LSTM sequence length. First, we investigated the effect of convolution kernel size and convolution layer number on the performance of the scheme. The results of the experiment are shown in FIG. 9 a. Experimental results show that these two parameters have a significant impact on the performance of the scheme. In this experiment, the scheme achieved the best performance when the convolution kernel size was 7 times 7 and the number of convolution layers was 3. We then investigated the effect of LSTM sequence length on pattern performance. The results of the experiment are shown in FIG. 9 b. The experimental results show that the length of the LSTM sequence also has a significant effect on the performance of the protocol. In this experiment, our method achieved the best performance when the length was 5 days. If the "length" value continues to increase, the performance variation is relatively small.

Performance under different structures: we compare the proposed scheme with the scheme without the embedded graph module. For the scheme without the embedded graph module, the MAPE and RMSE indexes are 0.1835 and 10.093 respectively; for the scheme with the embedded graph module, the two indexes of MAPE and RMSE are 0.1752 and 9.782 respectively. Without an embedded module, performance is significantly degraded, although it can still extract features from the temporal and spatial dimensions. We believe that spatial correlation exists not only between adjacent regions, but also between remote regions. The similarity between the regions extracted by the graph embedding module exceeds the limit of the regions. It can effectively explore such correlations in remote areas, thus having a significant impact on the performance of the proposed solution.

Claims

1. A credible monitoring method of a cold chain supply chain is characterized in that the structure of a credible monitoring system adopted by the method mainly comprises a business layer, a physical layer and a data persistence layer; the business layer mainly comprises business bodies such as manufacturers, wholesalers, distributors, consumers, government regulatory agencies and the like; the service layer is used for abstracting and mapping with the real service logic; the cold chain product entity and the related information realize the physical circulation and the information circulation in the service interaction process among all the main bodies of the service layer; the entity layer mainly relates to a plurality of different entity process objects such as a production process, multi-stage cold chain storage, multi-stage cold chain transportation and the like; in order to ensure the safety of the whole life cycle of a cold chain product, an internet of things monitoring module is introduced to monitor different entity process objects; the data persistence layer relates to block chains and cloud storage; the service data of the service layer and the monitoring data of the Internet of things of the entity layer are stored in a cloud storage platform; the data are uploaded to a block chain platform at the storage address of the cloud storage platform, the hash value of the data and the like so as to enhance the credibility of the data; the method is also characterized by mainly comprising the following steps:

2. The method for credibly monitoring the supply chain of the cold chain according to claim 1, wherein the step (1) comprises the following specific steps:

(1) cold chain product registration:

3. The method for credibly monitoring a cold chain supply chain according to claim 1, wherein the step (2) comprises the following specific steps:

4. The cold chain supply chain credible monitoring method of claim 1, wherein the step (3) comprises the following specific steps:

(3.3) monitoring of the Internet of things in cold-chain storage is organized mainly according to storage partition numbers or storage shelf numbers and other modes; after the internet of things monitoring module finishes the regular collection of cold-chain storage monitoring data, a data storage request is sent to the cloud storage platform, and the storage position number, the time, the data and the related digital signature corresponding to the internet of things monitoring data are sent to the cloud storage platform together; the cloud storage platform receives the request, performs data storage operation, stores related data in the cloud storage platform, and feeds back a monitoring data storage address to the Internet of things monitoring module;

(3.4) monitoring the Internet of things in cold-chain logistics transportation is mainly organized according to container numbers or vehicle numbers and other modes; after the monitoring module of the internet of things finishes the regular collection of the cold-chain logistics monitoring data, a data storage request is sent to the cloud storage platform, and the vehicle number, the time, the data and the related digital signature corresponding to the monitoring data of the internet of things are sent to the cloud storage platform together; the cloud storage platform receives the request, performs data storage operation, stores related data in the cloud storage platform, and feeds back a monitoring data storage address to the monitoring layer of the internet of things.

5. The method for credibly monitoring a cold chain supply chain according to claim 1, wherein the step (4) comprises the following specific steps:

6. A cold chain storage inventory management method is characterized by comprising the following steps:

where T is the number of samples and τ is the model accuracy.

taking the mean value and the variance of the Gaussian model obtained in the step (3) as uncertainty measurement values, and respectively assigning the uncertainty measurement values to mu and sigma in the cumulative distribution function²。