Medical transaction data account checking method and system

Abstract

The invention discloses a medical transaction data reconciliation method and system, and the method comprises the steps: enabling multi-channel transaction data to be stored in a plurality of first nodes in a scattered manner, and obtaining a plurality of pieces of first distributed data, performing fragmentation processing on the first distributed data corresponding to each first node to obtain a plurality of pieces of fragmented data, storing the plurality of pieces of fragmented data in a second node different from the first node to form second distributed data, and performing hierarchical encryption on the plurality of pieces of second distributed data to obtain a plurality of pieces of encrypted data, and processing the plurality of encrypted data to obtain transaction information, and inputting the transaction information into a preset neural network model for account checking to obtain an account checking result. According to the method, medical data islands are broken through distributed storage, fund transaction data circulation and cooperative processing are promoted, hierarchical encryption and block chain technologies are adopted, data security and credibility are guaranteed, fraud and errors are prevented, intelligent reconciliation and manual intervention are combined, efficiency is improved, the error rate is reduced, and reconciliation result reliability is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of medical transactions, and in particular to a multi-channel-based medical fund transaction and reconciliation system. Background Art

[0002] With the deepening advancement of medical information technology, medical fund transaction scenarios have become significantly more diverse. In the current medical service system, patients can complete payment through various channels such as the medical insurance settlement system, WeChat, Alipay, bank transfers, and credit cards. However, data integration and sharing between different medical systems is difficult, resulting in serious data silos and affecting the efficiency and accuracy of fund transactions. At the same time, fund management within the medical system involves complex business processes such as outpatient billing, inpatient settlement, drug and consumables procurement, and medical insurance reimbursement settlement. These processes place higher demands on transaction data security and privacy protection, as well as network security during storage and transmission. The current reconciliation process is overly reliant on manual labor, which is not only inefficient, costly, and prone to errors, but also lacks a real-time detection and processing mechanism for abnormal transactions. Summary of the Invention

[0003] In order to solve the above problems existing in the prior art, the present invention provides a medical transaction data reconciliation method and system. The technical problem to be solved by the present invention is achieved through the following technical solutions: A medical transaction data reconciliation method, comprising: Receiving multi-channel transaction data and distributing and storing the multi-channel transaction data in a plurality of first nodes to obtain a plurality of first distributed data, wherein the transaction data is sourced from a hospital HIS system, a medical insurance settlement system, and a settlement platform; Sharding the first distributed data corresponding to each first node to obtain a plurality of shard data, wherein the plurality of shard data are stored in a second node different from the first node to form second distributed data; Hierarchically encrypting the plurality of the second distributed data to obtain a plurality of encrypted data, wherein the encrypted data includes: data block-level encryption information, field-level encryption information, and record-level encryption information; Processing the plurality of encrypted data to obtain transaction information; The transaction information is input into a preset neural network model for reconciliation to obtain a reconciliation result.

[0004] In a specific embodiment, sharding the first distributed data corresponding to each first node to obtain a plurality of shard data includes: Extracting key diagnosis and treatment information from the first distributed data of the hospital HIS system and the medical insurance settlement system as feature conditions; Inputting the characteristic condition into a disease diagnosis-related grouping algorithm to obtain disease diagnosis-related grouping information, and grouping the first distributed data of the hospital HIS system and the medical insurance settlement system according to the disease diagnosis-related grouping information to obtain grouped data; Processing the grouped data using a consistent hashing algorithm to obtain a hash key value; The hash key value is mapped to a consistent hash ring, and the sharding processing of the grouped data is completed according to the position relationship on the hash ring.

[0005] In a specific embodiment, sharding the first distributed data corresponding to each first node to obtain a plurality of shard data includes: Extracting key information from the first distributed data of the settlement platform; According to the key information, the first distributed data of the settlement platform is processed using a consistent hashing algorithm to obtain a hash key value; The hash key value is mapped to a consistent hash ring, and the sharding processing of the first distributed data is completed according to the position relationship on the hash ring; wherein the key information is the transaction order number and / or the payment ID.

[0006] In a specific embodiment, hierarchically encrypting the plurality of second distributed data to obtain the plurality of encrypted data includes: Dividing the second distributed data into a plurality of fixed-size data blocks, iteratively generating a chaotic sequence according to the initialized chaotic mapping parameters, performing an XOR operation on the chaotic sequence and the data blocks to obtain encrypted data blocks as data block-level encryption information; extracting personal biometric information from the second distributed data to convert the personal biometric information into an encryption key, and encrypting the personal information in the second distributed data according to the encryption key to obtain encrypted personal identity information as field-level encrypted information; Extracting personnel roles and permissions from the second distributed data to set different medical attributes, using the medical attributes as encryption parameters, and encrypting the diagnosis and treatment information, medical insurance settlement information, and payment status in the second distributed data according to the encryption parameters to obtain a medical attribute data block as record-level encrypted information; In a specific embodiment, processing the plurality of encrypted data to complete the transaction and obtain transaction information includes: Integrating the encrypted data to form a transaction data packet, performing enhanced encryption on the transaction data packet and constructing it into a blockchain transaction packet; Broadcasting the blockchain transaction package to multiple transaction nodes of the blockchain, wherein the multiple transaction nodes include multiple hospital HIS nodes, medical insurance settlement nodes, and settlement platform nodes; The blockchain transaction packages received by the multiple transaction nodes are decrypted, and transactions are performed according to the decrypted blockchain transaction packages to obtain transaction information.

[0007] In one embodiment, the encrypted data is integrated to form a transaction data packet, and the transaction data packet is subjected to enhanced encryption and constructed into a blockchain transaction packet, including: The encrypted data is converted and reorganized to form a transaction data packet; Use the blockchain's public key to encrypt the entire transaction data packet; The transaction data packet after enhanced encryption is constructed into a blockchain transaction package using a blockchain transaction construction tool; wherein the blockchain transaction package includes a sender address, a receiver address, transaction data and a digital signature.

[0008] In one embodiment, the transaction information is input into a pre-set neural network model for reconciliation, including: Inputting the transaction information into a pre-trained convolutional neural network to extract multiple feature values; Fusing the plurality of eigenvalues ​​to form a fused eigenvector; A reconciliation model based on a long short-term memory network is constructed, and the fused feature vector is input into the reconciliation model based on the long short-term memory network.

[0009] In a specific embodiment, the reconciliation model based on the long short-term memory network includes: an input layer, an LSTM layer, a feature interaction layer, an anomaly detection layer, a fully connected layer and an output layer, wherein the input layer includes a data preprocessing sublayer and a sequence construction sublayer; the LSTM layer includes an input gate sublayer, a forget gate sublayer, a memory unit sublayer and an output gate sublayer; the feature interaction layer includes a multidimensional feature cross sublayer and a feature embedding sublayer; the anomaly detection layer includes a statistical analysis sublayer, a clustering analysis sublayer and an anomaly scoring sublayer; the fully connected layer includes a classification sublayer, a regression sublayer and an activation function sublayer; the output layer includes a reconciliation result output sublayer and a risk assessment sublayer.

[0010] In one specific embodiment, the reconciliation result includes a confidence value; Accordingly, after inputting the fused feature vector into the reconciliation model based on the long short-term memory network, the following steps are further included: When the confidence value is lower than a preset confidence threshold, a manual intervention process is initiated; When the confidence value is higher than a preset confidence threshold, a corresponding report is automatically generated based on the reconciliation result.

[0011] The present invention also provides a multi-channel medical fund transaction and reconciliation system, including: a receiving storage unit, configured to receive multi-channel transaction data and disperse the multi-channel transaction data to multiple first nodes to obtain multiple first distributed data, wherein the transaction data is sourced from a hospital HIS system, a medical insurance settlement system, and a settlement platform; a sharding unit, configured to shard the first distributed data corresponding to each first node to obtain a plurality of shard data, wherein the plurality of shard data are stored in a second node different from the first node to form second distributed data; An encryption unit, configured to hierarchically encrypt the plurality of second distributed data to obtain a plurality of encrypted data, wherein the encrypted data includes: data block-level encryption information, field-level encryption information, and record-level encryption information; a processing unit, configured to process the plurality of encrypted data to obtain transaction information; The reconciliation unit is used to input the transaction information into a preset neural network model for reconciliation to obtain a reconciliation result.

[0012] Beneficial effects of the present invention: The medical transaction data reconciliation method and system of the present invention, through the distributed storage of multi-source data, breaks the data silos between medical systems, promotes the circulation and collaborative processing of financial transaction data, and improves data processing efficiency; ensures the security of medical transaction data in storage and transmission through layered encryption technology, prevents data leakage and illegal access; makes medical fund transactions tamper-proof through blockchain technology, enhances data security and credibility, and prevents fund fraud and errors; improves efficiency and reduces error rate through intelligent reconciliation, monitors transaction data in real time during reconciliation, promptly discovers abnormal transaction patterns, and ensures the accuracy and integrity of fund transactions; ensures reconciliation efficiency and result reliability through the combination of manual and intelligent reconciliation; the system can expand and integrate new data sources and transaction types, adapt to the growth of data and complex scenarios in the medical industry, and ensure long-term availability and adaptability.

[0013] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of a medical transaction data reconciliation method provided by an embodiment of the present invention; Figure 2 This is a distributed node diagram of a medical transaction data reconciliation method provided by an embodiment of the present invention; Figure 3 This is a module block diagram of a medical transaction data reconciliation system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0016] Example 1 In one embodiment, see Figure 1 , Figure 1 This is a flowchart of a medical transaction data reconciliation method. The specific steps are as follows: S1. Receiving multi-channel transaction data to disperse the multi-channel transaction data and store it in multiple first nodes to obtain multiple first distributed data; The multi-channel transaction data sources include hospital HIS systems, medical insurance settlement systems, and settlement platforms (such as WeChat Pay, Alipay, and UnionPay). The transaction data is transmitted in different formats and protocols. Upon receipt, the transaction data undergoes preliminary format conversion to a unified internal processing format. Specifically, for HL7-formatted data from hospital HIS systems, HL7 messages are parsed to extract key fields such as personal information, medical treatment information, and expense information. For XML-formatted data from medical insurance settlement systems, XML documents are parsed to extract information such as personal information, medical insurance settlement information, medical treatment information, and medical insurance policy information. For JSON-formatted data from settlement platforms (such as WeChat Pay and Alipay), JSON objects are parsed to extract information such as personal information, transaction order number, payment amount, payment status, and payment ID. For ISO 8583-formatted data from settlement platforms (such as UnionPay), ISO 8583 messages are parsed to extract information such as personal information, transaction order number, transaction type, transaction amount, payment status, payment ID, and card number information. Preliminary verification of the received transaction data is performed to ensure its integrity and consistency. Optionally, check whether the required fields of the transaction data are complete, whether the format of the transaction data meets the requirements, etc. Temporarily store the converted data in a buffer zone for subsequent processing.

[0017] In one embodiment of the present invention, see Figure 2 , Figure 2 It is a distributed node diagram of a medical transaction data reconciliation method. According to the source of multi-channel transaction data, it can be distributed and stored in multiple first nodes, each node has independent storage and computing capabilities; it can also dynamically allocate the first node, and after obtaining the first distributed data, it can also verify the integrity of the first distributed data. It automatically triggers retransmission for data packets that fail verification. For example, Figure 2As shown, the transaction data of the hospital HIS system A is stored in the first node A, the transaction data of the hospital HIS system B is stored in the first node B, the transaction data of the medical insurance settlement system C is stored in the first node C, the transaction data of the medical insurance settlement system D is stored in the first node D, the transaction data of the settlement platform E is stored in the first node E, and the transaction data of the settlement platform F is stored in the first node F. The hospital HIS system A, the hospital HIS system B, the medical insurance settlement system C, and the medical insurance settlement system D can be systems in the same province or city, or systems in different provinces or cities.

[0018] S2. Slicing the first distributed data corresponding to each first node to obtain a plurality of shard data, wherein the plurality of shard data is stored in a second node different from the first node to form a second distributed data; The consistent hashing algorithm is combined with the diagnosis related groups (DRGs) to perform sharding on the first distributed data of the hospital HIS system and the medical insurance settlement system, and the consistent hashing algorithm is used to perform sharding on the first distributed data of the settlement platform; In one embodiment of the present invention, the first distributed data of the hospital HIS system and the medical insurance settlement system is sharded using a consistent hashing algorithm in combination with diagnosis-related groups (DRGs). The specific steps are as follows: S201. Extract key medical information from the first distributed data of the hospital HIS system and the medical insurance settlement system as feature conditions; S202. Input the characteristic condition into the DRG algorithm to obtain DRG information, and group the first distributed data of the hospital HIS system and the medical insurance settlement system according to the DRG information to obtain grouped data; S203. Process the packet data using a consistent hashing algorithm to obtain a hash key value; S204. Map the hash key value to a consistent hash ring, and complete the sharding process of the grouped data according to the position relationship on the hash ring.

[0019] In one embodiment of the present invention, key medical information (e.g., diagnosis code, disease name, treatment item code, and procedure type) is extracted from the first distributed data in the hospital HIS and the medical insurance settlement system as feature conditions. For example, information such as discharge diagnosis codes (e.g., ICD-10 codes) and primary treatment procedure codes (e.g., ICD-9-CM codes) is extracted from the medical information in the hospital HIS. The extracted feature conditions are input into a DRG algorithm, which groups the feature conditions based on factors such as disease complexity and treatment modality. The DRG algorithm can use specific grouping schemes, such as MS-DRG or CHS-DRG. For example, according to the CHS-DRG scheme, medical information with a diagnosis of "acute myocardial infarction" accompanied by complex treatment procedures (e.g., coronary artery bypass grafting) is grouped into one DRG group, while medical information with a diagnosis of "pneumonia" without complications or comorbidities is grouped into another DRG group. Grouping the first distributed data in the hospital HIS and the medical insurance settlement system based on DRG information allows data belonging to the same DRG group to be grouped together. This facilitates the subsequent unified processing and analysis of similar medical data. For example, all medical records belonging to the "Acute Myocardial Infarction Complex Treatment" DRG group can be grouped together, while all medical records belonging to the "Pneumonia Without Complications" DRG group can be grouped together. A consistent hashing algorithm is applied to the grouped data to generate a hash key. Generating the hash key may involve hashing the data identifiers (such as the hospitalization number or transaction ID). For example, a consistent hashing algorithm is applied to the hospitalization number of each medical record in the "Acute Myocardial Infarction Complex Treatment" DRG group to generate a hash key. The generated hash key is mapped to a consistent hash ring to determine the data's position within the ring. The size and node distribution of the consistent hash ring can be dynamically adjusted based on system scalability and load balancing requirements. For example, the generated hash key is mapped to a consistent hash ring with multiple positions. The grouped data is then distributed across multiple shards based on the position of the hash key within the ring. This ensures even data distribution and system load balancing, while also facilitating subsequent parallel computing and data retrieval.

[0020] In one embodiment of the present invention, the first distributed data of the settlement platform is sharded using a consistent hashing algorithm. The specific steps are as follows: S211 extracts key information from the first distributed data of the settlement platform; S212. Based on the key information, use a consistent hashing algorithm to process the first distributed data to obtain a hash key value; S213. Map the hash key value to a consistent hash ring, and complete the sharding processing of the first distributed data according to the positional relationship on the hash ring; wherein the key information is the transaction order number and / or payment ID.

[0021] In one embodiment of the present invention, key information, including the transaction order number and / or payment ID, is extracted from the first distributed data of the settlement platform. A consistent hashing algorithm is used to perform a hash operation on the first distributed data, selecting the transaction order number and / or payment ID as the hash key and calculating its hash value through the hash function. The hash value is then mapped to a consistent hash ring, a virtual annular space where hash values ​​are arranged in order of size. Based on the position of the hash key value on the ring, the first distributed data is distributed to multiple shards to obtain sharded data. This ensures uniform data distribution and system load balancing, while facilitating subsequent parallel computing and data retrieval.

[0022] In one embodiment of the present invention, each shard interval corresponds to a second node, and the shard data is stored in the corresponding second node to ensure high availability and fast access to the data. Each second node is responsible for storing and managing a portion of the shard data, and performing data integrity verification and preliminary processing. Figure 2 As shown, the second node A1 to the second node An are obtained after the first node A is sharded, and the second node B1 to the second node Bn, the second node C1 to the second node Cn, the second node D1 to the second node Dn, the second node E1 to the second node En, and the second node F1 to the second node Fn are obtained similarly.

[0023] S3. The second distributed data is encrypted hierarchically to obtain multiple encrypted data, wherein the encrypted data includes; data block-level encryption information, field-level encryption information and record-level encryption information; Furthermore, the second distributed data is divided into multiple data blocks of fixed size, a chaotic sequence is iteratively generated according to the initialization chaotic mapping parameters, the chaotic sequence is XORed with the data block to obtain the encrypted data block as the data block-level encryption information; personal biometric information is extracted from the second distributed data to convert the personal biometric information into an encryption key, and the personal information in the second distributed data is encrypted according to the encryption key to obtain the encrypted personal identity information as the field-level encryption information; the roles and permissions of the personnel are extracted from the second distributed data to set different medical attributes, and the medical attributes are used as encryption parameters, and the diagnosis and treatment information, medical insurance settlement information and payment status in the second distributed data are encrypted according to the encryption parameters to obtain medical attributes The data block is used as record-level encrypted information; specifically, a unified personnel role and permission model is defined: a unified medical personnel role and permission model is constructed, covering different roles such as doctors, nurses, medical insurance administrators, and financial personnel, and the accessible information scope and operation permissions are defined for different roles; for example, doctors have the permission to view and modify patient diagnosis and treatment information, nurses have the permission to view patient diagnosis and treatment information, medical insurance administrators have the permission to view medical insurance settlement information, financial personnel have the permission to view payment status, etc.; when encrypting record-level data such as diagnosis and treatment information, based on the above-mentioned unified permission model, the role and permission information are used as encryption parameters; using the CP-ABE encryption algorithm, with role and permission information as attributes, the diagnosis and treatment information, medical insurance settlement information and payment status are encrypted to obtain medical attribute data blocks.

[0024] S4. Processing the plurality of encrypted data to obtain transaction information; specifically comprising the following steps: S41. Integrate the encrypted data to form a transaction data packet, encrypt the transaction data packet and construct a blockchain transaction packet; In one embodiment of the present invention, the encrypted data in the second node, including medical information, medical insurance settlement information, and payment status, is integrated. This data undergoes format conversion and reorganization to form a transaction data packet that meets the transaction format requirements of the blockchain. Building on the existing encryption, the entire transaction data packet is encrypted using the blockchain's public key to ensure data security within the blockchain network. The blockchain's transaction construction tool is used to construct the encrypted transaction data packet into a blockchain transaction packet. The blockchain transaction packet contains information such as the sender's address, the recipient's address, the transaction data (including encrypted personal identity information, medical information, medical insurance settlement information, and payment status), and a digital signature.

[0025] S42. Broadcasting the blockchain transaction package to multiple transaction nodes of the blockchain; wherein the multiple transaction nodes include multiple hospital HIS nodes, medical insurance settlement nodes, and settlement platform nodes; In one embodiment of the present invention, a constructed blockchain transaction package is broadcast to the entire network through multiple transaction nodes of the blockchain network. The blockchain transaction package is transmitted between the transaction nodes. Each transaction node temporarily stores the blockchain transaction package in a memory pool, waiting for verification and processing. After receiving the blockchain transaction package, the transaction node in the blockchain network will perform preliminary verification, including checking the format, signature and encryption status of the blockchain transaction package, and then propagate it to other connected transaction nodes.

[0026] S43. Decrypting the blockchain transaction packets received by the multiple transaction nodes, and performing transactions based on the decrypted blockchain transaction packets to obtain transaction information; In one embodiment of the present invention, multiple transaction nodes match blockchain transaction packets based on the sender and receiver addresses within the packets and then decrypt them. The blockchain transaction packets are initially decrypted using the public key stored on the blockchain. Subsequently, data is decrypted at the data block, field, and record levels, depending on the layered encryption hierarchy. Based on the decrypted data, transaction operations are completed at the hospital HIS node, the medical insurance settlement node, and the settlement platform node. For example, transactions are conducted through the medical insurance settlement node, the hospital financial account at the hospital HIS node, or the WeChat, Alipay, or UnionPay systems at the third-party payment node. Transaction information is recorded in a transaction log, including transaction time, transaction amount, identity information of both parties, and transaction timestamps, for audit and traceability purposes. Transaction information is organized into blocks and written to the blockchain, creating an immutable transaction record. Transaction data is regularly backed up and archived to ensure long-term data preservation and recoverability. Based on the execution of the transaction, feedback information on the success of the transaction is generated. The feedback information includes key information such as the transaction status (success or failure), transaction time, and transaction amount. The transaction result feedback information is sent to both parties of the transaction, such as the hospital HIS, medical insurance department, settlement platform, etc.; both parties of the transaction can use the feedback information to promptly understand the status and results of the transaction to ensure the smooth completion of the transaction.

[0027] S5. Input the transaction information into a pre-set neural network model to obtain a reconciliation result. The specific steps are as follows: S51. Inputting the transaction information into a pre-trained convolutional neural network to extract multiple feature values; In one embodiment of the present invention, before the transaction information is input into the convolutional neural network (CNN) to extract feature values, the transaction information is pre-processed, including normalization, standardization and other operations to ensure that the data is suitable for CNN model processing. For example, information such as transaction time, transaction amount, transaction type, etc. is converted into a unified numerical range. Key features in the transaction information are selected, such as transaction type, transaction amount, transaction timestamp, identity information of both parties to the transaction, etc., and the selected features are serialized to form a format suitable for CNN model input, and multiple feature values ​​are extracted. Specifically, the feature value can be a feature value related to the transaction time, including the transaction timestamp and the transaction time period code; the feature value can be a feature value related to the transaction amount, including the original transaction amount and the amount segmentation code; the feature value can be a feature value related to the transaction type, including the transaction type code and the transaction type frequency; the feature value can be a feature value related to the identity of both parties to the transaction, including the identity code of both parties to the transaction and the historical transaction record features of both parties to the transaction; wherein the transaction timestamp converts the transaction time into a timestamp, expressed as an integer value; the transaction time period code divides a day into different time periods and assigns a code to each time period; the original transaction amount directly uses the actual amount of the transaction. The amount segmentation coding divides the transaction amount into segments according to its size range and assigns a code to each segment; the transaction type coding encodes different transaction types, such as registration fees, examination fees, and treatment fees; the transaction type frequency statistics the frequency of each transaction type within a period of time, for example, a hospital may have 500 registration transactions and 300 examination transactions in one day; the transaction party identity coding encodes the identities of the two parties involved in the transaction (such as different departments of the hospital, medical insurance institutions, patients, etc.); the transaction party historical transaction record features extract features from the historical transaction records of the transaction parties, such as the number of transactions between the two parties in the past week and the average transaction amount.

[0028] S52. The plurality of eigenvalues ​​are fused to form a fused feature vector; In one embodiment of the present invention, the feature values ​​extracted from the transaction information are fused through a specific fusion algorithm to form a comprehensive feature vector. The fusion algorithm can be simple splicing or weighted averaging. Specifically, simple splicing can arrange all the extracted feature values ​​in a certain order and combine them into a vector. For example, the transaction timestamp, transaction time period code, original transaction amount, amount segmentation code, transaction type code, transaction type frequency, transaction party identity code, transaction party historical transaction record features, etc. are arranged in sequence to form a fused feature vector. The weighted average can be to assign a weight to each feature based on the importance and relevance of each feature value, and then calculate the weighted average. For example, for the transaction amount feature, a larger weight can be assigned because the amount is usually more critical in the reconciliation process to form a fused feature vector.

[0029] S53. Constructing a reconciliation model based on a long short-term memory network, inputting the fused feature vector into the reconciliation model based on the long short-term memory network, The reconciliation model includes: an input layer, an LSTM layer, a feature interaction layer, an anomaly detection layer, a fully connected layer and an output layer, wherein the input layer includes a data preprocessing sublayer and a sequence construction sublayer; the LSTM layer includes an input gate sublayer, a forget gate sublayer, a memory unit sublayer and an output gate sublayer; the feature interaction layer includes a multidimensional feature cross sublayer and a feature embedding sublayer; the anomaly detection layer includes a statistical analysis sublayer, a clustering analysis sublayer and an anomaly scoring sublayer; the fully connected layer includes a classification sublayer, a regression sublayer and an activation function sublayer; the output layer includes a reconciliation result output sublayer and a risk assessment sublayer.

[0030] Specifically, in one embodiment of the present invention, the input layer specifically includes the following sublayers: Data preprocessing sublayer: Preprocess the fused feature vector, which can be normalized, standardized, and remove redundant information. The fused feature vector includes transaction timestamp, transaction time period code, original transaction amount, amount segmentation code, transaction type code, transaction type frequency, identity codes of both parties to the transaction, and historical transaction record features of both parties to the transaction.

[0031] Sequence construction sublayer: Serializes the fused feature vectors to form a time series format suitable for LSTM model input. Specifically, the fused feature vectors are arranged in chronological order of transaction time to form a time series. The format of the serialized fused feature vector sequence is [[transaction timestamp 1, transaction time period code 1, original transaction amount 1, amount segment code 1, transaction type code 1, transaction type frequency 1, transaction party identity code 1, historical transaction record feature 1], [transaction timestamp 2, transaction time period code 2, original transaction amount 2, amount segment code 2, transaction type code 2, transaction type frequency 2, transaction party identity code 2, historical transaction record feature 2], ...]; each element represents the key feature of a moment, fully preserving the chronological order of transactions and the detailed feature information at each moment.

[0032] The LSTM layer specifically includes the following sublayers: Input gate sublayer: Inputs the current moment's fused feature vector, the previous moment's hidden state, and the previous moment's cell state. The previous moment's hidden state is the output of the LSTM model at the previous moment and is used to transmit historical information. The previous moment's cell state is the cell state of the LSTM model at the previous moment and is used to store long-term memory. The input gate sublayer controls how much of the current moment's input information will be written into the cell state, and uses the sigmoid function to determine which feature information needs to be updated. The output vector, which has the same dimension as the hidden state, is the input gating vector.

[0033] The forget gate sublayer takes as input the current fused feature vector and the previous hidden state to control which information in the cell state should be forgotten. The sigmoid function is used to determine which features should be forgotten. The forget gate sublayer outputs a vector with the same dimension as the hidden state, which is the forget gate vector.

[0034] Memory cell sublayer: Inputs the cell state at the previous moment, the input gating vector at the current moment, and the forget gating vector at the current moment, maintaining information transfer between moments and updating the cell state. Generates new candidate information that may be added to the cell state; outputs the updated cell state.

[0035] Output gate sublayer: Input the updated cell state and the fused feature vector at the current moment, control the output information at the current moment, determine which part of the feature information will be output through the sigmoid function, and output the hidden state at the current moment.

[0036] The feature interaction layer specifically includes the following sublayers: Multi-dimensional feature crossover sublayer: This layer performs multi-dimensional crossover on the hidden state output by the LSTM layer at the current moment to explore the potential relationships between different features. For example, it calculates the product of the transaction timestamp and the original transaction amount to analyze the impact of transaction time on the transaction amount. It also combines the identity codes of both transaction parties with the transaction type codes to explore the preferred transaction type of a specific transaction party. The layer then outputs the crossover feature matrix.

[0037] Feature embedding sublayer: Input the crossed feature matrix, embed the crossed feature matrix, map it to a higher-dimensional space to enhance the expressiveness of the features, and output a high-dimensional feature vector.

[0038] The anomaly detection layer specifically includes the following sublayers: The statistical analysis sublayer takes high-dimensional feature vectors as input, performs statistical analysis on them, and calculates statistical quantities such as their mean and variance to provide a basis for anomaly detection. For example, it calculates the mean and standard deviation of raw transaction amounts within different transaction time periods to determine whether the amount of a particular transaction within that time period deviates from the normal range. It also calculates the mean and variance of the frequency of specific transaction types within each time period to detect anomalies in transaction type frequencies. The sublayer then outputs the statistical vector and anomaly detection results for each feature.

[0039] The cluster analysis sublayer takes the statistical vector and anomaly judgment results as input, performs cluster analysis, and identifies normal and abnormal transaction patterns. For example, it uses the K-means algorithm to cluster transaction records, grouping similar transactions together, identifying abnormal transactions that deviate from normal patterns, and outputs cluster labels and abnormal transaction identifiers. Specifically, the cluster label is the cluster category to which each transaction record belongs, and the abnormal transaction identifier indicates whether each transaction record is an abnormal transaction.

[0040] The anomaly scoring sublayer takes as input the statistical vector and anomaly determination results from the statistical analysis sublayer, as well as the cluster labels and abnormal transaction identifiers from the cluster analysis sublayer. Based on the results of the statistical and cluster analysis, it calculates an anomaly score for each transaction record. For example, based on the transaction record's similarity to normal patterns, it assigns an anomaly score. A higher score indicates a higher likelihood of an anomaly. The output is an anomaly score for each transaction record, with a higher score indicating a higher likelihood of an anomaly.

[0041] The fully connected layer specifically includes the following sublayers: The classification sublayer takes the anomaly score output by the anomaly scoring sublayer and the high-dimensional feature vector output by the feature embedding sublayer as input, maps the features to the category space of reconciliation results, and outputs the probability of each category. For example, it determines whether a transaction record is reconciled successfully or failed. The classification sublayer also outputs a probability vector for the reconciliation result category, where each element represents the probability of the transaction record belonging to a certain reconciliation result category.

[0042] The regression sublayer takes as input the normalized value of the original transaction amount and the high-dimensional feature vector output by the feature embedding sublayer. It then linearly combines the input features through a fully connected layer to predict the exact transaction amount. The predicted value is then compared with the actual transaction amount, and an error metric (such as mean squared error) is calculated to assist in determining the reconciliation results. The predicted value and error metric are then output.

[0043] Activation function sublayer: Inputs the probability vector of the classification sublayer and the predicted value and error index of the regression sublayer, uses nonlinear activation functions such as ReLU and Sigmoid, etc. to increase the expressiveness of the model, and outputs the classification probability and regression result after activation processing.

[0044] The output layer specifically includes the following sublayers: Reconciliation result output sublayer: This layer inputs the classification probability output by the activation function sublayer of the fully connected layer and selects the category with the highest probability as the final reconciliation result. For example, if the "reconciliation successful" category has the highest probability, the transaction record is considered reconciled successfully, and the probability value can be used as the confidence value. The final reconciliation result and confidence value are output.

[0045] The risk assessment sublayer takes as input the anomaly score output by the anomaly scoring sublayer, the reconciliation results from the reconciliation result output sublayer, and the error indicator from the regression sublayer to assess the risk level of the transaction record and output a risk score. For example, based on the anomaly score and other risk factors, a risk score is assigned to the transaction record to help identify high-risk transactions. Ultimately, the risk score and risk level of the transaction record are output.

[0046] S8. The reconciliation result includes a confidence value; accordingly, after inputting the fused feature vector into the LSTM-based reconciliation model, the process also includes initiating a manual intervention process when the confidence value is lower than a preset confidence threshold; and automatically generating a corresponding report based on the reconciliation result when the confidence value is higher than the preset confidence threshold.

[0047] Specifically, in one embodiment of the present invention, the reconciliation results output by the reconciliation model are accompanied by a confidence value, indicating the model's degree of certainty regarding the results. This confidence value is typically a score between 0 and 1, with higher scores indicating greater confidence in the model's results. The confidence value is calculated in the final layer of the reconciliation model (typically the output layer). The model output can be converted into a probability distribution using the Softmax function to obtain a confidence value for each category. A confidence threshold is set based on the actual application scenario and business needs. This determines whether the reconciliation results require further manual intervention. For example, in scenarios requiring high accuracy, such as medical fund reconciliation, the confidence threshold can be set to 0.9 or higher. When the confidence value of the reconciliation result falls below a preset confidence threshold, manual intervention is initiated. This manual intervention can occur through a manual review interface, expert system assistance, or manual decision-making. Conversely, when the confidence value of the reconciliation result exceeds the preset confidence threshold, the result is considered highly reliable, and a corresponding report can be automatically generated. Report content includes, but is not limited to, a summary of reconciliation results, transaction details, analysis of unusual transactions, and risk assessments and recommendations. Furthermore, reports can be exported in various formats, such as PDF, Excel, and HTML, for easy viewing and archiving.

[0048] By conducting confidence assessments on the reconciliation results output by the neural network and flexibly initiating manual intervention processes or automatically generating reports based on the assessment results, the accuracy and efficiency of reconciliation are improved while reducing the workload and cost of manual intervention.

[0049] In one embodiment, see Figure 3 , Figure 3 This is a module diagram of a medical transaction data reconciliation system, including: a receiving storage unit, configured to receive multi-channel transaction data and disperse the multi-channel transaction data to multiple first nodes to obtain multiple first distributed data, wherein the transaction data is sourced from a hospital HIS system, a medical insurance settlement system, and a settlement platform; a sharding unit, configured to shard the first distributed data corresponding to each first node to obtain a plurality of shard data, wherein the plurality of shard data are stored in a second node different from the first node to form second distributed data; An encryption unit, configured to hierarchically encrypt the plurality of second distributed data to obtain a plurality of encrypted data, wherein the encrypted data includes: data block-level encryption information, field-level encryption information, and record-level encryption information; a processing unit, configured to process the plurality of encrypted data to obtain transaction information; The reconciliation unit is used to input the transaction information into a preset neural network model for reconciliation to obtain a reconciliation result.

[0050] A medical transaction data reconciliation method and system of this embodiment first effectively extracts key features through a cleaning neural network, and learns and adapts to new fault modes through a memory detection neural network, thereby improving the accuracy and timeliness of fault feature extraction; secondly, the intelligence level of power grid operation and maintenance and fault warning capabilities are improved through data set updates and model supplementary training.

[0051] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0052] Although the present application is described herein with reference to various embodiments, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims in the process of implementing the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality.

[0053] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A medical transaction data reconciliation method, characterized in that: include: Receiving multi-channel transaction data and distributing and storing the multi-channel transaction data in a plurality of first nodes to obtain a plurality of first distributed data, wherein the transaction data is sourced from a hospital HIS system, a medical insurance settlement system, and a settlement platform; Sharding the first distributed data corresponding to each first node to obtain a plurality of shard data, wherein the plurality of shard data are stored in a second node different from the first node to form second distributed data; Hierarchically encrypting the plurality of the second distributed data to obtain a plurality of encrypted data, wherein the encrypted data includes: data block-level encryption information, field-level encryption information, and record-level encryption information; Processing the plurality of encrypted data to obtain transaction information; The transaction information is input into a preset neural network model for reconciliation to obtain a reconciliation result.

2. A medical transaction data reconciliation method according to claim 1, characterized in that: The first distributed data corresponding to each first node is sharded to obtain a plurality of sharded data, including: Extracting key diagnosis and treatment information from the first distributed data of the hospital HIS system and the medical insurance settlement system as feature conditions; Inputting the characteristic condition into a disease diagnosis-related grouping algorithm to obtain disease diagnosis-related grouping information, and grouping the first distributed data of the hospital HIS system and the medical insurance settlement system according to the disease diagnosis-related grouping information to obtain grouped data; Processing the grouped data using a consistent hashing algorithm to obtain a hash key value; The hash key value is mapped to a consistent hash ring, and the sharding processing of the grouped data is completed according to the position relationship on the hash ring.

3. A medical transaction data reconciliation method according to claim 1, characterized in that: The first distributed data corresponding to each first node is sharded to obtain a plurality of sharded data, including: Extracting key information from the first distributed data of the settlement platform; According to the key information, the first distributed data of the settlement platform is processed using a consistent hashing algorithm to obtain a hash key value; The hash key value is mapped to a consistent hash ring, and the sharding processing of the first distributed data is completed according to the position relationship on the hash ring; wherein the key information is the transaction order number and / or the payment ID.

4. A medical transaction data reconciliation method according to claim 1, characterized in that: Hierarchically encrypting the plurality of second distributed data to obtain a plurality of encrypted data, including: Dividing the second distributed data into a plurality of fixed-size data blocks, iteratively generating a chaotic sequence according to the initialized chaotic mapping parameters, performing an XOR operation on the chaotic sequence and the data blocks to obtain encrypted data blocks as data block-level encryption information; extracting personal biometric information from the second distributed data to convert the personal biometric information into an encryption key, and encrypting the personal information in the second distributed data according to the encryption key to obtain encrypted personal identity information as field-level encrypted information; The roles and permissions of personnel are extracted from the second distributed data to set different medical attributes, and the medical attributes are used as encryption parameters. The diagnosis and treatment information, medical insurance settlement information and payment status in the second distributed data are encrypted according to the encryption parameters to obtain medical attribute data blocks as record-level encryption information.

5. A medical transaction data reconciliation method according to claim 4, characterized in that: Processing the plurality of encrypted data to obtain transaction information includes: Integrating the encrypted data to form a transaction data packet, performing enhanced encryption on the transaction data packet and constructing it into a blockchain transaction packet; Broadcasting the blockchain transaction package to multiple transaction nodes of the blockchain, wherein the multiple transaction nodes include multiple hospital HIS nodes, medical insurance settlement nodes, and settlement platform nodes; The blockchain transaction packages received by the multiple transaction nodes are decrypted, and transactions are performed according to the decrypted blockchain transaction packages to obtain transaction information.

6. A medical transaction data reconciliation method according to claim 5, characterized in that: The encrypted data is integrated to form a transaction data packet, and the transaction data packet is subjected to enhanced encryption and constructed into a blockchain transaction packet, including: The encrypted data is converted and reorganized to form a transaction data packet; Use the blockchain's public key to encrypt the entire transaction data packet; The transaction data packet after enhanced encryption is constructed into a blockchain transaction package using a blockchain transaction construction tool; wherein the blockchain transaction package includes a sender address, a receiver address, transaction data and a digital signature.

7. A medical transaction data reconciliation method according to claim 1, characterized in that: Input the transaction information into a pre-set neural network model for reconciliation, including: Inputting the transaction information into a pre-trained convolutional neural network to extract multiple feature values; Fusing the plurality of eigenvalues ​​to form a fused eigenvector; A reconciliation model based on a long short-term memory network is constructed, and the fused feature vector is input into the reconciliation model based on the long short-term memory network.

8. A medical transaction data reconciliation method according to claim 7, characterized in that: The reconciliation model based on the long short-term memory network includes: an input layer, an LSTM layer, a feature interaction layer, an anomaly detection layer, a fully connected layer, and an output layer, wherein the input layer includes a data preprocessing sublayer and a sequence construction sublayer; the LSTM layer includes an input gate sublayer, a forget gate sublayer, a memory unit sublayer, and an output gate sublayer; the feature interaction layer includes a multidimensional feature cross sublayer and a feature embedding sublayer; the anomaly detection layer includes a statistical analysis sublayer, a clustering analysis sublayer, and an anomaly scoring sublayer; the fully connected layer includes a classification sublayer, a regression sublayer, and an activation function sublayer; the output layer includes a reconciliation result output sublayer and a risk assessment sublayer.

9. A medical transaction data reconciliation method according to claim 7, characterized in that: The reconciliation result includes a confidence value; Accordingly, after inputting the fused feature vector into the reconciliation model based on the long short-term memory network, the following steps are further included: When the confidence value is lower than a preset confidence threshold, a manual intervention process is initiated; When the confidence value is higher than a preset confidence threshold, a corresponding report is automatically generated based on the reconciliation result.

10. A multi-channel medical fund transaction and reconciliation system, characterized by: include: a receiving storage unit, configured to receive multi-channel transaction data and disperse the multi-channel transaction data to multiple first nodes to obtain multiple first distributed data, wherein the transaction data is sourced from a hospital HIS system, a medical insurance settlement system, and a settlement platform; a sharding unit, configured to shard the first distributed data corresponding to each first node to obtain a plurality of shard data, wherein the plurality of shard data are stored in a second node different from the first node to form second distributed data; An encryption unit, configured to hierarchically encrypt the plurality of second distributed data to obtain a plurality of encrypted data, wherein the encrypted data includes: data block-level encryption information, field-level encryption information, and record-level encryption information; a processing unit, configured to process the plurality of encrypted data to obtain transaction information; The reconciliation unit is used to input the transaction information into a preset neural network model for reconciliation to obtain a reconciliation result.

Images