Informatization asset identification method based on machine learning

Abstract

The invention discloses an informatization asset identification method based on machine learning. The method comprises the following steps: step 1, data acquisition and preprocessing; step 2, cross-platform data integration; 3, designing and training a context awareness model; and 4, dynamic asset identification and decision making. By designing the cross-platform data integration module and combining the machine learning model to perform data deduplication and consistency check, standardization and efficient integration of multi-source data are realized. Compared with a traditional manual data arrangement mode, the method has the advantages that structured and unstructured data can be automatically processed, and the accuracy and integrity of the integrated data are ensured, so that the condition of information redundancy or data conflict is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of medical informationization, and in particular to an informationization asset identification method based on machine learning. Background Art

[0002] With the rapid development of informatization in the medical industry, hospitals, medical insurance systems, equipment operation and maintenance platforms, etc. have generated a large amount of heterogeneous data. These data come from different sources and have different structures, including structured table data, unstructured text data, and real-time sensor data. Traditional asset identification methods usually rely on manual statistics and simple database queries, which makes it difficult to comprehensively and accurately integrate and analyze complex and diverse data. This method is not only inefficient, but also unable to respond to dynamic changes in the hospital environment in real time, which can easily lead to a lag in asset management information, thereby affecting the reasonable deployment and maintenance decisions of equipment.

[0003] On the other hand, asset management in modern hospitals requires that the use and health status of equipment can be monitored and identified in real time, so that maintenance, deployment and even scrapping can be carried out in a timely manner to ensure the normal operation of the equipment and the diagnosis and treatment efficiency of the hospital. At the same time, the management of medical equipment also needs to fully consider the dynamic factors in the equipment use environment, such as the equipment usage rate of different departments in different time periods, the workload of the equipment, etc. However, most asset management systems on the market currently lack the ability to perceive dynamic situations in complex hospital environments and cannot make flexible identification and decisions based on environmental changes.

[0004] With the development of machine learning and deep learning technologies, intelligent analysis and prediction based on big data has become a trend. By building a machine learning model, complex, multi-source data can be efficiently integrated and analyzed, especially the asset identification method based on contextual awareness, which can dynamically adjust the identification strategy to adapt to changes in the hospital environment, thereby improving the accuracy and real-time performance of asset identification. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] In view of the above-mentioned problems existing in the existing information-based asset identification method based on machine learning, the present invention is proposed.

[0007] Therefore, the purpose of the present invention is to provide an information-based asset identification method based on machine learning, which solves the problems of data heterogeneity, lack of contextual awareness, and delayed asset identification in the prior art through four steps: data collection and preprocessing, cross-platform data integration, design and training of contextual awareness models, and dynamic asset identification and decision-making.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: an information-based asset identification method based on machine learning, comprising the following steps:

[0009] Step 1: Data collection and preprocessing

[0010] Step 1.1: Collect multi-source data from different sources, such as hospital information systems, medical insurance systems, and equipment operation and maintenance platforms, including structured data, unstructured data, and real-time data;

[0011] Step 1.2: Use machine learning algorithms to preprocess the collected data, including data cleaning, missing value filling, data conversion, and data format standardization;

[0012] Step 2: Cross-platform data integration

[0013] Step 2.1: Design and implement a cross-platform data integration module to integrate data from different platforms based on a standardized data format;

[0014] Step 2.2: Use machine learning models to perform data deduplication and consistency checks to ensure that the integrated data is accurate and contains no redundant or conflicting information;

[0015] Step 3: Design and training of context-aware models

[0016] Step 3.1: Build a situation-aware model, taking the dynamic factors in the hospital environment as input variables;

[0017] Step 3.2: Train the machine learning model to identify asset usage patterns in different scenarios and dynamically adjust the asset identification strategy;

[0018] Step 3.3: Introduce deep learning algorithms to process contextual data to improve the model’s sensitivity and accuracy to time series data, thereby providing more accurate asset identification results in different contexts;

[0019] Step 4: Dynamic asset identification and decision making

[0020] Step 4.1: Based on the integrated data and the trained context-aware model, perform real-time asset identification. The model will adjust the identification strategy according to the current context and output the status of each asset.

[0021] Step 4.2: Design a dynamic decision-making module to make corresponding decision suggestions based on the asset identification results, and feed these suggestions back to the hospital management system or operation and maintenance platform;

[0022] Step 4.3: Introduce a feedback mechanism to re-input the result data after implementation into the model to continuously optimize the accuracy and adaptability of situational awareness and asset identification.

[0023] As a preferred solution of the information-based asset identification method based on machine learning described in the present invention, the use of the machine learning algorithm in step 1.2 specifically includes:

[0024] 1. Data Cleansing

[0025] Objective: Identify and remove outliers, duplicate data, and noise in the data to ensure data quality;

[0026] Algorithm: Anomaly detection method based on statistical analysis and machine learning, using the IQR method:

[0027] For a data set X = {x1, x2, ..., x n}, calculate the quartiles Q1 and Q3:

[0028] IQR=Q3-Q1

[0029] The criteria for determining outliers are:

[0030] x i <Q1-1.5×IQR

[0031] x i >Q3+1.5×IQR

[0032] 2. Missing Value Filling

[0033] Objective: To properly fill in missing values ​​in a dataset to avoid bias in analysis or model training;

[0034] Algorithm: Using K nearest neighbors:

[0035] Find the K neighbors that are most similar to the data point with missing values ​​and fill the missing values ​​with their mean or weighted mean:

[0036] 3. Data Conversion

[0037] Goal: Convert unstructured data into vector form that can be used in machine learning models;

[0038] Algorithms: word embedding methods in natural language processing and convolutional neural networks in image processing;

[0039] Word embedding: such as Word2Vec or GloVe, converts text data into a fixed-length vector:

[0040] Word2Vec(w)=vector(w),w∈vocabulary

[0041] The vector representation of a sentence or paragraph is:

[0042]

[0043] CNN image processing: Convolution operations convert images into feature maps:

[0044]

[0045] in, is the convolution kernel, b l is the bias, σ is the activation function;

[0046] 4. Standardization of data formats

[0047] Goal: Normalize the data so that data of different dimensions have the same scale to eliminate the dimensional differences between different features;

[0048] Algorithm: Min-Max Normalization or Z-score Normalization:

[0049] Min-Max Normalization:

[0050]

[0051] Z-score normalization:

[0052]

[0053] Among them, μ is the mean of the data and σ is the standard deviation.

[0054] As a preferred solution of the information-based asset identification method based on machine learning of the present invention, the specific processing process of using the machine learning algorithm in step 1.2 is as follows:

[0055] S1. Clean the data: Use the IQR method to detect and remove outliers, then remove or merge duplicate data to reduce data redundancy;

[0056] S2. Fill missing values: Apply the KNN method and select the appropriate filling method according to the characteristics of the data set to ensure the integrity of the data;

[0057] S3, data conversion: convert text data into vectors through word embedding, convert image data into feature maps through CNN, and provide structured data for subsequent analysis after unified processing;

[0058] S4. Standardized data: Standardize all features to ensure the balance between different features, so that the model can converge better and reduce deviation during training.

[0059] As a preferred solution of the information asset identification method based on machine learning described in the present invention, in step 2.2, the specific content of using the machine learning model to perform data deduplication and consistency check includes the selection of the model, the extraction of data features, the deduplication and consistency check strategies, specifically:

[0060] Among them, model selection includes:

[0061] Model type: In data deduplication and consistency checking, commonly used models include similarity measurement models, clustering models, and classification models. Specific choices:

[0062] Similarity measurement model: Based on the cosine similarity model, it is used to calculate the similarity between data items;

[0063] Clustering model: K-means, used to cluster similar data items and identify redundant data;

[0064] Classification model: Support vector machine, used to classify and determine whether the data is consistent;

[0065] Among them, data feature extraction includes:

[0066] Feature selection: Extract important features from the data set for deduplication and consistency checking. Different types of data require different feature extraction methods.

[0067] Structured data: Extract key feature values ​​of each column, such as unique identifier ID, timestamp, and numerical features;

[0068] Unstructured data: For text data, word embedding technology is used to convert text into vector representation. For image data, convolutional neural networks are used to extract image features.

[0069] Among them, deduplication strategies include:

[0070] A1. Calculate the similarity between data items;

[0071] For structured data, the distance between data items is calculated using Euclidean distance and Manhattan distance metrics:

[0072]

[0073] For text data, use cosine similarity or Jaccard similarity for comparison:

[0074]

[0075] A2. Set a similarity threshold. If the similarity between two data items is higher than the set threshold, they are considered duplicate data and one of them is removed.

[0076] The consistency check includes:

[0077] B1. For records with high similarity, use machine learning classification models to determine their consistency;

[0078] Model training: Use historical data to train the classification model, random forest, and the output of the model is the probability of whether two records are consistent;

[0079] Model formula: Random Forest, which determines whether records are consistent based on the voting results of multiple decision trees:

[0080]

[0081] Among them, h i (x) is the prediction result of the i-th tree;

[0082] B2. If the classification model determines that the two records are highly consistent, these records are merged to ensure data consistency; if they are inconsistent, the independent records are retained.

[0083] As a preferred solution of the information-based asset identification method based on machine learning described in the present invention, wherein: in the step 2.2, the processing flow of using the machine learning model to perform data deduplication and consistency check includes:

[0084] (1) Input data: Obtain data after cross-platform integration and extract relevant features;

[0085] (2) Deduplication processing: Calculate the similarity between data items, identify and delete duplicate data based on the set similarity threshold;

[0086] (3) Consistency check: For records with high similarity but not completely consistent, the classification model is used to make consistency judgments and merge consistent data items;

[0087] (4) Output results: Generate a data set without redundancy and after consistency check for use in the subsequent asset identification process.

[0088] As a preferred solution of the information-based asset identification method based on machine learning described in the present invention, the core of the context-aware model in step 3.1 is to use dynamic factors in the hospital environment, such as time, location, equipment usage, and personnel flow, as input variables, to perceive and understand environmental changes in real time, and then dynamically adjust the identification and management of assets, including:

[0089] 1) Input variables of the context-aware model

[0090] Time factors: such as working hours, non-working hours, night shifts, weekends, and holidays;

[0091] Location factors: the geographical location of different departments in the hospital, such as the emergency department, operating room, ICU, and outpatient department;

[0092] Equipment usage: including the current status of the equipment (in use, idle, under maintenance) and historical usage records;

[0093] Personnel flow: monitor the flow of personnel in the hospital, such as medical staff, patients, and technical support staff;

[0094] Environmental factors: such as temperature, humidity, and light intensity affect the operating status of the equipment.

[0095] 2) Construction of situational awareness model

[0096] Multimodal data fusion: Fusing information from different sensors and data sources to obtain more comprehensive and accurate situational information.

[0097] formula:

[0098]

[0099] Among them C t is the total situation perception result at time t, F i (t) is the data of different sensors, w i is the weight of each data source;

[0100] Bayesian Networks: Build Bayesian network models to handle uncertainty and probabilistic reasoning;

[0101] formula:

[0102]

[0103] Where C is the scenario, E i As input variables, the model infers the current scenario by calculating conditional probabilities;

[0104] Time series analysis: Use time series models to predict possible scenario changes in the future and adjust asset management strategies in advance;

[0105] formula:

[0106] h t =σ(W x ·x t +W h ·h t-1 +b)

[0107] Among them, h t is the hidden state at the current moment, x t is the input variable, W x and W h is the weight matrix and σ is the activation function.

[0108] As a preferred solution of the information-based asset identification method based on machine learning described in the present invention, after the context-aware model is established in step 3.2, a machine learning model is trained to enable it to identify asset usage patterns in different contexts and dynamically adjust asset identification strategies, including:

[0109] C1. Preparation of training data

[0110] Data collection: Collect a variety of data from the hospital environment, including equipment usage records, situational awareness data, and historical maintenance data;

[0111] Label generation: Label each data sample to indicate the asset usage pattern in a specific context, such as high usage, low usage, and equipment failure.

[0112] C2. Selection of machine learning models

[0113] Model type: Select supervised learning model, random forest and long short-term memory network;

[0114] Random forest: It performs integrated learning through multiple decision trees and is suitable for processing complex nonlinear relationships.

[0115] Long short-term memory network: processes time series data, can remember long-term dependencies, and is suitable for capturing temporal patterns of contextual changes;

[0116] C3. Model training process

[0117] Feature extraction: extracting features from the context-aware model as input to the machine learning model;

[0118] formula:

[0119] X=[C t ,S t ,U t ,…]

[0120] Among them, X is the input feature vector, which contains the context information C t , device status S t 、Use Mode U t ;

[0121] Training model: Use the training data set to train the model and minimize the loss function to improve the prediction accuracy of the model;

[0122] Loss function:

[0123]

[0124] Among them, y i is the true label, is the model prediction value, θ is the model parameter;

[0125] Model validation and optimization: Evaluate the performance of the model through cross-validation and use hyperparameter optimization to improve the performance of the model;

[0126] C4. Dynamically adjust asset identification strategy

[0127] Real-time inference and adjustment: After the model is deployed, the asset identification strategy is dynamically adjusted based on real-time context information;

[0128] Adaptive learning: As new data is continuously input, the model can learn online, adjust and optimize itself, and further improve its adaptability to different situations.

[0129] As a preferred solution of the information-based asset identification method based on machine learning described in the present invention, the dynamic decision module in step 4.2 is intended to automatically generate decision suggestions based on the asset identification results, and feed these suggestions back to the hospital management system or operation and maintenance platform to optimize the use and maintenance strategy of medical assets. The following is the specific content of the dynamic decision module:

[0130] D1. Architecture design of dynamic decision-making module

[0131] Input layer: receives the output of the asset identification model, including the current status, usage pattern, and health status of the equipment;

[0132] Decision engine: The core part, used to analyze input data and generate corresponding decision suggestions based on preset rules or learned patterns;

[0133] Output layer: Feedback the generated decision suggestions to the hospital management system or operation and maintenance platform so as to perform corresponding operations;

[0134] D2. Decision Making

[0135] Rule-based decision making:

[0136] A set of predefined methods are provided to handle common decision-making scenarios:

[0137] Equipment status: If the equipment status is marked as "faulty" or "abnormal", immediate maintenance is recommended;

[0138] Usage priority: If a device is used frequently during a specific time period, its usage priority is adjusted to avoid malfunctions caused by excessive use;

[0139] Maintenance plan: Based on the equipment's service life and historical maintenance records, it is recommended to schedule maintenance in advance, using preventive maintenance instead of post-repair;

[0140] Machine learning aids decision making:

[0141] Use machine learning models to learn the optimal decision-making strategy from historical data, continuously optimize the decision-making process, dynamically adjust decision recommendations, and strengthen learning formulas:

[0142]

[0143] Among them, Q(s,a) is the value of taking action a in state s, r is the reward, and γ is the discount factor;

[0144] D3. Generation of decision recommendations

[0145] Real-time analysis: Based on the input equipment status and usage, real-time analysis and generation of appropriate decision suggestions;

[0146] Multi-level recommendations: Generate multiple levels of recommendations for different device types and usage scenarios, including:

[0147] Maintenance recommendations: such as "Immediate maintenance", "Planned maintenance", "No maintenance required";

[0148] Usage suggestions: such as "adjust usage priority", "reallocate resources", "temporarily stop using";

[0149] Scrapping recommendations: If the equipment is near the end of its useful life, it is recommended to consider scrapping or replacing it;

[0150] D4. Decision Feedback and Execution

[0151] Feedback mechanism: Feedback the generated decision suggestions to the hospital management system or operation and maintenance platform through the interface to ensure that the suggestions can be adopted and implemented in a timely manner;

[0152] API interface: Transmit suggestions to the management system through RESTful API to achieve automated operations;

[0153] Execution monitoring: Monitor the execution of recommendations and continuously optimize decision rules and strategies based on feedback. If the execution effect is not ideal, improve the quality of decision making by adjusting methods or retraining models.

[0154] Beneficial effects of the present invention:

[0155] 1. Improve the accuracy and efficiency of data integration: The present invention realizes the standardization and efficient integration of multi-source data by designing a cross-platform data integration module and combining machine learning models to perform data deduplication and consistency checks. Compared with the traditional manual data sorting method, the present invention can automatically process structured and unstructured data and ensure the accuracy and integrity of data after integration, thereby avoiding information redundancy or data conflicts.

[0156] 2. Dynamically adjust asset identification strategies to improve identification accuracy: By introducing a context-aware model, the present invention can analyze dynamic factors in the hospital environment in real time and adjust asset identification strategies based on dynamic information such as different departments, time periods, and equipment usage rates. This dynamic adjustment mechanism greatly improves the accuracy and real-time performance of asset identification, especially during peak equipment usage periods or special situations, and can prioritize the identification and monitoring of key equipment to ensure its normal operation.

[0157] 3. Machine learning algorithms improve recognition accuracy: By training machine learning models and introducing deep learning algorithms to process contextual data, the present invention can effectively identify the usage patterns and health status of equipment, and dynamically adjust the recognition strategy to ensure that the model has higher prediction accuracy and sensitivity in different scenarios, especially the processing of time series data, making asset identification more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0158] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0159] Figure 1 It is a schematic diagram of the framework of the information-based asset identification method based on machine learning of the present invention. DETAILED DESCRIPTION

[0160] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0161] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0162] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0163] Secondly, the present invention is described in detail with reference to the schematic diagram. When describing the embodiments of the present invention in detail, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0164] Reference Figure 1 , provides an information-based asset identification method based on machine learning, including the following steps:

[0165] Step 1: Data collection and preprocessing

[0166] Step 1.1: Collect multi-source data from different sources, such as hospital information systems, medical insurance systems, and equipment operation and maintenance platforms, including structured data, unstructured data, and real-time data;

[0167] Step 1.2: Use machine learning algorithms to preprocess the collected data, including data cleaning, missing value filling, data conversion, and data format standardization;

[0168] Step 2: Cross-platform data integration

[0169] Step 2.1: Design and implement a cross-platform data integration module to integrate data from different platforms based on a standardized data format;

[0170] Step 2.2: Use machine learning models to perform data deduplication and consistency checks to ensure that the integrated data is accurate and contains no redundant or conflicting information;

[0171] Step 3: Design and training of context-aware models

[0172] Step 3.1: Build a situation-aware model, taking the dynamic factors in the hospital environment as input variables;

[0173] Step 3.2: Train the machine learning model to identify asset usage patterns in different scenarios and dynamically adjust the asset identification strategy;

[0174] Step 3.3: Introduce deep learning algorithms to process contextual data to improve the model’s sensitivity and accuracy to time series data, thereby providing more accurate asset identification results in different contexts;

[0175] Step 4: Dynamic asset identification and decision making

[0176] Step 4.1: Based on the integrated data and the trained context-aware model, perform real-time asset identification. The model will adjust the identification strategy according to the current context and output the status of each asset.

[0177] Step 4.2: Design a dynamic decision-making module to make corresponding decision suggestions based on the asset identification results, and feed these suggestions back to the hospital management system or operation and maintenance platform;

[0178] Step 4.3: Introduce a feedback mechanism to re-input the result data after implementation into the model to continuously optimize the accuracy and adaptability of situational awareness and asset identification.

[0179] The use of machine learning algorithms in step 1.2 specifically includes:

[0180] 1. Data Cleansing

[0181] Objective: Identify and remove outliers, duplicate data, and noise in the data to ensure data quality;

[0182] Algorithm: Anomaly detection method based on statistical analysis and machine learning, using the IQR method:

[0183] For a data set X = {x1, x2, ..., x n}, calculate the quartiles Q1 and Q3:

[0184] IQR=Q3-Q1

[0185] The criteria for determining outliers are:

[0186] x i <Q1-1.5×IQR

[0187] x i >Q3+1.5×IQR

[0188] 2. Missing Value Filling

[0189] Objective: To properly fill in missing values ​​in a dataset to avoid bias in analysis or model training;

[0190] Algorithm: Using K nearest neighbors:

[0191] Find the K neighbors that are most similar to the data point with missing values ​​and fill the missing values ​​with their mean or weighted mean:

[0192] 3. Data Conversion

[0193] Goal: Convert unstructured data into vector form that can be used in machine learning models;

[0194] Algorithms: word embedding methods in natural language processing and convolutional neural networks in image processing;

[0195] Word embedding: such as Word2Vec or GloVe, converts text data into a fixed-length vector:

[0196] Word2Vec(w)=vector(w),w∈vocabulary

[0197] The vector representation of a sentence or paragraph is:

[0198]

[0199] CNN image processing: Convolution operations convert images into feature maps:

[0200]

[0201] in, is the convolution kernel, b l is the bias, σ is the activation function;

[0202] 4. Standardization of data formats

[0203] Goal: Normalize the data so that data of different dimensions have the same scale to eliminate the dimensional differences between different features;

[0204] Algorithm: Min-Max Normalization or Z-score Normalization:

[0205] Min-Max Normalization:

[0206]

[0207] Z-score normalization:

[0208]

[0209] Among them, μ is the mean of the data and σ is the standard deviation.

[0210] Among them, the specific processing process of using machine learning algorithm in step 1.2 is:

[0211] S1. Clean the data: Use the IQR method to detect and remove outliers, then remove or merge duplicate data to reduce data redundancy;

[0212] S2. Fill missing values: Apply the KNN method and select the appropriate filling method according to the characteristics of the data set to ensure the integrity of the data;

[0213] S3, data conversion: convert text data into vectors through word embedding, convert image data into feature maps through CNN, and provide structured data for subsequent analysis after unified processing;

[0214] S4. Standardized data: Standardize all features to ensure the balance between different features, so that the model can converge better and reduce deviation during training.

[0215] Specifically, in step 2.2, the specific contents of using machine learning models to perform data deduplication and consistency checks include model selection, data feature extraction, deduplication and consistency check strategies, specifically:

[0216] Among them, model selection includes:

[0217] Model type: In data deduplication and consistency checking, commonly used models include similarity measurement models, clustering models, and classification models. Specific choices:

[0218] Similarity measurement model: Based on the cosine similarity model, it is used to calculate the similarity between data items;

[0219] Clustering model: K-means, used to cluster similar data items and identify redundant data;

[0220] Classification model: Support vector machine, used to classify and determine whether the data is consistent;

[0221] Among them, data feature extraction includes:

[0222] Feature selection: Extract important features from the data set for deduplication and consistency checking. Different types of data require different feature extraction methods.

[0223] Structured data: Extract key feature values ​​of each column, such as unique identifier ID, timestamp, and numerical features;

[0224] Unstructured data: For text data, word embedding technology is used to convert text into vector representation. For image data, convolutional neural networks are used to extract image features.

[0225] Among them, deduplication strategies include:

[0226] A1. Calculate the similarity between data items;

[0227] For structured data, the distance between data items is calculated using Euclidean distance and Manhattan distance metrics:

[0228]

[0229] For text data, use cosine similarity or Jaccard similarity for comparison:

[0230]

[0231] A2. Set a similarity threshold. If the similarity between two data items is higher than the set threshold, they are considered duplicate data and one of them is removed.

[0232] The consistency check includes:

[0233] B1. For records with high similarity, use machine learning classification models to determine their consistency;

[0234] Model training: Use historical data to train the classification model, random forest, and the output of the model is the probability of whether two records are consistent;

[0235] Model formula: Random Forest, which determines whether records are consistent based on the voting results of multiple decision trees:

[0236]

[0237] Among them, h i (x) is the prediction result of the i-th tree;

[0238] B2. If the classification model determines that the two records are highly consistent, these records are merged to ensure data consistency; if they are inconsistent, the independent records are retained.

[0239] Furthermore, in step 2.2, the processing flow of using the machine learning model to perform data deduplication and consistency check includes:

[0240] (1) Input data: Obtain data after cross-platform integration and extract relevant features;

[0241] (2) Deduplication processing: Calculate the similarity between data items, identify and delete duplicate data based on the set similarity threshold;

[0242] (3) Consistency check: For records with high similarity but not completely consistent, the classification model is used to make consistency judgments and merge consistent data items;

[0243] (4) Output results: Generate a data set without redundancy and after consistency check for use in the subsequent asset identification process.

[0244] The core of the situational awareness model in step 3.1 is to use dynamic factors in the hospital environment, such as time, location, equipment usage, and personnel flow, as input variables to perceive and understand environmental changes in real time, and then dynamically adjust the identification and management of assets, including:

[0245] 1) Input variables of the context-aware model

[0246] Time factors: such as working hours, non-working hours, night shifts, weekends, and holidays;

[0247] Location factors: the geographical location of different departments in the hospital, such as the emergency department, operating room, ICU, and outpatient department;

[0248] Equipment usage: including the current status of the equipment (in use, idle, under maintenance) and historical usage records;

[0249] Personnel flow: monitor the flow of personnel in the hospital, such as medical staff, patients, and technical support staff;

[0250] Environmental factors: such as temperature, humidity, and light intensity affect the operating status of the equipment.

[0251] 2) Construction of situational awareness model

[0252] Multimodal data fusion: Fusing information from different sensors and data sources to obtain more comprehensive and accurate situational information.

[0253] formula:

[0254]

[0255] Among them C t is the total situation perception result at time t, F i (t) is the data of different sensors, w i is the weight of each data source;

[0256] Bayesian Networks: Build Bayesian network models to handle uncertainty and probabilistic reasoning;

[0257] formula:

[0258]

[0259] Where C is the scenario, E i As input variables, the model infers the current scenario by calculating conditional probabilities;

[0260] Time series analysis: Use time series models to predict possible scenario changes in the future and adjust asset management strategies in advance;

[0261] formula:

[0262] h t =σ(W x ·x t +W h ·h t-1 +b)

[0263] Among them, h t is the hidden state at the current moment, x t is the input variable, W x and W h is the weight matrix and σ is the activation function.

[0264] Furthermore, after the context-aware model is established in step 3.2, a machine learning model is trained to enable it to identify asset usage patterns in different contexts and dynamically adjust asset identification strategies, including:

[0265] C1. Preparation of training data

[0266] Data collection: Collect a variety of data from the hospital environment, including equipment usage records, situational awareness data, and historical maintenance data;

[0267] Label generation: Label each data sample to indicate the asset usage pattern in a specific context, such as high usage, low usage, and equipment failure.

[0268] C2. Selection of machine learning models

[0269] Model type: Select supervised learning model, random forest and long short-term memory network;

[0270] Random forest: It performs integrated learning through multiple decision trees and is suitable for processing complex nonlinear relationships.

[0271] Long short-term memory network: processes time series data, can remember long-term dependencies, and is suitable for capturing temporal patterns of contextual changes;

[0272] C3. Model training process

[0273] Feature extraction: extracting features from the context-aware model as input to the machine learning model;

[0274] formula:

[0275] X=[C t ,S t ,U t ,…]

[0276] Among them, X is the input feature vector, which contains the context information C t , device status S t 、Use Mode U t ;

[0277] Training model: Use the training data set to train the model and minimize the loss function to improve the prediction accuracy of the model;

[0278] Loss function:

[0279]

[0280] Among them, y i is the true label, is the model prediction value, θ is the model parameter;

[0281] Model validation and optimization: Evaluate the performance of the model through cross-validation and use hyperparameter optimization to improve the performance of the model;

[0282] C4. Dynamically adjust asset identification strategy

[0283] Real-time inference and adjustment: After the model is deployed, the asset identification strategy is dynamically adjusted based on real-time context information;

[0284] Adaptive learning: As new data is continuously input, the model can learn online, adjust and optimize itself, and further improve its adaptability to different situations.

[0285] The dynamic decision module in step 4.2 is designed to automatically generate decision suggestions based on the asset identification results, and feed these suggestions back to the hospital management system or operation and maintenance platform to optimize the use and maintenance strategy of medical assets. The following is the specific content of the dynamic decision module:

[0286] D1. Architecture design of dynamic decision-making module

[0287] Input layer: receives the output of the asset identification model, including the current status, usage pattern, and health status of the equipment;

[0288] Decision engine: The core part, used to analyze input data and generate corresponding decision suggestions based on preset rules or learned patterns;

[0289] Output layer: Feedback the generated decision suggestions to the hospital management system or operation and maintenance platform so as to perform corresponding operations;

[0290] D2. Decision Making

[0291] Rule-based decision making:

[0292] A set of predefined methods are provided to handle common decision-making scenarios:

[0293] Equipment status: If the equipment status is marked as "faulty" or "abnormal", immediate maintenance is recommended;

[0294] Usage priority: If a device is used frequently during a specific time period, its usage priority is adjusted to avoid malfunctions caused by excessive use;

[0295] Maintenance plan: Based on the equipment's service life and historical maintenance records, it is recommended to schedule maintenance in advance, using preventive maintenance instead of post-repair;

[0296] Machine learning aids decision making:

[0297] Use machine learning models to learn the optimal decision-making strategy from historical data, continuously optimize the decision-making process, dynamically adjust decision recommendations, and strengthen learning formulas:

[0298]

[0299] Among them, Q(s,a) is the value of taking action a in state s, r is the reward, and γ is the discount factor;

[0300] D3. Generation of decision recommendations

[0301] Real-time analysis: Based on the input equipment status and usage, real-time analysis and generation of appropriate decision suggestions;

[0302] Multi-level recommendations: Generate multiple levels of recommendations for different device types and usage scenarios, including:

[0303] Maintenance recommendations: such as "Immediate maintenance", "Planned maintenance", "No maintenance required";

[0304] Usage suggestions: such as "adjust usage priority", "reallocate resources", "temporarily stop using";

[0305] Scrapping recommendations: If the equipment is near the end of its useful life, it is recommended to consider scrapping or replacing it;

[0306] D4. Decision Feedback and Execution

[0307] Feedback mechanism: Feedback the generated decision suggestions to the hospital management system or operation and maintenance platform through the interface to ensure that the suggestions can be adopted and implemented in a timely manner;

[0308] API interface: Transmit suggestions to the management system through RESTful API to achieve automated operations;

[0309] Execution monitoring: Monitor the execution of recommendations and continuously optimize decision rules and strategies based on feedback. If the execution effect is not ideal, improve the quality of decision making by adjusting methods or retraining models.

[0310] Specific, practical application examples:

[0311] In hospital environments, dynamic decision-making modules and situational awareness models can be applied to the management and maintenance of high-usage equipment, thereby optimizing maintenance cycles, reducing equipment downtime, and improving overall operation and maintenance efficiency.

[0312] Practical application example 1: Optimizing the maintenance cycle of MRI equipment

[0313] Background: A large hospital has multiple magnetic resonance imaging (MRI) devices. These devices are important diagnostic tools for the hospital and are used very frequently. Due to frequent use, MRI devices often face wear, failure and maintenance needs. If maintenance is not carried out in time, equipment downtime will cause interruptions in the diagnosis and treatment process and affect the patient's treatment progress.

[0314] Application dynamic decision module

[0315] Asset Identification: The context-aware model monitors the MRI equipment in real time, recording the equipment usage frequency, working status, historical fault records, and environmental factors (such as temperature, humidity, etc.).

[0316] Scenario analysis: The model identifies that MRI equipment is most frequently used between 9 a.m. and 5 p.m. on weekdays, and that equipment failure rates are higher during this period than at other times.

[0317] Decision-making recommendations:

[0318] Dynamically adjust maintenance cycles: It is recommended to perform preventive maintenance after the equipment is in high use (such as every night or on weekends) instead of waiting until the equipment has problems before repairing it.

[0319] Increase priority: Adjust the priority of the device according to its usage to ensure that the device is always in the best working condition during peak hours.

[0320] Execution feedback: These maintenance suggestions are fed back to the hospital operation and maintenance platform, which automatically generates maintenance work orders and arranges technicians to inspect and maintain the equipment during off-peak hours.

[0321] Results: By optimizing the maintenance cycle of MRI equipment, the hospital has significantly reduced the unplanned downtime of the equipment and improved diagnostic efficiency. Since the equipment is always in good working condition, the waiting time for patients is also significantly shortened, improving the overall operational efficiency of the hospital.

[0322] Practical application example 2: High-use ventilator management

[0323] Background: During the epidemic, the demand for ventilators in a hospital increased significantly, especially in the ICU ward. Due to frequent use, ventilator equipment is prone to malfunction, resulting in failure to meet clinical needs.

[0324] Application dynamic decision module

[0325] Asset Identification: The model monitors the use of ventilators in real time, including frequency of use, failure history, and various relevant contextual data.

[0326] Scenario analysis: The model found that the usage rate of ventilators was high during the night shift, but maintenance personnel were limited, resulting in equipment failures that could not be handled in a timely manner.

[0327] Decision-making recommendations:

[0328] Adjust maintenance periods: It is recommended to reserve a certain amount of spare equipment during the night shift and increase the maintenance frequency during low-usage periods during the day.

[0329] Prioritize equipment configuration: Dynamically adjust the allocation of ventilators based on the actual needs of the ICU, and prioritize the supply of equipment during high-demand periods.

[0330] Execution feedback: Suggestions are fed back to the hospital management system to adjust the allocation and maintenance arrangements of equipment in real time.

[0331] Results: The hospital effectively reduced the failure rate of ventilators, ensured the efficient operation of the equipment, and reduced the clinical risks caused by equipment failure.

[0332] A computer device includes a memory and a processor. The computer device may be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device may be a touch layer covering the display screen, or may be a key, trackball or touchpad provided on the housing of the computer device, or may be an external keyboard, touchpad or mouse, etc.

[0333] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, an information-based asset identification method based on machine learning is implemented as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0334] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An information-based asset identification method based on machine learning, characterized in that: The following steps are involved: Step 1: Data collection and preprocessing Step 1.1: Collect multi-source data from different sources, such as hospital information systems, medical insurance systems, and equipment operation and maintenance platforms, including structured data, unstructured data, and real-time data; Step 1.2: Use machine learning algorithms to preprocess the collected data, including data cleaning, missing value filling, data conversion, and data format standardization; Step 2: Cross-platform data integration Step 2.1: Design and implement a cross-platform data integration module to integrate data from different platforms based on a standardized data format; Step 2.2: Use machine learning models to perform data deduplication and consistency checks to ensure that the integrated data is accurate and contains no redundant or conflicting information; Step 3: Design and training of context-aware models Step 3.1: Build a situation-aware model, taking the dynamic factors in the hospital environment as input variables; Step 3.2: Train the machine learning model to identify asset usage patterns in different scenarios and dynamically adjust the asset identification strategy; Step 3.3: Introduce deep learning algorithms to process contextual data to improve the model’s sensitivity and accuracy to time series data, thereby providing more accurate asset identification results in different contexts; Step 4: Dynamic asset identification and decision making Step 4.1: Based on the integrated data and the trained context-aware model, perform real-time asset identification. The model will adjust the identification strategy according to the current context and output the status of each asset. Step 4.2: Design a dynamic decision-making module to make corresponding decision suggestions based on the asset identification results, and feed these suggestions back to the hospital management system or operation and maintenance platform; Step 4.3: Introduce a feedback mechanism to re-input the result data after implementation into the model to continuously optimize the accuracy and adaptability of situational awareness and asset identification.

2. The information-based asset identification method based on machine learning according to claim 1 is characterized in that: The use of machine learning algorithms in step 1.2 specifically includes:

1. Data Cleaning Objective: Identify and remove outliers, duplicate data, and noise in the data to ensure data quality; Algorithm: Anomaly detection method based on statistical analysis and machine learning, using the IQR method: For a data set X = {x1, x2, ..., x n }, calculate the quartiles Q1 and Q3: IQR=Q3-Q1 The criteria for determining outliers are: x i <Q1-1.5×IQR x i >Q3+1.5×IQR 2. Missing Value Filling Objective: To properly fill in missing values ​​in a dataset to avoid bias in analysis or model training; Algorithm: Using K nearest neighbors: Find the K neighbors that are most similar to the data point with missing values ​​and fill the missing values ​​with their mean or weighted mean:

3. Data Conversion Goal: Convert unstructured data into vector form that can be used in machine learning models; Algorithms: word embedding methods in natural language processing and convolutional neural networks in image processing; Word embedding: such as Word2Vec or GloVe, converts text data into a fixed-length vector: Word2Vec(w)=vector(w),w∈vocabulary The vector representation of a sentence or paragraph is: CNN image processing: Convolution operations convert images into feature maps: in, is the convolution kernel, b l is the bias, σ is the activation function; 4. Standardization of data formats Goal: Normalize the data so that data of different dimensions have the same scale to eliminate the dimensional differences between different features; Algorithm: Min-Max normalization or Z-score standardization: Min-Max Normalization: Z-score standardization: Among them, μ is the mean of the data and σ is the standard deviation.

3. The information-based asset identification method based on machine learning according to claim 2 is characterized in that: The specific processing process of using the machine learning algorithm in step 1.2 is as follows: S1. Clean the data: Use the IQR method to detect and remove outliers, then remove or merge duplicate data to reduce data redundancy; S2. Fill missing values: Apply the KNN method and select the appropriate filling method according to the characteristics of the data set to ensure the integrity of the data; S3, data conversion: convert text data into vectors through word embedding, convert image data into feature maps through CNN, and provide structured data for subsequent analysis after unified processing; S4. Standardized data: Standardize all features to ensure the balance between different features, so that the model can converge better and reduce deviation during training.

4. The information-based asset identification method based on machine learning according to claim 1 is characterized in that: In step 2.2, the specific contents of using machine learning models for data deduplication and consistency checking include model selection, data feature extraction, deduplication and consistency checking strategies, specifically: Among them, model selection includes: Model type: In data deduplication and consistency checking, commonly used models include similarity measurement models, clustering models, and classification models. Specific choices: Similarity measurement model: Based on the cosine similarity model, it is used to calculate the similarity between data items; Clustering model: K-means, used to cluster similar data items and identify redundant data; Classification model: Support vector machine, used to classify and determine whether the data is consistent; Among them, data feature extraction includes: Feature selection: Extract important features from the data set for deduplication and consistency checking. Different feature extraction methods are required for different types of data. Structured data: Extract key feature values ​​of each column, such as unique identifier ID, timestamp, and numerical features; Unstructured data: For text data, we use word embedding technology to convert text into vector representation. For image data, we use convolutional neural networks to extract image features. Among them, deduplication strategies include: A1. Calculate the similarity between data items; For structured data, the distance between data items is calculated using Euclidean distance and Manhattan distance metrics: For text data, use cosine similarity or Jaccard similarity for comparison: A2. Set a similarity threshold. If the similarity between two data items is higher than the set threshold, they are considered duplicate data and one of them is removed. The consistency check includes: B1. For records with high similarity, use machine learning classification models to determine their consistency; Model training: Use historical data to train the classification model, random forest, and the output of the model is the probability of whether two records are consistent; Model formula: Random Forest, which determines whether records are consistent based on the voting results of multiple decision trees: Among them, h i (x) is the prediction result of the i-th tree; B2. If the classification model determines that the two records are highly consistent, these records are merged to ensure data consistency; if they are inconsistent, the independent records are retained.

5. The information-based asset identification method based on machine learning according to claim 4 is characterized in that: In step 2.2, the processing flow of using the machine learning model to perform data deduplication and consistency check includes: (1) Input data: Obtain data after cross-platform integration and extract relevant features; (2) Deduplication processing: Calculate the similarity between data items, identify and delete duplicate data based on the set similarity threshold; (3) Consistency check: For records with high similarity but not completely consistent, the classification model is used to make consistency judgments and merge consistent data items; (4) Output results: Generate a data set without redundancy and after consistency check for use in the subsequent asset identification process.

6. The information-based asset identification method based on machine learning according to claim 1 is characterized in that: The core of the situational awareness model in step 3.1 is to use dynamic factors in the hospital environment, such as time, location, equipment usage, and personnel flow, as input variables to perceive and understand environmental changes in real time, and then dynamically adjust the identification and management of assets, including: 1) Input variables of the context-aware model Time factors: such as working hours, non-working hours, night shifts, weekends, and holidays; Location factors: the geographical location of different departments in the hospital, such as the emergency department, operating room, ICU, and outpatient department; Equipment usage: including the current status of the equipment (in use, idle, under maintenance) and historical usage records; Personnel flow: monitor the flow of personnel in the hospital, such as medical staff, patients, and technical support staff; Environmental factors: such as temperature, humidity, and light intensity affect the operating status of the equipment. 2) Construction of situational awareness model Multimodal data fusion: Fusing information from different sensors and data sources to obtain more comprehensive and accurate situational information. formula: Among them C t is the total situation perception result at time t, F i (t) is the data of different sensors, w i is the weight of each data source; Bayesian Networks: Build Bayesian network models to handle uncertainty and probabilistic reasoning; formula: Where C is the scenario, E i As input variables, the model infers the current scenario by calculating conditional probabilities; Time series analysis: Use time series models to predict possible scenario changes in the future and adjust asset management strategies in advance; formula: h t =σ(W x ·x t +W h ·h t-1 +b) Among them, h t is the hidden state at the current moment, x t is the input variable, W x and W h is the weight matrix and σ is the activation function.

7. The information-based asset identification method based on machine learning according to claim 6 is characterized in that: After the context-aware model is established in step 3.2, a machine learning model is trained to enable it to identify asset usage patterns in different contexts and dynamically adjust asset identification strategies, including: C1. Preparation of training data Data collection: Collect a variety of data from the hospital environment, including equipment usage records, situational awareness data, and historical maintenance data; Label generation: Label each data sample to indicate the asset usage pattern in a specific context, such as high usage, low usage, and equipment failure. C2. Selection of machine learning models Model type: Select supervised learning model, random forest and long short-term memory network; Random forest: It performs integrated learning through multiple decision trees and is suitable for processing complex nonlinear relationships. Long short-term memory network: processes time series data, can remember long-term dependencies, and is suitable for capturing temporal patterns of contextual changes; C3. Model training process Feature extraction: extracting features from the context-aware model as input to the machine learning model; formula: X=[C t ,S t ,U t ,…] Among them, X is the input feature vector, which contains the context information C t , device status S t 、Use Mode U t ; Training model: Use the training data set to train the model and minimize the loss function to improve the prediction accuracy of the model; Loss function: Among them, y i is the true label, is the model prediction value, θ is the model parameter; Model validation and optimization: Evaluate the performance of the model through cross-validation and use hyperparameter optimization to improve the performance of the model; C4. Dynamically adjust asset identification strategy Real-time inference and adjustment: After the model is deployed, the asset identification strategy is dynamically adjusted based on real-time context information; Adaptive learning: As new data is continuously input, the model can learn online, adjust and optimize itself, and further improve its adaptability to different situations.

8. The information-based asset identification method based on machine learning according to claim 1 is characterized in that: The dynamic decision module in step 4.2 is designed to automatically generate decision suggestions based on the asset identification results, and feed these suggestions back to the hospital management system or operation and maintenance platform to optimize the use and maintenance strategy of medical assets. The following is the specific content of the dynamic decision module: D1. Architecture design of dynamic decision-making module Input layer: receives the output of the asset identification model, including the current status, usage pattern, and health status of the equipment; Decision engine: The core part, used to analyze input data and generate corresponding decision suggestions based on preset rules or learned patterns; Output layer: Feedback the generated decision suggestions to the hospital management system or operation and maintenance platform so as to perform corresponding operations; D2. Decision Making Rule-based decision making: A set of predefined methods are provided to handle common decision-making scenarios: Equipment status: If the equipment status is marked as "fault" or "abnormal", immediate maintenance is recommended; Usage priority: If a device is used frequently during a specific time period, its usage priority is adjusted to avoid malfunctions caused by excessive use; Maintenance plan: Based on the equipment's service life and historical maintenance records, it is recommended to schedule maintenance in advance, using preventive maintenance instead of post-repair; Machine learning aids decision making: Use machine learning models to learn the best decision-making strategy from historical data, continuously optimize the decision-making process, dynamically adjust decision recommendations, and strengthen learning formulas: Among them, Q(s,a) is the value of taking action a in state s, r is the reward, and γ is the discount factor; D3. Generation of decision recommendations Real-time analysis: Based on the input equipment status and usage, real-time analysis and appropriate decision suggestions are generated; Multi-level recommendations: Generate multiple levels of recommendations for different device types and usage scenarios, including: Maintenance recommendations: such as "Immediate maintenance", "Planned maintenance", "No maintenance required"; Usage suggestions: such as "adjust usage priority", "reallocate resources", "temporarily stop using"; Scrapping recommendations: If the equipment is near the end of its useful life, it is recommended to consider scrapping or replacing it; D4. Decision Feedback and Execution Feedback mechanism: Feedback the generated decision suggestions to the hospital management system or operation and maintenance platform through the interface to ensure that the suggestions can be adopted and implemented in a timely manner; API interface: Transmit suggestions to the management system through RESTful API to achieve automated operations; Execution monitoring: Monitor the execution of recommendations and continuously optimize decision rules and strategies based on feedback. If the execution effect is not ideal, improve the decision quality by adjusting the method or retraining the model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the information-based asset identification method based on machine learning described in any one of claims 1 to 8 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 an information-based asset identification method based on machine learning as described in any one of claims 1 to 8 are implemented.

Images