Medical device digital supply chain batch intelligent management method and device

Abstract

The embodiments of the present application provide a method and apparatus for intelligent batch management of a digital supply chain for medical devices. The method includes: combining blockchain technology to ensure data security, achieving accurate demand forecasting through deep learning, and possessing intelligent inventory optimization capabilities. At the same time, the system can also quickly respond to market changes and realize the automation and intelligence of replenishment decisions. The present application effectively improves the overall operational efficiency of the medical device supply chain.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and specifically to a method and device for intelligent batch management of a digital supply chain for medical devices. Background Art

[0002] Batch management of medical devices faces numerous challenges in the current healthcare supply chain. Traditional management methods rely primarily on manual experience and fixed rules, often leading to inventory backlogs or shortages when faced with complex and changing medical needs. Although some institutions have begun to adopt information systems for management, significant deficiencies remain in areas such as data security, forecasting accuracy, and inventory optimization.

[0003] Current batch management systems generally suffer from low data reliability and single-source forecasting models. Most systems lack effective data tamper-proofing mechanisms, making it difficult to ensure information authenticity and traceability. Furthermore, existing demand forecasting methods are often overly simplistic, failing to fully consider time series characteristics and the integration of multi-source data, resulting in significant deviations between forecast results and actual demand.

[0004] Of particular note is the weakness of existing systems in handling medical device returns and optimizing inventory. Most systems are unable to accurately identify and categorize the status of returned devices and lack intelligent replenishment decision-making mechanisms. This leads to inefficient inventory management and high operating costs. Furthermore, the systems' adaptability and responsiveness to sudden demand or seasonal fluctuations are insufficient.

[0005] Therefore, the medical industry urgently needs a smarter and safer batch management solution. Summary of the Invention

[0006] In response to the problems in the existing technology, the present application provides a method and device for batch intelligent management of a digital supply chain of medical devices, which can effectively improve the overall operational efficiency of the medical device supply chain.

[0007] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application provides a method for intelligent batch management of a digital supply chain of medical devices, comprising:

[0009] Construct a unique box set ID based on the classification code of the medical device, encrypt the box set ID using the elliptic curve digital signature algorithm, perform a hash operation on the digital signature and medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the box set ID and the tamper-proof blockchain database;

[0010] Writing the box set ID and the medical device information into a multi-level index structure, verifying data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, constructing a time series sample on a daily basis for the verified historical consumption data, performing sliding window segmentation on the time series sample and extracting time series features, inputting the time series feature data into the input layer of a long short-term memory neural network model, performing maximum and minimum value normalization on the time series feature data through a forget gate and an input gate, training the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determining the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model;

[0011] The medical device information is bound to the box set ID in a batch matching manner, and feature extraction and image classification are performed on the returned medical devices based on the residual deep convolutional neural network. The image classification result is fused with the output value of the long short-term memory neural network model using evidence theory data to obtain the time series correlation coefficient and root mean square error value of the fused data. A dynamic weight allocation matrix based on a genetic algorithm is constructed according to the time series correlation coefficient and the root mean square error value, and the weight parameters of the long short-term memory neural network model are updated online according to the dynamic weight allocation matrix. The updated weight parameters and their corresponding distribution strategies and replenishment opportunities are recorded in the tamper-proof blockchain database. Based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices, the triggering probability of the replenishment opportunity is calculated. When the triggering probability exceeds the preset threshold, a replenishment list is automatically generated.

[0012] Furthermore, the method constructs a unique box set ID based on the classification code of the medical device, encrypts the box set ID using the elliptic curve digital signature algorithm, performs a hash operation on the digital signature and the medical device information using a symmetric key to generate an information digest, stores the information digest in a tamper-proof blockchain database, and establishes a mapping relationship between the box set ID and the tamper-proof blockchain database, including:

[0013] Taking the box set ID as input, a binary data block is obtained, the binary data block is mapped to a point on an elliptic curve, a public key pair is generated based on a base point P on the elliptic curve and a random number k, a signature value is calculated using the public key pair, and an encrypted digital signature is output based on the binary data block and the signature value;

[0014] The production date, expiration date, product batch number, and sterilization batch number of the medical device are read as medical device information, the medical device information and the digital signature are merged into a string to be processed, the hash value of the string to be processed is iteratively calculated through the SHA-256 hash function, the hash value and timestamp are organized into a block information summary using a Merkle tree structure, the block information summary is encrypted based on an asymmetric encryption algorithm, and then written into a tamper-proof blockchain database.

[0015] Furthermore, writing the box set ID and the medical device information into a multi-level index structure, and verifying data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, includes:

[0016] A three-layer index structure is constructed. The top-layer index records the storage location pointer of the box set ID, the middle-layer index stores the category and specification information of the medical device, and the bottom-layer index records the consumption history of the medical device according to the timestamp. The association relationship between the indexes at each layer is organized through a B+ tree structure. The data validity flag is set in the bottom-layer index, and the medical device information is compressed and stored using bitmap indexing technology.

[0017] Read the historical consumption data with timestamp from the underlying index, obtain the block information digest of the corresponding timestamp from the tamper-proof blockchain database, decrypt the block information digest using the public key of the elliptic curve digital signature algorithm, perform hash calculation on the decrypted information digest and the historical consumption data, and compare the calculation result with the hash value stored in the block information digest to see if they are consistent.

[0018] Furthermore, the verified historical consumption data is used to construct a time series sample in units of days, the time series sample is segmented into sliding windows, and time series features are extracted, including:

[0019] The historical consumption data is aggregated daily by time stamp to calculate the average daily consumption. The average daily consumption is normalized by standard deviation to eliminate the dimensionality effect. Missing time point data is supplemented by linear interpolation. High-frequency noise in the average daily consumption is removed by a low-pass filter to construct a multidimensional time series sample containing consumption, time periodicity, and seasonality characteristics.

[0020] The length of the sliding window is set to 14 days and the step length is 1 day. The statistical features of the data in each window, including the mean, variance, kurtosis, and skewness, are calculated. The frequency domain features of the window data are extracted using wavelet transform, and the periodic features of the window data are obtained through autocorrelation analysis. The statistical features, frequency domain features, and periodic features are combined to form a feature vector to construct a time series feature dataset.

[0021] Furthermore, the time series feature data is input into the input layer of the long short-term memory neural network model, the time series feature data is normalized to the maximum and minimum values ​​through a forget gate and an input gate, the hidden layer parameters of the long short-term memory neural network model are trained using an adaptive moment estimation optimizer, and the future average daily consumption and fluctuation range of medical devices are determined based on the trained long short-term memory neural network model, including:

[0022] The time series feature data is divided into an 80% training set and a 20% validation set. The maximum and minimum value intervals are calculated for each feature dimension. The MinMax normalization method is used to map the feature values ​​to the [0, 1] interval. A sigmoid activation function is set in the forget gate of the long short-term memory neural network to perform a nonlinear transformation on the feature data. The weight distribution of the feature data is adjusted through the tanh activation function of the input gate. The dropout mechanism is used to randomly inactivate some neurons to prevent overfitting.

[0023] A bidirectional long short-term memory neural network structure is constructed, which includes two hidden layers. The number of neurons in each layer is set to 128. The Xavier initialization method is used to initialize the hidden layer parameters. The Adam optimizer is used to iteratively update the network parameters. The loss function is constructed based on the root mean square error, and the learning rate is set to 0.001. When the loss value of the validation set does not decrease for five consecutive epochs, the early stopping mechanism is activated. The trained model parameters are saved and used to predict the consumption range for the next 30 days.

[0024] Furthermore, the medical device information is bound to the box set ID in a batch matching manner, feature extraction and image classification are performed on the returned medical devices based on a residual deep convolutional neural network, and the image classification results are fused with the output values ​​of the long short-term memory neural network model through evidence theory data to obtain the time series correlation coefficient and root mean square error value of the fused data, including:

[0025] A 50-layer residual deep convolutional neural network structure was built, with a 3×3 convolution kernel set to extract local features of medical device images. Batch normalization layers and ReLU activation functions were used to enhance feature expression. Short-circuit connections were added to the residual blocks to alleviate the vanishing gradient problem. A global average pooling layer was used to reduce feature dimensionality. The fully connected layer output the probability distribution of the medical device's integrity classification, which was then output as the image classification result.

[0026] Based on the DS evidence theory, a basic probability distribution function of the two evidence sources is constructed. The support function and likelihood function are calculated to represent the credibility of the image classification result and the predicted output value, respectively. The Dempster combination rule is used to perform orthogonal sum operations on the two evidence sources. The Pearson correlation coefficient and root mean square error of the fused data and the original data are calculated to evaluate the fusion effect.

[0027] Furthermore, a dynamic weight allocation matrix based on a genetic algorithm is constructed according to the time series correlation coefficient and the root mean square error value, and the weight parameters of the long short-term memory neural network model are updated online according to the dynamic weight allocation matrix. The updated weight parameters and their corresponding distribution strategies and replenishment timing are recorded in the tamper-proof blockchain database. Based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices, the trigger probability of the replenishment timing is calculated. When the trigger probability exceeds a preset threshold, a replenishment list is automatically generated, including:

[0028] The time series correlation coefficient and root mean square error value are used as input variables of the fitness function. The population individuals are initialized using real number encoding, with the crossover probability set to 0.8 and the mutation probability set to 0.1. A tournament selection strategy is used to select high-quality individuals. A new generation of population is generated through arithmetic crossover operator and Gaussian mutation operator. The weight allocation scheme is iteratively optimized until the maximum number of iterations is reached. The optimal individual is decoded into a dynamic weight allocation matrix for updating the weight parameters of the long short-term memory neural network model.

[0029] A replenishment decision model is constructed based on a Bayesian network. The current inventory level, predicted consumption, safety stock threshold and historical replenishment records of medical devices are used as input nodes of conditional probability. The maximum likelihood estimation method is used to learn the network parameters. The posterior probability distribution of replenishment timing is calculated through the variable elimination algorithm. The posterior probability distribution is normalized to obtain the trigger probability. When the trigger probability is greater than 0.8, the replenishment demand is written into the tamper-proof blockchain database.

[0030] In a second aspect, the present application provides a medical device digital supply chain batch intelligent management device, comprising:

[0031] An information encryption module is used to construct a unique box set ID based on the classification code of the medical device, encrypt the box set ID using an elliptic curve digital signature algorithm, perform a hash operation on the digital signature and medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the box set ID and the tamper-proof blockchain database;

[0032] a consumption prediction module, configured to write the box set ID and the medical device information into a multi-level index structure, verify data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, construct a time series sample on a daily basis for the verified historical consumption data, perform sliding window segmentation on the time series sample and extract time series features, input the time series feature data into the input layer of a long short-term memory neural network model, perform maximum and minimum value normalization on the time series feature data through a forget gate and an input gate, train the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determine the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model;

[0033] A replenishment monitoring module is used to bind the medical device information with the box set ID in a batch matching manner, perform feature extraction and image classification on the returned medical devices based on the residual deep convolutional neural network, perform evidence theory data fusion on the image classification results and the output value of the long short-term memory neural network model to obtain the time series correlation coefficient and root mean square error value of the fused data, construct a dynamic weight allocation matrix based on a genetic algorithm according to the time series correlation coefficient and the root mean square error value, and update the weight parameters of the long short-term memory neural network model online according to the dynamic weight allocation matrix, record the updated weight parameters and their corresponding distribution strategies and replenishment timing in the tamper-proof blockchain database, calculate the trigger probability of the replenishment timing based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices, and automatically generate a replenishment list when the trigger probability exceeds the preset threshold.

[0034] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for batch intelligent management of a digital supply chain of medical devices are implemented.

[0035] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for batch intelligent management of a digital supply chain of medical devices.

[0036] In a fifth aspect, the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the medical device digital supply chain batch intelligent management method.

[0037] As can be seen from the above technical solution, this application provides a method and device for intelligent batch management of a digital supply chain for medical devices. This system ensures data security by integrating blockchain technology, achieves accurate demand forecasting through deep learning, and possesses intelligent inventory optimization capabilities. Furthermore, the system can quickly respond to market changes, automating and intelligentizing replenishment decisions, thereby effectively improving the overall operational efficiency of the medical device supply chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0039] Figure 1 This is one of the flow charts of the method for batch intelligent management of a digital supply chain of medical devices in an embodiment of the present application;

[0040] Figure 2 This is the second flow chart of the method for batch intelligent management of a digital supply chain of medical devices in an embodiment of the present application;

[0041] Figure 3 This is the third flow chart of the method for batch intelligent management of a digital supply chain of medical devices in an embodiment of the present application;

[0042] Figure 4 This is the fourth flow chart of the method for batch intelligent management of a digital supply chain of medical devices in an embodiment of the present application;

[0043] Figure 5 This is the fifth flow chart of the method for batch intelligent management of a digital supply chain of medical devices in an embodiment of the present application;

[0044] Figure 6 This is the sixth flow chart of the method for batch intelligent management of a digital supply chain of medical devices in an embodiment of the present application;

[0045] Figure 7 This is the seventh flow chart of the method for batch intelligent management of a digital supply chain of medical devices in an embodiment of the present application;

[0046] Figure 8 This is a structural diagram of a medical device digital supply chain batch intelligent management device in an embodiment of the present application;

[0047] Figure 9 Schematic diagram of the structure of the electronic device in the embodiment of the present application.

[0048] Reference numerals:

[0049] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver program storage unit 9144, antenna 9111, speaker 9131, microphone 9132. DETAILED DESCRIPTION

[0050] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0051] The acquisition, storage, use, and processing of data in this application's technical solution comply with relevant national laws and regulations.

[0052] In light of the existing challenges, this application proposes a method and device for intelligent batch management of a digital supply chain for medical devices. This system utilizes blockchain technology to ensure data security, uses deep learning to achieve accurate demand forecasting, and incorporates intelligent inventory optimization capabilities. Furthermore, the system can rapidly respond to market changes, automating and intelligentizing replenishment decisions, thereby effectively improving the overall operational efficiency of the medical device supply chain.

[0053] In order to effectively improve the overall operational efficiency of the medical device supply chain, this application provides an embodiment of a medical device digital supply chain batch intelligent management method, see Figure 1 The medical device digital supply chain batch intelligent management method specifically includes the following contents:

[0054] Step S101: Construct a unique box set ID based on the classification code of the medical device, encrypt the box set ID using the elliptic curve digital signature algorithm, perform a hash operation on the digital signature and medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the box set ID and the tamper-proof blockchain database;

[0055] Optionally, in this embodiment, this step implements a unique identifier generation and tamper-resistant storage mechanism for medical device case sets. First, the system constructs a case set ID based on the medical device classification code. Using a hierarchical encoding method, it encodes information such as device category, specification, model, and production batch into a fixed-length string, ensuring uniqueness and traceability throughout the supply chain.

[0056] During the encryption process, the system uses the Elliptic Curve Digital Signature Algorithm (ECDSA) for security. This algorithm, based on the elliptic curve discrete logarithm problem, offers shorter key lengths and higher computational efficiency compared to the traditional RSA algorithm while maintaining the same security strength. The system first generates a public and private key pair: the private key is used to generate the digital signature, and the public key is used to verify the signature's validity.

[0057] Symmetric keys are generated using a high-strength key derivation function, using a random number seed and a key expansion algorithm to generate key material of sufficient length. The system uses this symmetric key to encrypt digital signatures and critical medical device information, then calculates the message digest using the SHA-256 hash algorithm. The hashing process is one-way and collision-resistant, effectively preventing information tampering.

[0058] Message summaries are stored using distributed ledger technology, writing data to a tamper-proof blockchain database. Each block contains multiple transaction records, each of which includes fields such as a timestamp, the hash of the previous block, and a message summary. Blocks are linked via cryptographic hashes to form an immutable chain structure. The system uses a consensus mechanism to ensure data consistency and smart contracts to automatically perform data verification and updates.

[0059] To improve data retrieval efficiency, the system establishes a mapping between box set IDs and the blockchain database. This mapping table uses a distributed hash table structure, enabling rapid location and access to specific records. The system uses a consistent hashing algorithm to achieve uniform data distribution and a replication mechanism to ensure high data availability.

[0060] This technical solution effectively addresses the unique identification and secure storage of medical device identifiers through the integration of encryption algorithms and blockchain technology. The system ensures data immutability and traceability, providing reliable technical support for the full lifecycle management of medical devices. Furthermore, the efficient computational capabilities of elliptic curve cryptography ensure the system's performance in large-scale applications.

[0061] Furthermore, the solution's distributed nature and consensus mechanism effectively avoid single-point-of-failure risks and enhance system reliability. The hash chain-based storage structure and smart contract-based automated execution reduce human intervention and enhance the automation of data management. The establishment of mapping relationships optimizes data retrieval efficiency and provides a convenient data access interface for upper-level applications.

[0062] Step S102: writing the box set ID and the medical device information into a multi-level index structure, verifying data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, constructing a time series sample on a daily basis for the verified historical consumption data, performing sliding window segmentation on the time series sample and extracting time series features, inputting the time series feature data into the input layer of a long short-term memory neural network model, performing maximum and minimum value normalization processing on the time series feature data through a forget gate and an input gate, training the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determining the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model;

[0063] Optionally, in this embodiment, this step implements multi-level index storage and an LSTM-based time series prediction mechanism for medical device consumption data. The system first constructs a multi-level index structure, using a B+ tree as the underlying data structure to organize case set IDs and medical device information into a hierarchical index system. Each index node contains a key-value pair and a pointer, supporting efficient range queries and data updates.

[0064] During data consistency verification, the system extracts historical consumption data from a multi-level index structure and simultaneously obtains a summary of the corresponding time period from the blockchain database. By comparing the hash values ​​of the two, it ensures that the data has not been tampered with. Verified data is aggregated by day to construct a time series sample containing consumption, timestamps, and other relevant features.

[0065] Time series feature extraction utilizes a sliding window technique, with the window size dynamically adjusted based on the periodicity of the data. Within each window, the system calculates statistical features (mean, variance, kurtosis, etc.) and frequency domain features (Fourier transform coefficients), and captures periodic patterns in the data through autocorrelation analysis. These features collectively form a feature vector describing the patterns of medical device consumption.

[0066] The input layer of the LSTM neural network model receives processed time series feature data. Using the forget gate mechanism, the model adaptively determines the extent to which historical information is retained or discarded. The input gate controls the rate at which new information enters the memory cell. Feature data undergoes minimum and maximum value normalization, mapping the value range to the [0, 1] interval to eliminate dimensionality effects and improve model training stability.

[0067] The model training process uses the Adaptive Moment Estimation (AMSGrad) optimizer, a modified version of the Adam algorithm. This optimizer adaptively adjusts the learning rate of each parameter by recording the exponential moving average of historical gradients. This mechanism effectively prevents drastic fluctuations in the learning rate and improves the stability of model convergence. Hidden layer parameters are updated using the backpropagation algorithm, optimizing network weights by minimizing the mean squared error between predicted and actual values.

[0068] The trained LSTM model captures both the long-term dependencies and short-term fluctuations in medical device consumption. By performing probability distribution analysis on the model output, the system can predict average daily consumption over a period of time and estimate its fluctuation range. This provides reliable data support for subsequent inventory management and replenishment decisions.

[0069] This technical solution addresses the challenges of efficiently storing and accurately predicting medical device consumption data. A multi-level indexing structure provides fast data access, blockchain verification ensures data credibility, and an LSTM-based deep learning model accurately models complex time series patterns. The system boasts excellent scalability and predictive accuracy, providing strong support for intelligent management of medical devices.

[0070] Step S103: The medical device information is bound to the box set ID in a batch matching manner, feature extraction and image classification are performed on the returned medical devices based on the residual deep convolutional neural network, and the image classification result is fused with the output value of the long short-term memory neural network model using evidence theory data to obtain the time series correlation coefficient and root mean square error value of the fused data, and a dynamic weight allocation matrix based on a genetic algorithm is constructed according to the time series correlation coefficient and the root mean square error value, and the weight parameters of the long short-term memory neural network model are updated online according to the dynamic weight allocation matrix, and the updated weight parameters and their corresponding distribution strategies and replenishment timing are recorded in the tamper-proof blockchain database, and the trigger probability of the replenishment timing is calculated based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices. When the trigger probability exceeds the preset threshold, a replenishment list is automatically generated.

[0071] Optionally, in this embodiment, this step implements the complete process of intelligent medical device allocation and replenishment management. First, the system uses batch matching technology to bind medical device information to case set IDs, establishing a unique mapping relationship to ensure device traceability. The matching process utilizes a parallel processing mechanism to improve the efficiency of large-scale data processing.

[0072] During the processing of returned devices, the system uses a residual deep convolutional neural network for image analysis. The network architecture comprises multiple residual blocks, each consisting of a convolutional layer, a batch normalization layer, and an activation function. The introduction of residual connections effectively alleviates the vanishing gradient problem in deep networks and improves feature extraction accuracy. The convolution operation uses kernels of varying scales to capture multi-scale features of device images, enabling accurate classification of device integrity and usage status.

[0073] The evidence-based data fusion process combines image classification results with the predictions of the LSTM model. The system constructs a basic probability distribution function based on the Dempster-Shafer theory and calculates the evidence conflict degree and combination rules to obtain the fused decision result. This fusion mechanism fully utilizes the complementarity of multi-source data to improve the reliability of predictions.

[0074] The system constructs a dynamic weight allocation matrix based on the time series correlation coefficient and root mean square error calculated from the fused data as evaluation metrics. A genetic algorithm is used to optimize the weight allocation scheme, continuously evolving optimal weight combinations through operations such as encoding, crossover, and mutation. The fitness function comprehensively considers prediction accuracy and model complexity, ensuring performance while mitigating overfitting risks.

[0075] The LSTM model's online updates utilize incremental learning, adjusting network parameters based on a dynamic weight matrix. A sliding window mechanism is used during the update process to maintain a balance between historical and new data, ensuring model adaptability and stability. The updated weight parameters and corresponding distribution strategies are recorded in the blockchain database, ensuring data immutability.

[0076] Replenishment timing is determined based on a comprehensive evaluation of multiple indicators. The system calculates safety stock thresholds for each medical device type and, based on predicted consumption trends and fluctuations, constructs a probabilistic model to estimate the probability of triggering a replenishment. When the trigger probability exceeds the preset threshold, the system automatically generates a replenishment list containing device type, quantity, and priority.

[0077] This technical solution achieves intelligent and automated medical device replenishment management through the organic combination of deep learning, evidence theory, and genetic algorithms. The system boasts strong environmental adaptability and reliable decision-making, effectively addressing the subjectivity and lag issues inherent in traditional replenishment management. Furthermore, blockchain-based data storage ensures traceability and transparency in the decision-making process, providing an innovative technical solution for medical device supply chain management.

[0078] The implementation of this solution significantly improved the accuracy of medical device inventory management, reduced the risk of inventory overstocking and shortages, and optimized supply chain operational efficiency. The system's adaptive nature enables it to continuously optimize decision-making models, providing strong support for refined management in medical institutions.

[0079] As can be seen from the foregoing description, the intelligent batch management method for a digital supply chain of medical devices provided in this embodiment of the application can ensure data security by integrating blockchain technology, achieve accurate demand forecasting through deep learning, and possess intelligent inventory optimization capabilities. Furthermore, the system can rapidly respond to market changes, automating and intelligentizing replenishment decisions, thereby effectively improving the overall operational efficiency of the medical device supply chain.

[0080] In one embodiment of the medical device digital supply chain batch intelligent management method of the present application, see Figure 2 , and can also include the following:

[0081] Step S201: Obtain a binary data block using the box set ID as input, map the binary data block to a point on an elliptic curve, generate a public key pair based on a base point P on the elliptic curve and a random number k, calculate a signature value using the public key pair, and output an encrypted digital signature based on the binary data block and the signature value;

[0082] Step S202: Read the production date, expiration date, product batch number, and sterilization batch number of the medical device as medical device information, merge the medical device information and the digital signature into a string to be processed, iteratively calculate the hash value of the string to be processed through the SHA-256 hash function, organize the hash value and timestamp into a block information summary using a Merkle tree structure, encrypt the block information summary based on an asymmetric encryption algorithm, and then write it into a tamper-proof blockchain database.

[0083] Optionally, in this embodiment, these two steps implement the digital signature generation and blockchain storage mechanism for the medical device case set. In step S201, the system first converts the case set ID into a binary data block and uses a point compression algorithm to map this data to points on an elliptic curve. The elliptic curve uses the NIST-recommended P-256 curve, which offers excellent security and computational efficiency. The base point P is a generator on the curve and has a large prime order, ensuring the difficulty of solving the discrete logarithm problem.

[0084] The random number k is generated using a cryptographically secure random number generator, which collects system noise as a random source through an entropy pool. Based on the random number k and the base point P, the system calculates a temporary public key R = kP and simultaneously generates a private key d and the corresponding public key Q = dP. The signature value is calculated using a modified ECDSA algorithm consisting of two components (r, s), where r is the x-coordinate of the temporary public key R, and s is calculated using the message hash value, the random number k, and the private key d.

[0085] In step S202, the system processes the medical device's key attribute information. This information is standardized and concatenated with the digital signature to form a string to be processed. The SHA-256 hash function iteratively calculates this string, leveraging its avalanche effect to ensure that the output hash value is highly sensitive to even the slightest changes in the input.

[0086] The Merkle tree is constructed in a bottom-up manner, using hash values ​​as leaf nodes. Every two nodes are combined to calculate the hash value of the parent node, until a root hash is generated. Timestamps are in UTC format, accurate to the millisecond level, and together with the hash value, form a block summary. This tree structure supports efficient data verification and partial update operations.

[0087] Block digests are encrypted using the RSA-2048 asymmetric encryption algorithm. The system generates a unique key pair for each participant: a public key for encryption and a private key for decryption. The encryption process uses the OAEP padding scheme, enhancing resistance to chosen-ciphertext attacks. The encrypted data is written to the blockchain database, with each block linked to the previous one via a hash pointer, forming an immutable chain structure.

[0088] This technical solution combines elliptic curve cryptography with blockchain technology to achieve secure storage and tamper-proofing of medical device information. ECDSA digital signatures ensure data authenticity and integrity, while a Merkle tree-based block structure provides efficient data verification. The system supports distributed storage and access control, effectively addressing the security risks of traditional centralized storage.

[0089] The implementation of this solution significantly improves the security and reliability of medical device information management. Through a dual mechanism of cryptographic protection and blockchain storage, data security is ensured during transmission and storage. The system boasts excellent scalability and interoperability, providing reliable technical support for the full lifecycle management of medical devices. Furthermore, an efficient Merkle tree-based verification mechanism reduces the computational overhead of data verification and improves system responsiveness.

[0090] In one embodiment of the medical device digital supply chain batch intelligent management method of the present application, see Figure 3 , and can also include the following:

[0091] Step S301: Construct a three-layer index structure, where the top-layer index records the storage location pointer of the box set ID, the middle-layer index stores the category and specification information of the medical device, and the bottom-layer index records the consumption records of the medical device according to the timestamp. The association relationship between the indexes at each layer is organized through a B+ tree structure, and a data validity flag is set in the bottom-layer index. The medical device information is compressed and stored using bitmap indexing technology;

[0092] Step S302: Read the historical consumption data with timestamps from the underlying index, obtain the block information digest of the corresponding timestamp from the tamper-proof blockchain database, decrypt the block information digest using the public key of the elliptic curve digital signature algorithm, perform hash calculation on the decrypted information digest and the historical consumption data, and compare the calculation result with the hash value stored in the block information digest to see if they are consistent.

[0093] Optionally, in this embodiment, these two steps establish an efficient multi-layer indexing system for medical device information and its data verification mechanism. In step S301, the system employs a three-tiered indexing architecture to organize and manage medical device information. The top-level index utilizes a sparse index structure, mapping the storage bin set ID to the physical storage location. Each index entry contains an ID value and a pointer to the actual data block. This design reduces the number of index entries and optimizes memory usage.

[0094] The middle-level index uses a dense indexing approach, detailing the categories and specifications of medical devices. Index items are sorted by category code, with each category containing sub-items with specific specifications, forming a hierarchical organizational structure. This index supports range queries and fuzzy matching, improving retrieval efficiency. The bottom-level index uses a time-series database storage model, recording medical device usage in timestamp order. Each record includes information such as timestamp, operation type, and quantity change.

[0095] A B+ tree structure organizes the relationships between these three levels of indexing. Non-leaf nodes in the B+ tree store index key values ​​and child node pointers. Leaf nodes contain pointers to actual data records and are interconnected via a linked list. This structure supports efficient range queries and sequential access. The data validity flag in the underlying index uses a single byte to indicate the record's status, including valid, deleted, and expired.

[0096] Bitmap indexing technology converts medical device information into a compressed binary format by mapping attribute values ​​to bit vectors. For attributes with limited value ranges, such as specifications and models, dictionary encoding is used before creating a bitmap index, significantly reducing storage space. Bitmap indexes support fast Boolean operations, improving the processing efficiency of complex queries.

[0097] In step S302, the system implements a data integrity verification mechanism. It first reads historical consumption data, including complete time series information, from the underlying index. Based on the timestamp, the system retrieves the corresponding block information digest from the blockchain database. This process uses the time index to accelerate searches and ensure rapid location of the target block.

[0098] The block digest is decrypted using the public key of the Elliptic Curve Digital Signature Algorithm. The decryption process uses batch mode to process verification requests for multiple blocks simultaneously, improving system throughput. The decrypted digest contains the hash value of the original data, a timestamp, and other metadata.

[0099] The system uses the same hashing algorithm to hash historical consumption data and generate a new hash value. This hashing process considers all key fields of the data, ensuring comprehensive data integrity verification. By comparing the newly generated hash value with the original hash value stored in the block, the system can detect whether the data has been tampered with.

[0100] This technical solution addresses the challenges of data organization efficiency and integrity verification in medical device information management. A multi-layer index structure provides flexible data access, while bitmap indexing technology enables efficient data compression and querying. Furthermore, a blockchain-based verification mechanism ensures data credibility.

[0101] The implementation has yielded significant results, improving system query performance, increasing storage space utilization, and reliably ensuring data integrity. This solution is particularly well-suited for the storage and management of large-scale medical device information, providing medical institutions with an efficient and reliable means of information management.

[0102] In one embodiment of the medical device digital supply chain batch intelligent management method of the present application, see Figure 4 , and can also include the following:

[0103] Step S401: Calculate the daily average consumption by aggregating historical consumption data by day according to timestamps, normalize the daily average consumption by standard deviation to eliminate dimensionality effects, use linear interpolation to supplement missing time point data, and use a low-pass filter to remove high-frequency noise in the daily average consumption to construct a multidimensional time series sample containing consumption, time periodicity, and seasonality characteristics;

[0104] Step S402: Set the length of the sliding window to 14 days and the step length to 1 day, calculate the statistical features of the data in each window, including the mean, variance, kurtosis, and skewness, use wavelet transform to extract the frequency domain features of the window data, and obtain the periodic features of the window data through autocorrelation analysis. Combine the statistical features, frequency domain features, and periodic features to form a feature vector, and construct a time series feature data set.

[0105] Optionally, in this embodiment, these two steps construct a time-series feature extraction system for medical device consumption data. In step S401, historical consumption data is first aggregated at the daily level according to timestamps. The system calculates average daily consumption using a weighted average method, with weights dynamically adjusted based on the data's credibility and integrity. For records spanning multiple days, the system allocates them to the corresponding dates based on the time proportion.

[0106] Standard deviation normalization uses the Z-score method to convert the average daily consumption into a standard normal distribution. The specific calculation formula is z = (x - μ) / σ, where x is the original value, μ is the mean, and σ is the standard deviation. This processing method eliminates dimensional differences between different medical devices, making the data comparable. For missing time point data, the system uses piecewise linear interpolation to supplement and maintain data continuity. The interpolation calculation takes into account the trends of adjacent data points to ensure the rationality of the interpolation results.

[0107] Low-pass filtering uses a Butterworth filter, with the cutoff frequency dynamically set based on the data characteristics. Zero-phase filtering is used during the filtering process to avoid phase distortion. The filtered data retains the main trend changes while removing the influence of random and abnormal fluctuations. The construction of the multidimensional time series sample considers three dimensions: basic consumption, cyclical changes, and seasonal characteristics. Cyclic characteristics are extracted through Fourier analysis, while seasonal characteristics are obtained through moving average decomposition.

[0108] In step S402, a 14-day sliding window is set based on the typical cycle of medical device usage patterns. The window sliding step is 1 day to ensure the continuity and integrity of feature extraction. Within each window, the system calculates four basic statistics: the mean reflects the overall level, the variance indicates the degree of fluctuation, the kurtosis describes the peakedness of the distribution, and the skewness indicates the symmetry of the distribution.

[0109] The wavelet transform uses the discrete wavelet transform (DWT) method and the Daubechies wavelet basis function. Through a five-layer decomposition, frequency domain features at different scales are obtained. The wavelet coefficients reflect the energy distribution of the time series at different frequency bands and can capture local and sudden changes in the data.

[0110] Autocorrelation analysis uses the partial autocorrelation function (PACF), calculating the lag period with a maximum value of half the window length. By analyzing the changing pattern of the autocorrelation coefficient, cyclical components of the time series can be identified. Significant autocorrelation coefficients correspond to important cyclical features.

[0111] The feature vector is constructed using a feature concatenation approach, combining different types of features in a predefined order. Statistical features occupy four dimensions. Frequency domain features are reduced to 10 principal components through principal component analysis. Periodic features are constructed by selecting the five most significant autocorrelation coefficients. This ultimately results in a 19-dimensional feature vector.

[0112] This technical solution solves the problem of feature extraction and representation of medical device consumption data. Through multi-level feature extraction, the system can comprehensively capture the various characteristics of consumption patterns, providing a reliable data foundation for subsequent predictive analysis.

[0113] Implementation results demonstrate that this solution effectively extracts key features of time series, resulting in feature vectors with good discriminability and stability. The system boasts high processing efficiency, moderate computational overhead, and excellent practicality. The extracted features effectively support a variety of analytical tasks, including anomaly detection and trend forecasting.

[0114] In one embodiment of the medical device digital supply chain batch intelligent management method of the present application, see Figure 5 , and can also include the following:

[0115] Step S501: Split the time series feature data into an 80% training set and a 20% validation set, calculate the maximum and minimum value intervals for each feature dimension, use the MinMax normalization method to map the feature values ​​to the [0, 1] interval, set a sigmoid activation function in the forget gate of the long short-term memory neural network to perform a nonlinear transformation on the feature data, adjust the weight distribution of the feature data through the tanh activation function of the input gate, and use the dropout mechanism to randomly inactivate some neurons to prevent overfitting;

[0116] Step S502: Construct a bidirectional long short-term memory neural network structure containing two hidden layers, set the number of neurons in each layer to 128, initialize the hidden layer parameters using the Xavier initialization method, use the Adam optimizer to iteratively update the network parameters, construct a loss function based on the root mean square error, set the learning rate to 0.001, and activate the early stopping mechanism when the loss value of the validation set does not decrease for five consecutive epochs. Save the trained model parameters for predicting the consumption range for the next 30 days.

[0117] Optionally, in this embodiment, these two steps construct a deep learning-based medical device consumption prediction system. In step S501, the time series feature data is first divided into a training set and a validation set using a stratified random sampling method, maintaining consistency in the temporal distribution of the two datasets. This data division takes temporal continuity into account, avoiding the potential for temporal feature discontinuity that may result from random division.

[0118] Feature normalization uses the MinMax method to calculate the maximum value max(f) and minimum value min(f) for each feature dimension f, and then convert the feature value x to (x-min(f)) / (max(f)-min(f)). This normalization method maintains the shape of the feature distribution while mapping all features to the same value range, which is beneficial for neural network training convergence.

[0119] The forget gate in the LSTM network uses a sigmoid activation function to convert the input to a value between 0 and 1, controlling the degree of retention of historical information. The nonlinear nature of the sigmoid function enables the network to learn complex temporal dependencies. The input gate uses the tanh activation function to map data to the range -1 to 1. This symmetric activation function helps alleviate the vanishing gradient problem.

[0120] The dropout mechanism randomly disconnects neurons during training, with a probability of 0.2. This regularization technique introduces randomness, forcing the network to learn more robust feature representations and effectively preventing overfitting. Dropout varies dynamically within each training batch but is disabled during prediction to ensure stability.

[0121] In step S502, the bidirectional LSTM network design allows the model to simultaneously consider past and future contextual information. Two hidden layers process the input sequence in the forward and reverse directions, respectively, and then combine the outputs from both directions. The number of neurons in each layer is 128, determined based on the input feature dimensionality and the complexity of the prediction task.

[0122] The Xavier initialization method sets the initial values ​​of the weights based on the input and output dimensions of the layer, ensuring that the variance of the output of each layer is approximately equal. Specifically, the weights are sampled from a uniform distribution with mean 0 and variance 2 / (nin+nout), where nin and nout are the input and output dimensions of the layer, respectively.

[0123] The Adam optimizer combines the advantages of momentum and adaptive learning rate methods, maintaining an independent learning rate for each parameter. The optimizer's hyperparameters are set to: β1 = 0.9 (exponential decay rate for the first-order moment estimate) and β2 = 0.999 (exponential decay rate for the second-order moment estimate). The root mean square error loss function penalizes the magnitude of prediction errors and is particularly suitable for regression prediction tasks.

[0124] Setting the learning rate to 0.001 is a relatively conservative choice to ensure training stability. An early stopping mechanism monitors the loss on the validation set. If the loss does not improve for five consecutive epochs, the current optimal model parameters are saved and training is stopped. This mechanism effectively prevents overfitting while conserving computing resources.

[0125] This technical solution solves the time series modeling problem in medical device consumption forecasting, capturing complex temporal dependencies through deep learning methods. The bidirectional LSTM structure fully utilizes contextual information to improve forecast accuracy.

[0126] Implementation results demonstrate that this solution achieves excellent results in both forecast accuracy and model generalization. The system accurately predicts consumption ranges for the next 30 days, providing reliable decision support for medical device inventory management. The model training is stable, computationally efficient, and possesses excellent practical value.

[0127] In one embodiment of the medical device digital supply chain batch intelligent management method of the present application, see Figure 6 , and can also include the following:

[0128] Step S601: Build a 50-layer residual deep convolutional neural network structure, set a 3×3 convolution kernel to extract local features of the medical device image, enhance feature expression capabilities through batch normalization layers and ReLU activation functions, add short-circuit connections in the residual blocks to alleviate the gradient vanishing problem, use a global average pooling layer to reduce feature dimensions, output the probability distribution of the medical device integrity classification in the fully connected layer, and output the probability distribution as the image classification result;

[0129] Step S602: Based on the DS evidence theory, a basic probability distribution function of the two evidence sources is constructed. The support function and likelihood function are calculated to represent the credibility of the image classification result and the predicted output value respectively. The Dempster combination rule is used to perform orthogonal sum operations on the two evidence sources. The Pearson correlation coefficient and root mean square error value of the fused data and the original data are calculated to evaluate the fusion effect.

[0130] Optionally, in this embodiment, these two steps construct a deep learning-based medical device integrity assessment system and its result fusion mechanism. In step S601, a 50-layer residual deep convolutional neural network is used for image feature extraction and classification. This network architecture draws on the design principles of ResNet, with each residual block containing two 3×3 convolutional layers, a batch normalization layer, and a ReLU activation function.

[0131] The 3×3 convolution kernel effectively captures local features such as texture and edges in medical device images while maintaining a moderate receptive field. The convolution operation uses a sliding method with a stride of 1, and zero padding is used on the edges to maintain the feature map size. The batch normalization layer normalizes the output of each feature channel, accelerating network training convergence and providing regularization. The ReLU activation function introduces nonlinear transformations to enhance the network's feature representation capabilities.

[0132] The short-circuit connections in the residual block implement identity mapping, allowing the network to directly learn the residual part. This design significantly alleviates the vanishing gradient problem in deep networks. The global average pooling layer compresses the spatial dimensions of the feature map, preserving information in the channel dimension and reducing the number of parameters. The fully connected layer maps the features to a classification space, with the number of output nodes equal to the number of categories in the integrity level. The output is converted to a probability distribution using the softmax function, representing the probability of the image belonging to each category.

[0133] In step S602, a data fusion framework is constructed based on the DS evidence theory. Basic probability assignment functions are constructed for both the image classification results and the predicted output values. The basic probability of the image classification is determined based on the probability value output by the softmax function, while the basic probability of the predicted value is calculated based on the confidence level of the prediction interval.

[0134] The support function represents the strength of evidence directly supporting a hypothesis, taking into account the accuracy and reliability of the data. For image classification, support primarily comes from the model's classification confidence, while support for prediction outputs is based on historical prediction accuracy. The likelihood function represents the strength of evidence indirectly supporting a hypothesis, calculated using conditional probability.

[0135] The Dempster combination rule is used to combine information from two sources of evidence. The combination process first calculates the conflict factor K, which represents the degree of contradiction between the sources. Then, an orthogonal sum operation is used to calculate the basic probability distribution after fusion. This process considers the reliability weights of the evidence sources and dynamically adjusts the influence of information from different sources.

[0136] Fusion effectiveness is evaluated using two metrics: the Pearson correlation coefficient measures the correlation between the fusion result and the original data, and the root mean square error (RMSE) reflects the accuracy of the fusion result. A correlation coefficient closer to 1 indicates that the fusion result more closely preserves the changing trends of the original data. A smaller RMS error indicates that the fusion result is closer to the true value.

[0137] This technical solution addresses the accuracy and reliability issues in medical device integrity assessments. Deep learning models provide powerful image feature extraction and classification capabilities, while the evidence theory framework enables the effective fusion of multi-source information.

[0138] Implementation results demonstrate that this solution accurately identifies the intact condition of medical devices, with highly reliable fusion results. The system operates stably, with high processing efficiency, meeting practical application requirements. Multi-source information fusion significantly improves the reliability of assessment results, providing strong support for medical device management.

[0139] In one embodiment of the medical device digital supply chain batch intelligent management method of the present application, see Figure 7 , and can also include the following:

[0140] Step S701: Using the time series correlation coefficient and the root mean square error value as input variables of the fitness function, initializing the population individuals using real number coding, setting the crossover probability to 0.8 and the mutation probability to 0.1, using the tournament selection strategy to select high-quality individuals, generating a new generation of population through the arithmetic crossover operator and Gaussian mutation operator, iteratively optimizing the weight allocation scheme until the maximum number of iterations is reached, and decoding the obtained optimal individuals into a dynamic weight allocation matrix for updating the weight parameters of the long short-term memory neural network model;

[0141] Step S702: A replenishment decision model is constructed based on a Bayesian network. The current inventory level, predicted consumption, safety stock threshold, and historical replenishment records of medical devices are used as input nodes of conditional probability. The maximum likelihood estimation method is used to learn network parameters. The posterior probability distribution of replenishment timing is calculated through a variable elimination algorithm. The posterior probability distribution is normalized to obtain the trigger probability. When the trigger probability is greater than 0.8, the replenishment demand is written into the tamper-proof blockchain database.

[0142] Optionally, in this embodiment, these two steps construct a medical device intelligent replenishment decision-making system. In step S701, a weighted optimization scheme based on a genetic algorithm is designed. The fitness function comprehensively considers the time series correlation coefficient and the root mean square error, and calculates the individual fitness values ​​using a weighted summation method. The weight coefficients are determined through cross-validation.

[0143] Real-number encoding uses floating-point numbers to represent individual chromosomes, with each gene bit corresponding to a weight parameter in the neural network. The initial population is generated through random sampling, with the sampling range determined based on the statistical distribution of the weight parameters. The population size is set to 100 to maintain population diversity while limiting computational overhead.

[0144] The tournament selection strategy randomly selects k individuals (k = 3) each time and selects the one with the highest fitness to advance to the next generation. This selection method maintains a certain level of selection pressure while preventing dominant individuals from dominating the population too quickly. The arithmetic crossover operator linearly combines the corresponding gene positions of two parent individuals, with the combination coefficient randomly generated from the interval [0, 1]. The Gaussian mutation operator adds normally distributed random perturbations to the selected gene positions, with the standard deviation decreasing with the number of iterations, thus achieving a transition from global exploration to local refined search.

[0145] The maximum number of iterations was set to 200, and the convergence condition was set: the iterations were terminated early when the relative change in the optimal fitness for 20 consecutive generations was less than 0.001. The optimal individual was decoded into a weight matrix through linear transformation and used to update the parameters of the LSTM model.

[0146] In step S702, the Bayesian network model uses a directed acyclic graph (DAG) structure, where edges between nodes represent conditional dependencies between variables. The network structure is determined through a combination of expert knowledge and data-driven approaches, ensuring that the model reflects the key factors and logical relationships in medical device replenishment decisions.

[0147] Conditional probability table parameters are learned using maximum likelihood estimation, with the expectation-maximization algorithm handling potential missing values ​​in the data. For continuous variables, a Gaussian mixture model is used for probability density estimation. The variable elimination algorithm calculates posterior probabilities using dynamic programming, and its complexity is related to the width of the network structure.

[0148] The trigger probability is normalized using the softmax function to ensure that the probability value is in the range [0, 1] and the sum is 1. The trigger threshold is set to 0.8 based on a cost-benefit analysis to strike a balance between preventing inventory shortages and avoiding overstocking.

[0149] The tamper-proof blockchain database utilizes a consortium chain architecture and a consensus mechanism based on a practical Byzantine fault-tolerance algorithm. Replenishment demand information is packaged into transaction blocks, which contain key information such as timestamps, forecast basis, and decision-making process. Blockchain technology ensures the transparency and immutability of replenishment decisions.

[0150] This technical solution addresses the challenges of intelligent and reliable decision-making for medical device replenishment. Genetic algorithm optimization improves the accuracy of the prediction model, while Bayesian networks provide interpretable decision support.

[0151] Implementation results demonstrate that this solution accurately identifies replenishment opportunities and effectively reduces inventory costs. The system exhibits excellent adaptability, dynamically adjusting decision-making strategies based on actual conditions. The introduction of blockchain technology ensures traceability and data security during the decision-making process, providing an innovative solution for medical device supply chain management.

[0152] In order to effectively improve the overall operational efficiency of the medical device supply chain, the present application provides an embodiment of a medical device digital supply chain batch intelligent management device for implementing all or part of the content of the medical device digital supply chain batch intelligent management method, see Figure 8 The medical device digital supply chain batch intelligent management device specifically includes the following contents:

[0153] An information encryption module 10 is configured to construct a unique case set ID based on the classification code of the medical device, encrypt the case set ID using an elliptic curve digital signature algorithm, perform a hash operation on the digital signature and the medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the case set ID and the tamper-proof blockchain database;

[0154] The consumption prediction module 20 is configured to write the box set ID and the medical device information into a multi-level index structure, verify data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, construct a time series sample on a daily basis for the verified historical consumption data, perform sliding window segmentation on the time series sample and extract time series features, input the time series feature data into the input layer of a long short-term memory neural network model, perform maximum and minimum value normalization on the time series feature data through a forget gate and an input gate, train the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determine the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model;

[0155] The replenishment monitoring module 30 is used to bind the medical device information with the box set ID in a batch matching manner, perform feature extraction and image classification on the returned medical devices based on the residual deep convolutional neural network, perform evidence theory data fusion on the image classification results and the output value of the long short-term memory neural network model to obtain the time series correlation coefficient and root mean square error value of the fused data, construct a dynamic weight allocation matrix based on the genetic algorithm according to the time series correlation coefficient and the root mean square error value, and update the weight parameters of the long short-term memory neural network model online according to the dynamic weight allocation matrix, record the updated weight parameters and their corresponding distribution strategies and replenishment timing in the tamper-proof blockchain database, calculate the trigger probability of the replenishment timing based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices, and automatically generate a replenishment list when the trigger probability exceeds the preset threshold.

[0156] As can be seen from the foregoing description, the medical device digital supply chain batch intelligent management device provided in this embodiment can ensure data security by integrating blockchain technology, achieve accurate demand forecasting through deep learning, and possess intelligent inventory optimization capabilities. Furthermore, the system can quickly respond to market changes, automating and intelligentizing replenishment decisions, thereby effectively improving the overall operational efficiency of the medical device supply chain.

[0157] From a hardware perspective, in order to effectively improve the overall operational efficiency of the medical device supply chain, this application provides an embodiment of an electronic device for implementing all or part of the method for batch intelligent management of a digital medical device supply chain. The electronic device specifically includes the following:

[0158] A processor, a memory, a communication interface, and a bus; wherein the processor, the memory, and the communication interface communicate with each other via the bus; the communication interface is used to realize information transmission between the medical device digital supply chain batch intelligent management device and related equipment such as the core business system, the user terminal, and the related database; the logic controller can be a desktop computer, a tablet computer, and a mobile terminal, etc., but this embodiment is not limited to this. In this embodiment, the logic controller can be implemented with reference to the embodiment of the medical device digital supply chain batch intelligent management method and the embodiment of the medical device digital supply chain batch intelligent management device in the embodiment, and their contents are merged here, and the repeated parts are not repeated.

[0159] It is understandable that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0160] In practical applications, part of the method for batch intelligent management of a digital supply chain for medical devices can be executed on the electronic device side as described above, or all operations can be completed on the client device. The specific selection can be based on the processing capabilities of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed on the client device, the client device may also include a processor.

[0161] The client device may include a communication module (i.e., a communication unit) that can establish a communication connection with a remote server to implement data transmission with the server. The server may include a server on the task scheduling center side, and in other implementation scenarios, may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0162] Figure 9 Schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Figure 9 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that the Figure 9 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0163] In one embodiment, the function of the medical device digital supply chain batch intelligent management method can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0164] Step S101: Construct a unique box set ID based on the classification code of the medical device, encrypt the box set ID using the elliptic curve digital signature algorithm, perform a hash operation on the digital signature and medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the box set ID and the tamper-proof blockchain database;

[0165] Step S102: writing the box set ID and the medical device information into a multi-level index structure, verifying data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, constructing a time series sample on a daily basis for the verified historical consumption data, performing sliding window segmentation on the time series sample and extracting time series features, inputting the time series feature data into the input layer of a long short-term memory neural network model, performing maximum and minimum value normalization processing on the time series feature data through a forget gate and an input gate, training the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determining the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model;

[0166] Step S103: The medical device information is bound to the box set ID in a batch matching manner, feature extraction and image classification are performed on the returned medical devices based on the residual deep convolutional neural network, and the image classification result is fused with the output value of the long short-term memory neural network model using evidence theory data to obtain the time series correlation coefficient and root mean square error value of the fused data, and a dynamic weight allocation matrix based on a genetic algorithm is constructed according to the time series correlation coefficient and the root mean square error value, and the weight parameters of the long short-term memory neural network model are updated online according to the dynamic weight allocation matrix, and the updated weight parameters and their corresponding distribution strategies and replenishment timing are recorded in the tamper-proof blockchain database, and the trigger probability of the replenishment timing is calculated based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices. When the trigger probability exceeds the preset threshold, a replenishment list is automatically generated.

[0167] As can be seen from the above description, the electronic device provided in the embodiments of this application utilizes blockchain technology to ensure data security, uses deep learning to achieve accurate demand forecasting, and possesses intelligent inventory optimization capabilities. Furthermore, the system can rapidly respond to market changes, automating and intelligentizing replenishment decisions, thereby effectively improving the overall operational efficiency of the medical device supply chain.

[0168] In another embodiment, the medical device digital supply chain batch intelligent management device can be configured separately from the central processor 9100. For example, the medical device digital supply chain batch intelligent management device can be configured as a chip connected to the central processor 9100, and the function of the medical device digital supply chain batch intelligent management method can be realized through the control of the central processor.

[0169] like Figure 9 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Figure 9 In addition, the electronic device 9600 may also include all components shown in Figure 9 For components not shown, reference may be made to the prior art.

[0170] like Figure 9 As shown, the central processing unit 9100 is sometimes also referred to as a controller or operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0171] Memory 9140 can be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It can store the aforementioned failure-related information and also store programs that execute the relevant information. The CPU 9100 can execute the programs stored in memory 9140 to implement information storage or processing.

[0172] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 may be, for example, a keypad or touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display objects such as images and text. The display may be, for example, an LCD display, but is not limited thereto.

[0173] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), or a SIM card. Alternatively, it may be a memory that retains information even when power is off, can be selectively erased, and is provided with more data. Examples of such memory are sometimes referred to as EPROMs. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 by the central processing unit 9100.

[0174] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various driver programs for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0175] The communication module 9110 is a transmitter / receiver that transmits and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

[0176] Based on different communication technologies, multiple communication modules 9110 can be provided in the same electronic device, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby implementing typical telecommunication functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Furthermore, the audio processor 9130 is also coupled to the central processing unit 9100, enabling local recording via the microphone 9132 and playback of stored audio via the speaker 9131.

[0177] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all steps of the method for batch intelligent management of a digital supply chain of medical devices in the above-mentioned embodiments, where the execution subject is a server or a client. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, all steps of the method for batch intelligent management of a digital supply chain of medical devices in the above-mentioned embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0178] Step S101: Construct a unique box set ID based on the classification code of the medical device, encrypt the box set ID using the elliptic curve digital signature algorithm, perform a hash operation on the digital signature and medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the box set ID and the tamper-proof blockchain database;

[0179] Step S102: writing the box set ID and the medical device information into a multi-level index structure, verifying data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, constructing a time series sample on a daily basis for the verified historical consumption data, performing sliding window segmentation on the time series sample and extracting time series features, inputting the time series feature data into the input layer of a long short-term memory neural network model, performing maximum and minimum value normalization processing on the time series feature data through a forget gate and an input gate, training the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determining the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model;

[0180] Step S103: The medical device information is bound to the box set ID in a batch matching manner, feature extraction and image classification are performed on the returned medical devices based on the residual deep convolutional neural network, and the image classification result is fused with the output value of the long short-term memory neural network model using evidence theory data to obtain the time series correlation coefficient and root mean square error value of the fused data, and a dynamic weight allocation matrix based on a genetic algorithm is constructed according to the time series correlation coefficient and the root mean square error value, and the weight parameters of the long short-term memory neural network model are updated online according to the dynamic weight allocation matrix, and the updated weight parameters and their corresponding distribution strategies and replenishment timing are recorded in the tamper-proof blockchain database, and the trigger probability of the replenishment timing is calculated based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices. When the trigger probability exceeds the preset threshold, a replenishment list is automatically generated.

[0181] As can be seen from the foregoing description, the computer-readable storage medium provided in the embodiments of this application utilizes blockchain technology to ensure data security, achieves accurate demand forecasting through deep learning, and possesses intelligent inventory optimization capabilities. Furthermore, the system can rapidly respond to market changes, automating and intelligently implementing replenishment decisions, thereby effectively improving the overall operational efficiency of the medical device supply chain.

[0182] The embodiments of the present application also provide a computer program product capable of implementing all steps of the method for batch intelligent management of a digital supply chain of medical devices in the above-mentioned embodiments, where the execution subject is a server or a client. When the computer program / instructions are executed by a processor, the steps of the method for batch intelligent management of a digital supply chain of medical devices are implemented. For example, the computer program / instructions implement the following steps:

[0183] Step S101: Construct a unique box set ID based on the classification code of the medical device, encrypt the box set ID using the elliptic curve digital signature algorithm, perform a hash operation on the digital signature and medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the box set ID and the tamper-proof blockchain database;

[0184] Step S102: writing the box set ID and the medical device information into a multi-level index structure, verifying data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, constructing a time series sample on a daily basis for the verified historical consumption data, performing sliding window segmentation on the time series sample and extracting time series features, inputting the time series feature data into the input layer of a long short-term memory neural network model, performing maximum and minimum value normalization processing on the time series feature data through a forget gate and an input gate, training the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determining the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model;

[0185] Step S103: The medical device information is bound to the box set ID in a batch matching manner, feature extraction and image classification are performed on the returned medical devices based on the residual deep convolutional neural network, and the image classification result is fused with the output value of the long short-term memory neural network model using evidence theory data to obtain the time series correlation coefficient and root mean square error value of the fused data, and a dynamic weight allocation matrix based on a genetic algorithm is constructed according to the time series correlation coefficient and the root mean square error value, and the weight parameters of the long short-term memory neural network model are updated online according to the dynamic weight allocation matrix, and the updated weight parameters and their corresponding distribution strategies and replenishment timing are recorded in the tamper-proof blockchain database, and the trigger probability of the replenishment timing is calculated based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices. When the trigger probability exceeds the preset threshold, a replenishment list is automatically generated.

[0186] As can be seen from the foregoing description, the computer program product provided in the embodiments of this application utilizes blockchain technology to ensure data security, uses deep learning to achieve accurate demand forecasting, and possesses intelligent inventory optimization capabilities. Furthermore, the system can rapidly respond to market changes, automating and intelligentizing replenishment decisions, thereby effectively improving the overall operational efficiency of the medical device supply chain.

[0187] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0188] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0189] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0190] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0191] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for intelligent batch management of a medical device digital supply chain, characterized in that: The method comprises: Construct a unique box set ID based on the classification code of the medical device, encrypt the box set ID using the elliptic curve digital signature algorithm, perform a hash operation on the digital signature and medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the box set ID and the tamper-proof blockchain database; Writing the box set ID and the medical device information into a multi-level index structure, verifying data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, constructing a time series sample on a daily basis for the verified historical consumption data, performing sliding window segmentation on the time series sample and extracting time series features, inputting the time series feature data into the input layer of a long short-term memory neural network model, performing maximum and minimum value normalization on the time series feature data through a forget gate and an input gate, training the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determining the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model; The medical device information is bound to the box set ID in a batch matching manner, and feature extraction and image classification are performed on the returned medical devices based on the residual deep convolutional neural network. The image classification result is fused with the output value of the long short-term memory neural network model using evidence theory data to obtain the time series correlation coefficient and root mean square error value of the fused data. A dynamic weight allocation matrix based on a genetic algorithm is constructed according to the time series correlation coefficient and the root mean square error value, and the weight parameters of the long short-term memory neural network model are updated online according to the dynamic weight allocation matrix. The updated weight parameters and their corresponding distribution strategies and replenishment opportunities are recorded in the tamper-proof blockchain database. Based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices, the triggering probability of the replenishment opportunity is calculated. When the triggering probability exceeds the preset threshold, a replenishment list is automatically generated.

2. The method for batch intelligent management of a digital supply chain of medical devices according to claim 1, characterized in that: The method comprises: constructing a unique box set ID based on the classification code of the medical device, encrypting the box set ID using the elliptic curve digital signature algorithm, performing a hash operation on the digital signature and the medical device information using a symmetric key to generate an information digest, storing the information digest in a tamper-proof blockchain database, and establishing a mapping relationship between the box set ID and the tamper-proof blockchain database, including: Taking the box set ID as input, a binary data block is obtained, the binary data block is mapped to a point on an elliptic curve, a public key pair is generated based on a base point P on the elliptic curve and a random number k, a signature value is calculated using the public key pair, and an encrypted digital signature is output based on the binary data block and the signature value; The production date, expiration date, product batch number, and sterilization batch number of the medical device are read as medical device information, the medical device information and the digital signature are merged into a string to be processed, the hash value of the string to be processed is iteratively calculated through the SHA-256 hash function, the hash value and timestamp are organized into a block information summary using a Merkle tree structure, the block information summary is encrypted based on an asymmetric encryption algorithm, and then written into a tamper-proof blockchain database.

3. The method for batch intelligent management of a digital supply chain of medical devices according to claim 1, characterized in that: The step of writing the box set ID and the medical device information into a multi-level index structure and verifying data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database includes: A three-layer index structure is constructed. The top-layer index records the storage location pointer of the box set ID, the middle-layer index stores the category and specification information of the medical device, and the bottom-layer index records the consumption history of the medical device according to the timestamp. The association relationship between the indexes at each layer is organized through a B+ tree structure. The data validity flag is set in the bottom-layer index, and the medical device information is compressed and stored using bitmap indexing technology. Read the historical consumption data with timestamp from the underlying index, obtain the block information digest of the corresponding timestamp from the tamper-proof blockchain database, decrypt the block information digest using the public key of the elliptic curve digital signature algorithm, perform hash calculation on the decrypted information digest and the historical consumption data, and compare the calculation result with the hash value stored in the block information digest to see if they are consistent.

4. The method for batch intelligent management of a digital supply chain of medical devices according to claim 1, characterized in that: The process of constructing a time series sample on a daily basis for the verified historical consumption data, segmenting the time series sample into sliding windows, and extracting time series features includes: The historical consumption data is aggregated daily by time stamp to calculate the average daily consumption. The average daily consumption is normalized by standard deviation to eliminate the dimensionality effect. Missing time point data is supplemented by linear interpolation. High-frequency noise in the average daily consumption is removed by a low-pass filter to construct a multidimensional time series sample containing consumption, time periodicity, and seasonality characteristics. The length of the sliding window is set to 14 days and the step length is 1 day. The statistical features of the data in each window, including the mean, variance, kurtosis, and skewness, are calculated. The frequency domain features of the window data are extracted using wavelet transform, and the periodic features of the window data are obtained through autocorrelation analysis. The statistical features, frequency domain features, and periodic features are combined to form a feature vector to construct a time series feature dataset.

5. The method for batch intelligent management of medical device digital supply chain according to claim 1, characterized in that: The method includes inputting the time series feature data into the input layer of the long short-term memory neural network model, performing maximum and minimum value normalization processing on the time series feature data through a forget gate and an input gate, training the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determining the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model, including: The time series feature data is divided into an 80% training set and a 20% validation set. The maximum and minimum value intervals are calculated for each feature dimension. The MinMax normalization method is used to map the feature values ​​to the [0, 1] interval. A sigmoid activation function is set in the forget gate of the long short-term memory neural network to perform a nonlinear transformation on the feature data. The weight distribution of the feature data is adjusted through the tanh activation function of the input gate. The dropout mechanism is used to randomly inactivate some neurons to prevent overfitting. A bidirectional long short-term memory neural network structure is constructed, which includes two hidden layers. The number of neurons in each layer is set to 128. The Xavier initialization method is used to initialize the hidden layer parameters. The Adam optimizer is used to iteratively update the network parameters. The loss function is constructed based on the root mean square error, and the learning rate is set to 0.

001. When the loss value of the validation set does not decrease for five consecutive epochs, the early stopping mechanism is activated. The trained model parameters are saved and used to predict the consumption range for the next 30 days.

6. The method for batch intelligent management of a digital supply chain of medical devices according to claim 1, characterized in that: The medical device information is bound to the box set ID in a batch matching manner, feature extraction and image classification are performed on the returned medical devices based on a residual deep convolutional neural network, and the image classification results are fused with the output value of the long short-term memory neural network model through evidence theory data to obtain the time series correlation coefficient and root mean square error value of the fused data, including: A 50-layer residual deep convolutional neural network structure was built, with a 3×3 convolution kernel set to extract local features of medical device images. Batch normalization layers and ReLU activation functions were used to enhance feature expression. Short-circuit connections were added to the residual blocks to alleviate the vanishing gradient problem. A global average pooling layer was used to reduce feature dimensionality. The fully connected layer output the probability distribution of the medical device's integrity classification, which was then output as the image classification result. Based on the DS evidence theory, a basic probability distribution function of the two evidence sources is constructed. The support function and likelihood function are calculated to represent the credibility of the image classification result and the predicted output value, respectively. The Dempster combination rule is used to perform orthogonal sum operations on the two evidence sources. The Pearson correlation coefficient and root mean square error of the fused data and the original data are calculated to evaluate the fusion effect.

7. The method for batch intelligent management of a digital supply chain of medical devices according to claim 1, characterized in that: The method comprises constructing a dynamic weight allocation matrix based on a genetic algorithm according to the time series correlation coefficient and the root mean square error value, and performing online updating of the weight parameters of the long short-term memory neural network model according to the dynamic weight allocation matrix. The updated weight parameters and their corresponding allocation strategies and replenishment timing are recorded in the tamper-proof blockchain database. The method comprises calculating the trigger probability of the replenishment timing based on the allocation strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices. When the trigger probability exceeds a preset threshold, the method comprises: The time series correlation coefficient and root mean square error value are used as input variables of the fitness function. The population individuals are initialized using real number encoding, with the crossover probability set to 0.8 and the mutation probability set to 0.

1. A tournament selection strategy is used to select high-quality individuals. A new generation of population is generated through arithmetic crossover operator and Gaussian mutation operator. The weight allocation scheme is iteratively optimized until the maximum number of iterations is reached. The optimal individual is decoded into a dynamic weight allocation matrix for updating the weight parameters of the long short-term memory neural network model. A replenishment decision model is constructed based on a Bayesian network. The current inventory level, predicted consumption, safety stock threshold and historical replenishment records of medical devices are used as input nodes of conditional probability. The maximum likelihood estimation method is used to learn the network parameters. The posterior probability distribution of replenishment timing is calculated through the variable elimination algorithm. The posterior probability distribution is normalized to obtain the trigger probability. When the trigger probability is greater than 0.8, the replenishment demand is written into the tamper-proof blockchain database.

8. A medical device digital supply chain batch intelligent management device, characterized by: The device comprises: An information encryption module is used to construct a unique box set ID based on the classification code of the medical device, encrypt the box set ID using an elliptic curve digital signature algorithm, perform a hash operation on the digital signature and medical device information using a symmetric key to generate an information digest, store the information digest in a tamper-proof blockchain database, and establish a mapping relationship between the box set ID and the tamper-proof blockchain database; a consumption prediction module, configured to write the box set ID and the medical device information into a multi-level index structure, verify data consistency based on the historical consumption data stored in the multi-level index structure and the information summary in the tamper-proof blockchain database, construct a time series sample on a daily basis for the verified historical consumption data, perform sliding window segmentation on the time series sample and extract time series features, input the time series feature data into the input layer of a long short-term memory neural network model, perform maximum and minimum value normalization on the time series feature data through a forget gate and an input gate, train the hidden layer parameters of the long short-term memory neural network model using an adaptive moment estimation optimizer, and determine the future average daily consumption and fluctuation range of medical devices based on the trained long short-term memory neural network model; A replenishment monitoring module is used to bind the medical device information with the box set ID in a batch matching manner, perform feature extraction and image classification on the returned medical devices based on the residual deep convolutional neural network, perform evidence theory data fusion on the image classification results and the output value of the long short-term memory neural network model to obtain the time series correlation coefficient and root mean square error value of the fused data, construct a dynamic weight allocation matrix based on a genetic algorithm according to the time series correlation coefficient and the root mean square error value, and update the weight parameters of the long short-term memory neural network model online according to the dynamic weight allocation matrix, record the updated weight parameters and their corresponding distribution strategies and replenishment timing in the tamper-proof blockchain database, calculate the trigger probability of the replenishment timing based on the distribution strategies recorded in the tamper-proof blockchain database and the safety stock thresholds of various medical devices, and automatically generate a replenishment list when the trigger probability exceeds the preset threshold.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the medical device digital supply chain batch intelligent management method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the medical device digital supply chain batch intelligent management method according to any one of claims 1 to 7 are implemented.

Images