Intelligent blood product management method and system based on multi-dimensional data

Abstract

The invention relates to an intelligent blood product management method and system based on multi-dimensional data, and solves the problems that a manual prediction mode is poor in accuracy and cannot meet real-time changing blood demands of all departments, so that supply and demand of blood products are not matched, and even resource waste or insufficient supply is caused. The method comprises the following steps: acquiring blood demand information of each department of a hospital, wherein the blood demand information comprises blood type categories and demand quantities of different blood type categories; according to the blood demand information of each department of the hospital, a blood packaging robot is started to package blood according to the blood demand information to form a blood product, and a bar code or an RFID tag is pasted on the blood product; and the blood distribution robot is started to store the blood products and perform blood distribution of each department of the hospital along a preset distribution route. The blood transfusion system has the advantages that the working efficiency of the blood transfusion department and the blood use safety are remarkably improved, and timely and accurate blood transfusion service is provided for patients.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical treatment and logistics distribution technology, and in particular to an intelligent blood product management method and system based on multi-dimensional data. Background Art

[0002] At present, hospitals still manage blood products in a relatively traditional way. Although some automation technologies have been applied to the storage and distribution of blood products, there are still many deficiencies in blood demand analysis, prediction, and intelligent packaging of blood products. Existing technologies cannot accurately predict the blood demand of each department, nor can they intelligently respond to the demand for special blood products.

[0003] Most existing technical solutions rely on medical staff to make manual predictions based on experience or simple statistical data, and then package and distribute the blood. Some hospitals have initially introduced automated distribution systems, but the level of intelligence is limited and accurate predictions and personalized services cannot be achieved. Summary of the invention

[0004] In order to improve the intelligence level of hospital blood product management, accurately predict the blood needs of various departments, improve packaging and distribution efficiency, and ensure rapid response to special blood needs, the present application provides an intelligent blood product management method and system based on multi-dimensional data.

[0005] In the first aspect, the present application provides an intelligent blood product management method based on multi-dimensional data, which adopts the following technical solutions: An intelligent blood product management method based on multi-dimensional data, comprising: Obtain blood demand information of each department in the hospital, including blood type categories and demand for different blood types; According to the blood demand information of each department of the hospital, the blood packaging robot is started to package the blood into blood products according to the blood demand information, and a barcode or RFID tag is attached to the blood products; The blood distribution robot is activated to store blood products and distribute blood to various departments of the hospital along the preset distribution routes.

[0006] By adopting the above technical solutions, the efficiency and safety of hospital blood management can be significantly improved. By accurately obtaining the blood needs of each department, this method can ensure that the requirements of different blood types and demand quantities are met in a timely and accurate manner. The use of robots for blood packaging and distribution not only greatly saves manpower, but also reduces the risk of human error. The application of barcodes or RFID tags further enhances the traceability and management convenience of blood products. Overall, this method helps to realize the intelligent, automated and precise management of hospital blood.

[0007] Optionally, obtain blood demand information for each department of the hospital including: Get the preset time steps of each department in the hospital; The preset time step of the same department of the hospital is input into the trained ARIMA model of the corresponding department to generate the predicted blood demand information of the corresponding department of the hospital.

[0008] By adopting the above technical solution and applying the ARIMA model to analyze the preset time step data of each department, the future blood demand of each department in the hospital can be accurately predicted. This method not only improves the predictability and planning of blood management, but also helps hospitals to plan blood inventory in advance and ensure the timely supply of blood for clinical use. At the same time, the forecast data can provide a scientific basis for the hospital's blood management decisions and reduce the cost of blood waste and excessive reserves. In general, this solution will help hospitals achieve accurate prediction and intelligent management of blood needs.

[0009] Optionally, obtaining the trained ARIMA model of the corresponding department includes: Collect the historical blood usage data of the corresponding departments and perform data preprocessing, wherein the historical blood usage data of each department includes but is not limited to the historical blood usage volume, blood usage type, and blood usage time of each department, and the data preprocessing includes but is not limited to filling missing values, processing outliers, and smoothing noise; Parameter estimation based on information criteria to determine the optimal order of the ARIMA model; Using the selected parameters, the historical data is fitted by maximum likelihood estimation to train the coefficients of the ARIMA model; Compare the model prediction results with historical data and calculate the mean square error. If and only if the mean square error is less than the preset error value, the ARIMA model training of the corresponding department is determined to have completed.

[0010] By adopting the above technical solution, through scientific data preprocessing and precise parameter estimation, the ARIMA model of the corresponding department was successfully trained. The application of this model can achieve high-precision prediction of the department's future blood demand and effectively improve the management efficiency of blood resources. By comparing with the actual historical data, the mean square error of the model is less than the preset value, showing its superior prediction performance. This method not only helps to improve the rational allocation of medical resources, but also reduces resource waste while ensuring clinical blood use, which is of great significance to the sustainable development and optimization of the medical system.

[0011] Optionally, obtain blood demand information for each department of the hospital including: Analyze whether the difference between the blood demand predicted by the ARIMA model and the actual blood demand of each department of the hospital within the preset time period exceeds the preset difference; If not, maintain the original setting; If yes, then obtain the historical forecast data of the ARIMA model, the actual blood demand data, and the patient's diagnosis and treatment dynamics information, including but not limited to the patient's condition changes, surgery arrangements, and treatment plan adjustment information; The historical forecast data of the ARIMA model, the actual blood demand data, and the patient diagnosis and treatment dynamic information are integrated, and the features closely related to blood demand are extracted through feature engineering, and the extracted features are converted into numerical forms suitable for machine learning models; Use the processed data set to train the gradient boosting tree model and use cross-validation techniques to select the best model parameters; Using the preset weighted average method, the ARIMA model and the gradient boosting tree model are integrated to form a new prediction model; According to the new prediction model, the blood demand information of each department in the hospital is obtained.

[0012] By adopting the above technical solutions and cleverly combining the ARIMA model and the gradient boosting tree model, the accuracy of blood demand forecasting in various departments of the hospital has been significantly improved. By analyzing the difference between the forecast and the actual demand in real time and dynamically adjusting the forecasting model, accurate prediction of blood demand is ensured. The integration of dynamic information on patient diagnosis and treatment makes the model more flexible and responsive in real time, and can adapt to changing clinical situations. This solution not only helps hospitals to accurately manage blood resources and reduce waste, but also provides patients with more timely and efficient medical services, reflecting the important progress of medical intelligence.

[0013] Optionally, forming a new prediction model includes: Use the validation set to evaluate the ARIMA model and the gradient boosting tree model, calculate the mean square error of different models, and use the inverse of the mean square error of each model as the weight coefficient of the corresponding model; According to the weight coefficients of the corresponding models obtained through analysis and the weighted average method, the prediction results of the ARIMA model and the gradient boosting tree model are fused to form a new prediction model.

[0014] By adopting the above technical solutions and evaluating the validation set, we assigned weight coefficients based on the inverse of the mean square error to the ARIMA model and the gradient boosting tree model. Using the weighted average method, we successfully combined the prediction results of the two models and built a new prediction model. This model brings together the advantages of the two single models and significantly improves the accuracy and robustness of the prediction. This comprehensive prediction method helps us grasp the data trends more comprehensively and optimize decision-making. In general, the fusion model provides a more powerful tool for prediction.

[0015] Optionally, based on the blood demand information of each department of the hospital, the blood packaging robot is started to package blood according to the blood demand information to form blood products, including: Analyze whether there is any special blood product demand information proactively issued by various departments of the hospital; If yes, then according to the special blood product demand information, the blood packaging robot is started to package the blood to form special blood products first according to the special blood product demand information, and then continues to package the blood to form blood products according to the blood demand information; If no, remain unchanged.

[0016] By adopting the above technical solutions, the blood needs of various departments in the hospital are intelligently analyzed and the blood packaging process is optimized. In particular, for the needs of special blood products, the robot's work priority can be quickly adjusted to ensure that key issues are resolved in a timely manner. This improves the flexibility and efficiency of medical services, reduces patient waiting time, and also improves the rational allocation and utilization of blood resources. Overall, this innovative measure helps to improve the operational efficiency of the medical system and patient satisfaction, and demonstrates the great potential of intelligent medical equipment in medical services.

[0017] Optional special blood demand information proactively issued by hospital departments include: Obtain patient disease-related information uploaded to the system by various departments of the hospital, including specific diseases, personal blood transfusion allergy history, age and physical condition information; Analyze whether the patient disease-related information uploaded to the system by various departments of the hospital contains specific diseases or personal blood transfusion allergy history, and whether there is allergy to some blood components; If yes, the patient's disease-related information is used as input information to the trained special blood product prediction model, and the special blood demand information is output as the special blood demand information proactively issued by various departments of the hospital; If no, remain unchanged.

[0018] By adopting the above technical solutions, deep learning of patient disease information can achieve accurate prediction of specific blood needs. For patients with specific diseases and a history of blood transfusion allergy, personalized blood demand information can be quickly generated, effectively avoiding medical risks caused by improper blood transfusion. At the same time, this intelligent process greatly improves the response speed and accuracy of medical services, ensuring the safety and timeliness of blood use for patients. Overall, this innovative measure not only improves the quality of medical care, but also demonstrates the broad application prospects of artificial intelligence in the medical field.

[0019] Optional barcode or RFID labeling of blood products includes: Analyze whether blood products are special blood products; If yes, then the patient's disease-related information is used as input information to the urgency analysis model, and the urgency of the patient served by the special blood product is output; Attach barcodes or RFID tags to blood products and add urgency information, as well as special identification patterns; If not, keep the original setting.

[0020] By adopting the above technical solutions, barcodes or RFID tags are affixed to blood products, and urgency information and special identification patterns are added, which significantly improves the efficiency and accuracy of blood management. For special blood products, this measure can quickly convey key information about the patient's urgency and blood characteristics, helping medical staff to make quick and accurate decisions. At the same time, the application of labels also optimizes the inventory management process and improves the traceability of blood resources. Overall, this innovative approach ensures the safety and effectiveness of the use of blood products and brings important improvements to medical services.

[0021] Optionally, starting the blood distribution robot to store blood products and distribute blood to various departments of the hospital along a preset distribution route includes: Start the blood distribution robot to identify the barcode or RFID tag on the packaged blood products, give priority to assembling the blood products with urgency information, and reserve the remaining blood products in the preset department blood product reserve priority order according to the remaining reserve space of the blood distribution robot; The blood distribution robot is started to sort the distribution order of departments from high to low according to the urgency of different departments and distribute blood products. The distribution order of the remaining departments is determined by the shortest path planning based on the last distribution location of the blood distribution robot.

[0022] By adopting the above technical solutions, using blood distribution robots, and intelligently identifying barcodes or RFID tags, efficient and accurate blood product distribution is achieved. Prioritizing the distribution of blood products with high urgency ensures the timeliness of blood supply in emergency situations. At the same time, combined with distribution route optimization and reserve priority setting, distribution efficiency and resource utilization are improved. This solution not only ensures the safety and timeliness of blood use for patients, but also reduces the workload of medical staff, demonstrating the important value of intelligent medical equipment in the modern medical service system.

[0023] In the second aspect, the present application provides an intelligent blood management system based on multi-dimensional data, which adopts the following technical solutions: An intelligent blood management system based on multi-dimensional data comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein the program can implement the intelligent blood product management method based on multi-dimensional data as described in the first aspect when loaded and executed by the processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the overall process of an intelligent blood product management method based on multi-dimensional data in an embodiment of the present application.

[0025] Figure 2 It is a flowchart of another embodiment of the present application for obtaining blood demand information of various departments in a hospital.

[0026] Figure 3 It is a flowchart of another embodiment of the present application for obtaining the ARIMA model of the corresponding department that has completed training.

[0027] Figure 4 It is a flowchart of another embodiment of the present application for obtaining blood demand information of various departments in a hospital. DETAILED DESCRIPTION

[0028] The present application is further described in detail below in conjunction with the accompanying drawings.

[0029] Reference Figure 1 , is an intelligent blood product management method based on multi-dimensional data disclosed in this application, comprising: Step S100, obtaining blood demand information of each department of the hospital.

[0030] The blood demand information includes blood type categories and the demand for different blood type categories.

[0031] Specifically, the blood demand information of each department of the hospital can be obtained through the hospital's internal information system or dedicated software platform.

[0032] For example: suppose the hospital has a central information system through which each department can submit blood demand applications. The system background will automatically summarize and analyze these data to obtain the total demand for each blood type. Step S100 is to obtain the latest blood demand information in real time by connecting to the interface of this information system. This information may exist in the form of a data table or database record, including key fields such as blood type (such as type A, type B, type AB, type O), demand (such as XX bags or XX milliliters), etc. After obtaining this information, the subsequent blood packaging and distribution work can be carried out more accurately and efficiently.

[0033] Step S200, based on the blood demand information of each department of the hospital, start the blood packaging robot to package the blood into blood products according to the blood demand information, and attach a barcode or RFID tag to the blood product.

[0034] Specifically, the blood packaging robot will receive the blood demand information of each department of the hospital obtained in step S100, and automatically select the corresponding blood type and quantity from the blood bank for packaging. During the packaging process, the robot uses machine vision technology to ensure that each blood package is accurate. After packaging, the robot will attach a barcode or RFID tag to each blood product. These tags can not only track and record the information of the blood, but also ensure the safety and traceability of the blood during use.

[0035] For example, if a hospital needs 10 bags of type A blood and 5 bags of type B blood, the blood packaging robot will automatically select the qualified blood from the inventory, and then carry out pollution-free and leak-free safe packaging according to the prescribed process. After packaging, the robot will affix the corresponding barcode or RFID tag on each bag, which records the key information of the bag of blood, such as blood type, collection date, and expiration date, for subsequent management and tracking.

[0036] Step S300, start the blood distribution robot to store blood products and distribute blood to various departments of the hospital along a preset distribution route.

[0037] Specifically, blood products are delivered directly to the storage space of the blood distribution robot through automated conveying devices (such as conveyor belts). The blood distribution robot is equipped with a specially designed refrigerated storage space with an internal temperature and humidity control system to ensure that the storage conditions of blood products meet the requirements. The storage space is divided into different areas or grids, each of which can be used to place blood products of a specific type or degree of urgency, for easy classification and quick access. When receiving blood products, the blood distribution robot reads the barcode or RFID tag, enters the relevant information of the blood products (such as blood type, demand, urgency, expiration date, etc.) into its own control system, and synchronizes it with the hospital's central information system. Based on the information received, the robot's control system automatically assigns a storage location for each blood product and records it in the system.

[0038] In addition, the blood delivery robot uses laser radar, infrared sensors, etc. to achieve accurate environmental perception and obstacle avoidance, ensuring stable navigation in the complex environment of the hospital. At the same time, the robot communicates with the central control system in real time through the wireless network to receive delivery tasks and update route information. Intelligent algorithms help the robot to efficiently determine the delivery order based on the urgency of each department and the shortest path planning.

[0039] For example, if a delivery robot receives a task to deliver type A blood to the emergency room, it will plan the optimal route based on the built-in map and sensor data. During the delivery process, if it encounters obstacles or emergencies, the robot will adjust the route in real time to ensure that the blood products are delivered in time.

[0040] In addition, between step S200 and step S300, the situation where there are multiple blood distribution robots can also be considered. For this situation, the task allocation principles are as follows: Urgency priority: Prioritize blood distribution tasks with high urgency to robots that are closer to the demand department and are idle. Priority to expiring: For blood products that are about to expire, regardless of the urgency of the demand department, priority is given to the robot that can deliver the fastest. Priority to temperature limit: Blood product tasks with strict temperature restrictions are assigned to robots with corresponding storage conditions, or to robots that can deliver to a suitable storage environment in the shortest time.

[0041] Reference Figure 2 , obtain blood demand information of each department of the hospital including: Step S210, obtaining the preset time step of each department of the hospital.

[0042] Specifically, the technology involved in step S210 is mainly data acquisition and processing technology. In order to obtain the preset time steps of each department of the hospital, it may be necessary to retrieve this information from a central database or configuration file. The time step refers to the time interval selected when performing time series analysis, which is crucial for the subsequent prediction model training.

[0043] For example: suppose that there is a database in the hospital's information system that is specifically used to store the operating parameters of each department. In step S210, this database will be connected first, and then the preset time step of each department will be obtained by executing a specific query statement. This time step may be in days, hours, or minutes, depending on the hospital's operational needs and the sophistication of data analysis. Once this information is obtained, it will be input into the ARIMA model of the corresponding department for subsequent predictive analysis. In this way, each department can predict future blood needs based on its own unique operating rhythm, thereby improving prediction accuracy and resource allocation efficiency.

[0044] Step S220, input the preset time step of the same department of the hospital into the ARIMA model of the corresponding department that has completed training, and generate the blood demand information predicted by the corresponding department of the hospital.

[0045] Specifically, the technologies involved in step S220 are mainly time series prediction and machine learning algorithms. The ARIMA model (autoregressive integrated moving average model) is a statistical method widely used in time series prediction. In this step, we have obtained the preset time steps of each department of the hospital through step S210. Next, we input these data into the trained ARIMA model.

[0046] For example, suppose we have trained an ARIMA model for a department in a hospital (such as surgery). The model has learned and captured the time series characteristics of blood demand in surgery through historical blood demand data. Now, we input the preset time step (for example, daily or weekly) of surgery into this model, and the model can predict the blood demand of the department in the future (such as the next week or month) based on these historical data and time steps.

[0047] Reference Figure 3 , obtain the corresponding department ARIMA model that has completed training, including: Step SA00, collect the historical blood usage data of the corresponding department and perform data preprocessing.

[0048] Among them, the historical blood usage data of each department includes but is not limited to the historical blood usage volume, blood usage type, and blood usage time of each department, and data preprocessing includes but is not limited to filling missing values, processing outliers, and smoothing noise.

[0049] Step SB00, perform parameter estimation based on the information criterion to determine the optimal order of the ARIMA model.

[0050] The techniques involved in step SB00 are mainly parameter estimation and the application of information criteria. When determining the optimal order of the ARIMA model, we need to find a balance point, that is, the best compromise between the complexity of the model and its ability to fit the data. This is usually achieved by using information criteria (such as AIC - Akaike Information Criterion, or BIC - Bayesian Information Criterion).

[0051] In step SB00, we first set different order combinations (i.e., the p, d, and q values ​​in ARIMA(p,d,q)). For each order, we use the maximum likelihood estimation method to fit the historical data and calculate the corresponding AIC or BIC value. These information criterion values ​​take into account the complexity of the model (the higher the order, the more complex the model) and the degree of fit of the model to the data (measured by the likelihood function).

[0052] After evaluating all possible orders, we will select the order combination with the smallest AIC or BIC value as the optimal order. This optimal order represents the best balance between fitting the data and avoiding excessive complexity. In this way, we can ensure that the trained ARIMA model can not only fit the historical data well, but also make accurate predictions about future changes in blood demand.

[0053] Step SC00, using the selected parameters, fit the historical data through maximum likelihood estimation to train the coefficients of the ARIMA model.

[0054] Among them, the techniques involved in step SC00 are mainly maximum likelihood estimation and model fitting. After determining the optimal order of the ARIMA model, we need to use these orders and complete and accurate historical data obtained through data preprocessing to train the coefficients of the model. Maximum likelihood estimation is a statistical method whose basic principle is to find the parameter value that is most likely to cause the current observation data to appear.

[0055] Assume that we have completed data collection and preprocessing for the internal medicine department (step SA00), and determined that ARIMA(1,1,1) is the optimal model order based on the information criterion (step SB00). Now, we proceed to step SC00, which is to use maximum likelihood estimation to fit the data and train the model coefficients.

[0056] In the specific operation, we will input the historical blood usage of the internal medicine department, blood usage type (if considering the independent prediction of different types of blood usage, it may be necessary to build a separate model for each type), and blood usage time into the ARIMA (1,1,1) model. Then, through the maximum likelihood estimation method, the autoregressive coefficient, difference order, and moving average coefficient in the model are calculated to minimize the difference between the predicted data generated by the model and the historical observation data.

[0057] Step SD00, compare the model prediction results with historical data, calculate the mean square error, and determine that the ARIMA model of the corresponding department has completed training if and only if the mean square error is less than the preset error value.

[0058] Specifically, assume that the ARIMA model training for the surgical department has been completed, and the future blood usage forecast results based on historical data have been obtained. In order to verify the accuracy of this prediction model, we will compare the model's prediction results with the actual surgical historical blood usage data on a monthly or daily basis. In the comparison process, we will use the mean square error (MSE) as an important evaluation indicator. The mean square error can help us quantify the degree of difference between the model prediction and the actual observation. The operation method is as follows: We put the model's predicted data and the actual surgical blood usage data in the same table, compare the data at the corresponding time points, calculate the prediction error at each time point, and then find the average of the square values ​​of these errors to get the MSE.

[0059] If the calculated MSE is smaller than the error threshold we set previously (this threshold is set based on the hospital's requirements for prediction accuracy and the fluctuation of historical data), then we can conclude that this ARIMA model has a high accuracy for predicting surgical blood demand, and the model training work can be considered completed.

[0060] Reference Figure 4, obtain blood demand information of each department of the hospital including: Step S110, analyzing whether the difference between the blood demand predicted by the ARIMA model and the actual blood demand of each department of the hospital within a preset time period exceeds a preset difference. If not, execute step S120; if yes, execute step S130.

[0061] Among them, step S110 mainly performs difference analysis between the blood demand predicted by the ARIMA model and the actual blood demand. Specifically, this step involves comparing the difference between the blood demand prediction value generated by the ARIMA model and the actual blood demand of each department of the hospital within a preset time period.

[0062] For example, suppose we set a weekly time period and the surgical department of the hospital is the object of analysis. First, we will obtain the daily blood demand data of the surgical department predicted by the ARIMA model in the past week. In fact, these data are the results predicted by the model based on historical data, representing how much blood the model believes the surgical department should need every day this week.

[0063] Next, we collect data on the actual daily blood demand of the surgery department over the past week. These data are real and reflect how much blood the surgery department actually used every day.

[0064] Then, the predicted daily blood demand is compared with the actual daily blood demand on a daily basis, and the daily difference is calculated. If the difference on a certain day exceeds the preset difference threshold (this threshold is set based on the hospital's tolerance for prediction errors and historical data fluctuations), it means that the ARIMA model's prediction results on that day are significantly different from the actual demand.

[0065] If the difference of all days in the entire weekly time period does not exceed the preset difference, it means that the prediction of the ARIMA model in this week is relatively accurate, and the original settings can be maintained (step S120). If the difference of any day exceeds the preset difference, it means that the prediction of the ARIMA model may not be accurate enough at some times, and further adjustment and optimization are needed.

[0066] Step S120, maintaining the original settings.

[0067] Step S130, obtaining historical prediction data of the ARIMA model, actual blood demand data, and patient diagnosis and treatment dynamic information.

[0068] Specifically, the historical forecast data of the ARIMA model is the model's forecast record of the blood demand of various departments in the hospital over the past period of time. The actual blood demand data reflects the actual blood usage of various departments in the hospital during the corresponding period of time. These data are an important basis for evaluating the accuracy of the forecast model. By comparing with the actual data, the gap between the predicted value and the actual value can be intuitively seen. Patient diagnosis and treatment dynamic information includes but is not limited to changes in the patient's condition, surgical arrangements, and treatment plan adjustment information.

[0069] For example, suppose a hospital's surgical department has a series of major surgeries scheduled in the near future, which are expected to require a large amount of blood transfusions. In this case, by collecting and analyzing patient diagnosis and treatment dynamic information, the hospital can predict in advance that the demand for blood in the surgical department will increase. At the same time, combining the historical forecast data of the ARIMA model with the actual blood demand data, the hospital can more accurately determine the growth trend of blood demand, so as to prepare and dispatch blood inventory in advance.

[0070] Step S140, the historical prediction data of the ARIMA model, the actual blood demand data and the patient diagnosis and treatment dynamic information are fused, and features closely related to blood demand are extracted through feature engineering, and the extracted features are converted into numerical forms suitable for machine learning models.

[0071] Specifically, in this step, we need to deeply integrate the historical forecast data of the ARIMA model, the actual blood demand data, and the patient diagnosis and treatment dynamic information, and extract features closely related to blood demand.

[0072] Here are some specific features that may be extracted and their examples: Trend characteristics of historical forecast data: By analyzing the ARIMA model's past forecast data, we can extract statistical features such as the mean, median, maximum, and minimum of the forecast values, which reflect the overall trend of historical forecasts.

[0073] Fluctuation characteristics of actual blood demand data: For actual blood demand data, we can calculate the standard deviation, variance, kurtosis, skewness, etc. of the demand to describe the shape and dispersion of the data distribution. These characteristics help us understand the stability and volatility of actual blood demand.

[0074] Classification features of patient diagnosis and treatment dynamic information: According to the changes in the patient's condition, we can classify the patient's status into categories such as "stable", "worsening", and "improved"; surgical arrangements can be converted into features such as the number and type of surgery (such as emergency surgery, elective surgery), and duration of surgery; adjustments to treatment plans can be converted into the frequency of adjustments, the type of adjustments (such as drug changes, dosage adjustments), etc. Time series related features: Considering that blood demand may be affected by factors such as seasons, holidays, weekdays and weekends, we can extract features related to specific time points, such as months, quarters, whether it is a weekend, whether it is a holiday, etc. Cross-features: In addition to the basic features mentioned above, we can also create new cross-features by combining different features. For example, you can calculate the "average blood demand on weekdays when the patient's condition worsens." Such features may reveal the impact of certain specific patient conditions and time factors on blood demand.

[0075] Assume that in step S140, we find that the blood demand will increase significantly when "the patient's condition deteriorates" and "the surgery is scheduled as an emergency surgery". At this time, we can add this as an important cross-feature to the dataset for training subsequent machine learning models. Such features not only help the model to more accurately predict the blood demand in a specific situation, but also help the hospital to allocate blood resources in advance.

[0076] Step S150, using the processed data set to train the gradient boosting tree model, using cross-validation technology to select the best model parameters.

[0077] The details are as follows: Dataset preparation: In step S140, we have integrated the historical forecast data of the ARIMA model, the actual blood demand data, and the patient diagnosis and treatment dynamic information, extracted the features closely related to blood demand through feature engineering, and converted these features into numerical forms suitable for machine learning models. These data will serve as the input of step S150.

[0078] Model selection: We chose to use the gradient boosted tree model for training. Gradient boosted tree is an efficient machine learning algorithm that improves prediction accuracy by building multiple weak learners (decision trees) and combining them into a strong learner. This model performs well when processing data with complex nonlinear relationships.

[0079] Training process: During the training process, we use the processed data set to train the gradient boosting tree model. Through continuous iteration and optimization, the model gradually learns the laws and patterns in the data, thereby improving the ability to predict future blood needs.

[0080] Cross-validation: In order to select the best model parameters, we use cross-validation technology. Specifically, the original data set is divided into multiple subsets, and one subset is selected each time as the validation set, and the remaining subsets are used as the training set. In this way, we can evaluate the performance of the model on different data sets and select the parameter combination that makes the model perform best.

[0081] Assume that we have prepared a dataset containing 1,000 samples, each of which contains features closely related to blood demand extracted through feature engineering. We can divide this dataset into 5 subsets, select one subset each time as the validation set, and the remaining 4 subsets as training sets.

[0082] We then train the gradient boosted tree model on each training set and evaluate the model's performance on the validation set. By comparing the performance of the model under different parameter combinations, we can select the best parameter combination. For example, we may find that when the learning rate is 0.1 and the maximum depth of the decision tree is 5, the model performs best on the validation set.

[0083] Finally, we retrain the entire dataset using this optimal parameter combination to obtain a gradient boosted tree model with excellent performance.

[0084] Step S160, using a preset weighted average method, the ARIMA model and the gradient boosting tree model are integrated to form a new prediction model.

[0085] In step S160, we use a preset weighted average method to fuse the ARIMA model and the gradient boosting tree model. This fusion method aims to combine the advantages of the two models to make up for their respective shortcomings. Specifically, we perform a weighted average on the predicted values ​​output by the two models, where the weights can be adjusted according to the actual situation.

[0086] Suppose we have a data set about the blood demand of a department in a hospital. The data set contains the actual blood demand over a period of time and the predicted values ​​given by the ARIMA model and the gradient boosting tree model respectively.

[0087] In step S160, we first determine the weights of the two models. This can be done through experiments or based on the advice of domain experts. For example, we can set the weight of the ARIMA model to 0.6 and the weight of the gradient boosting tree model to 0.4.

[0088] Next, we calculate the weighted average of the predictions of the two models. For each time point t, we have two predictions: ARIMA_pred(t) and GBRT_pred(t), which represent the predictions of the ARIMA model and the gradient boosted tree model at time point t, respectively. Then, the fused prediction value Fused_pred(t) can be calculated according to the following formula: Fused_pred(t)=0.6*ARIMA_pred(t)+0.4*GBRT_pred(t).

[0089] Step S170, obtaining blood demand information of each department of the hospital according to the new prediction model.

[0090] The new prediction model includes: Step S161, using the validation set to evaluate the ARIMA model and the gradient boosting tree model, calculating the mean square error of different models, and taking the inverse of the mean square error of each model obtained as the weight coefficient of the corresponding model.

[0091] Specifically, after obtaining the mean square error of each model, we take its reciprocal as the weight coefficient of the corresponding model. This approach is based on the following consideration: the smaller the mean square error, the better the prediction performance of the model, so we give such a model a greater weight. The reciprocal of the mean square error can just reflect the prediction accuracy of the model. The smaller the mean square error, the larger its reciprocal, that is, the larger the weight coefficient.

[0092] Assume that we have a validation set of 10 samples for evaluating the ARIMA model and the Gradient Boosted Trees model.

[0093] First, we use the ARIMA model and the gradient boosting tree model to predict these 10 samples respectively and get two sets of predicted values. Then, we compare the two sets of predicted values ​​with the actual values ​​and calculate the mean square error of the two models respectively. Assume that the mean square error of the ARIMA model is 0.8 and the mean square error of the gradient boosting tree model is 1.2. Calculate the weight coefficient according to the inverse of the mean square error. The weight coefficient of the ARIMA model is 1 / 0.8=1.25, and the weight coefficient of the gradient boosting tree model is 1 / 1.2≈0.83. In this way, we get the weight coefficients of the two models, which can be used in the subsequent weighted average fusion step. In this example, the ARIMA model has a larger weight coefficient because it has a smaller mean square error.

[0094] Step S162, according to the weight coefficients of the corresponding models obtained through analysis and the weighted average method, the prediction results of the ARIMA model and the gradient boosting tree model are integrated to form a new prediction model.

[0095] Specifically, for each prediction time point, we multiply the prediction value of the ARIMA model by its weight coefficient, and also multiply the prediction value of the gradient boosting tree model by its weight coefficient. Then, we add these two weighted prediction values ​​together to get the prediction result of the fusion model. This process can be expressed by the mathematical formula: Fusion prediction value = ARIMA prediction value * ARIMA weight coefficient + gradient boosting tree prediction value * gradient boosting tree weight coefficient.

[0096] In this way, we can take advantage of the strengths of both models and obtain a more robust and accurate prediction result.

[0097] Assume that at a certain prediction time point, the prediction value of the ARIMA model is 100, and its weight coefficient is 0.6; the prediction value of the gradient boosting tree model is 120, and its weight coefficient is 0.4. Then, according to the fusion method of step S162, we can calculate the prediction value of the fusion model as: 100*0.6+120*0.4=60+48=108.

[0098] According to the blood demand information of each department of the hospital, the blood packaging robot is started to package the blood according to the blood demand information to form blood products including: Step S210, analyzing whether there is any special blood product demand information proactively issued by various departments of the hospital.

[0099] In step S210, the system will actively retrieve and analyze the special blood product demand information from various departments of the hospital. This type of information is usually generated due to special cases or emergency needs of the department, and may involve blood of a specific blood type, blood products with added specific drugs, etc.

[0100] The key to this step is to accurately identify and classify the demand information. Once a special demand from a department is detected, the system will further analyze the details of the demand, such as the specific type, specification, quantity and required delivery time of the blood products required.

[0101] Suppose a hospital's emergency department receives a patient who is seriously injured in a car accident. The patient needs a large amount of O-type Rh-negative blood (i.e. "panda blood") and needs a blood transfusion as soon as possible.

[0102] In this case, the emergency department will send a special blood product requirement message through the hospital's internal information system, clearly indicating the blood type, quantity and urgency of the required blood.

[0103] When this demand information is captured by the system, step S210 is triggered. The system automatically analyzes this information, identifies it as a demand for special blood products, and extracts key data, such as blood type (O type Rh negative), quantity (e.g.: 1000 ml) and urgency of the demand (e.g.: immediately).

[0104] Then, based on this information, the system will give priority to arranging blood packaging robots to package blood according to this special demand, ensuring that the required blood products can be delivered to the emergency department in a timely and accurate manner to meet the patient's blood transfusion needs.

[0105] Step S220: If yes, then according to the special blood product demand information, the blood packaging robot is started to package the blood to form special blood products first according to the special blood product demand information, and then continues to package the blood to form blood products according to the blood demand information.

[0106] In step S220, once the system confirms that there is special blood product demand information proactively sent by various departments of the hospital, it will immediately trigger a priority processing mechanism, which ensures that the blood packaging robot can prioritize the blood packaging work according to these special demand information.

[0107] Here, "special blood product demand information" may include but is not limited to: urgent demand for specific blood types, demand for blood products of specific volumes, demand for blood products with added specific drugs or nutrients, etc. These special demands are usually different from ordinary blood demands and require more precise and rapid processing.

[0108] To meet these special needs, the blood packaging robot will be programmed to identify and prioritize these tasks. The robot will obtain detailed parameters in the demand information, such as the type, specification, quantity and priority of the required blood products, and then start the blood packaging work according to this information.

[0109] In addition, the system will ensure that after giving priority to special blood product needs, the blood packaging robot can continue to package blood according to regular blood demand information to maintain the efficient operation of the overall process.

[0110] For example: Suppose a hospital's surgical operating room suddenly needs a large amount of type A Rh positive blood for an important operation. The operating room sends a special blood product request message containing detailed information through the internal system.

[0111] After this information is captured and confirmed by the system, step S220 is activated immediately. The blood packaging robot receives this special demand information and immediately starts to prioritize this task. According to the instructions in the information, the robot quickly and accurately selects the type A Rh positive blood that meets the requirements from the inventory and performs the necessary inspection and packaging processes.

[0112] After completing the packaging of this batch of special-need blood products, the robot will return to its regular blood packaging tasks to ensure that the normal blood needs of other departments can also be met.

[0113] Step S230, if no, remains unchanged.

[0114] Special blood demand information proactively issued by various departments of the hospital include: Step S211, obtaining patient disease-related information uploaded to the system by various departments of the hospital, the disease-related information including specific disease, personal blood transfusion allergy history, age and physical condition information.

[0115] In step S211, each department of the hospital needs to upload the patient's disease-related information through the internal system. This information not only includes the patient's specific disease name, but also covers key data such as personal blood transfusion allergy history, age, and current physical condition.

[0116] Suppose a hospital's internal medicine department admits a 45-year-old patient with leukemia who has a history of mild allergic reactions to blood transfusions.

[0117] According to the requirements of step S211, the medical staff will first enter the patient's specific disease "leukemia" into the system. Then, they will record the patient's personal history of blood transfusion allergy in detail, including the type and severity of the allergic reaction and the blood components that caused the allergy. In addition, the medical staff will also enter the patient's age "45 years old" and a description of the current physical condition, such as whether there are other complications, mental condition, vital signs, etc.

[0118] Step S212, analyzing whether the patient disease-related information uploaded to the system by various departments of the hospital contains specific diseases or personal blood transfusion allergy history, resulting in allergies to some blood components.

[0119] The core task of step S212 is to conduct an in-depth analysis of the patient disease-related information uploaded by various departments of the hospital. The main purpose of this step is to identify whether there is a specific disease, or whether the patient has a personal history of allergy to blood transfusion, especially allergies to certain blood components.

[0120] To achieve this goal, this step may use advanced technologies such as natural language processing and pattern recognition to intelligently analyze and screen the uploaded information. Based on preset rules and algorithms, the system will automatically detect whether the disease information contains preset specific disease keywords, or whether the allergy history section mentions allergic reactions to specific blood components. Suppose the pediatric department of a hospital uploads disease-related information about a 5-year-old child, including a diagnosis of "acute lymphocytic leukemia" and "allergy to plasma proteins" during past blood transfusions. In step S212, the system will analyze this information in detail. First, the system will identify the specific disease of "acute lymphocytic leukemia", which is a disease that requires frequent blood transfusions. Secondly, the system will detect that the patient has a history of transfusion allergy of "allergy to plasma proteins".

[0121] Based on these analysis results, step S212 will determine that the patient's condition meets the screening criteria for specific conditions, thereby triggering the subsequent process, that is, inputting these key information into the special blood product prediction model to generate more accurate blood demand information.

[0122] Step S213, if yes, the patient's disease related information is used as input information, input into the trained special blood product prediction model, and the special blood demand information is output as the special blood demand information actively issued by various departments of the hospital.

[0123] In step S213, once the analysis result of step S212 shows that the patient's disease-related information contains a specific disease or a personal transfusion allergy history, the system will execute this step. The main task of this step is to import these key patient disease-related information as input into a special blood product prediction model that has been trained.

[0124] This prediction model is trained by machine learning algorithms using multi-dimensional data such as hospital transfusion history, patient disease information, and blood product use effects. It can intelligently predict the specific blood product needs that best suit the patient based on the input patient's specific disease, personal transfusion allergy history, age, and physical condition information. The model uses a variety of algorithm fusion methods, including but not limited to decision trees, support vector machines, neural networks, etc., to improve prediction accuracy. At the same time, the model also has self-optimization capabilities and can continuously learn and optimize prediction strategies during actual use.

[0125] The special blood demand information output by the prediction model will serve as the special blood demand information proactively issued by various departments of the hospital, directly affecting the subsequent blood collection, processing, and distribution processes.

[0126] Suppose a hospital receives a car accident victim who needs an emergency blood transfusion. The victim is diagnosed with a rare coagulation disorder and an allergy to a specific blood component.

[0127] In step S212, the system has identified this special situation. Therefore, in step S213, the system automatically inputs the patient's disease information, including the diagnosis of coagulation disorder and allergy to specific blood components, into the special blood product prediction model.

[0128] Based on this information, the prediction model quickly generates a specific blood demand information, which clearly requires that the required blood must be specially processed, not contain the blood components that the injured person is allergic to, and must be prepared in the shortest possible time for emergency use.

[0129] Step S214: if no, remain unchanged.

[0130] Barcode or RFID labeling of blood products includes: Step S2A0, analyzing whether the blood product is a special blood product.

[0131] The core task of step S2A0 is to determine whether the blood product is a special blood product. When executing this step, the system or operator will conduct a detailed inspection and analysis of each blood product according to pre-set standards. These standards may include but are not limited to blood type, treatment method, purpose, and whether it contains specific components. For example, blood of certain rare blood types, blood products that have been specially treated, or blood used for specific medical procedures may be considered special blood products.

[0132] In order to make this judgment accurately, automated equipment may be used to assist, such as barcode scanners, RFID readers, etc. These devices can quickly identify the label information on blood products and compare it with the preset standards in the system.

[0133] Step S2B0, if yes, then the patient's disease-related information is used as input information, input into the urgency analysis model, and the urgency of the patients served by the special blood products is output.

[0134] The core of step S2B0 is to use the urgency analysis model to evaluate the urgency of patients served by special blood products. This model is an algorithm based on machine learning, which receives the patient's disease-related information as input and then predicts the patient's urgency based on this information.

[0135] In step S2B0, we first collect and organize relevant information about the patient's disease, including but not limited to the type and severity of the disease, the patient's age, gender, physical condition, and past medical history, etc. Then, we use this information as input data and input it into the urgency analysis model.

[0136] The urgency analysis model will quantify the urgency of each patient based on the input disease-related information combined with its internal algorithms and logic. The scoring criteria may include factors such as the severity of the disease and the treatment response time. Ultimately, the model will output an urgency score to indicate the urgency of the patient's need for special blood products.

[0137] Suppose a patient who was seriously injured in a car accident was sent to the hospital. After examination, the patient needed Rh negative AB blood products for transfusion. Since Rh negative AB blood is relatively rare, this type of blood product is considered a special blood product.

[0138] In step S2A0, the system will identify the blood product as a special blood product. Then, in step S2B0, the system will input the patient's relevant information into the urgency analysis model. This information may include: multiple severe fractures and internal organ injuries caused by the car accident, age 30, no previous medical history, etc.

[0139] The urgency analysis model will use this information and its internal algorithm to score the patient's urgency. Assuming the model outputs a score of 90 (the score range is 1-100, with higher scores indicating higher urgency), this score means that the patient's need for Rh-negative AB blood is very urgent.

[0140] Step S2C0, affix a barcode or RFID tag to the blood product and add urgency information, and mark a special identification pattern.

[0141] The core task of step S2C0 is to attach barcodes or RFID tags to the packages of blood products that have been identified as special products, and add the urgency information analyzed in the previous step to these tags. This is done to facilitate medical personnel to quickly understand the special characteristics and urgency of blood products, thereby ensuring that patients who need it most can receive timely treatment.

[0142] When performing this step, the operator will use professional printing equipment to print out barcodes or RFID labels and stick them on the packaging of blood products. At the same time, these labels not only contain basic blood product information, but also add a special logo pattern, which is designed to remind medical personnel of the special nature of the blood product and the precautions when using it.

[0143] Urgency information is also encoded into the label, usually in the form of color coding, numerical ratings, or other easily identifiable forms, so that when medical personnel scan the label, they can quickly obtain information about the urgency of the patient served by the blood product.

[0144] Step S2D0, if no, maintain the original setting.

[0145] The blood distribution robot is activated to store blood products and distribute blood to various departments of the hospital along the preset distribution route, including: Step S310, start the blood distribution robot to identify the barcode or RFID tag on the packaged blood products, give priority to assembling blood products with urgency information, and reserve the remaining blood products according to the preset department blood product reserve priority order based on the remaining reserve space of the blood distribution robot.

[0146] Specifically, the core of step S310 is that the blood distribution robot determines the priority and order of loading by identifying the barcode or RFID tag on the packaged blood products. First, the robot will give priority to assembling blood products with urgency information to ensure that the most urgently needed blood can be delivered quickly. Secondly, the robot will load the remaining blood products according to its own remaining reserve space and the preset departmental blood product reserve priority order. In this process, the blood distribution robot will use its equipped barcode scanner or RFID reader to quickly and accurately read the label information on each blood product. This information not only includes basic information such as the type and quantity of blood, but also key data such as urgency. The robot processes and analyzes the read data through the built-in algorithm to determine the priority and order of loading.

[0147] Assume that a blood delivery robot in a hospital is about to start its delivery mission. In step S310, it first identifies a blood product of type O with urgency information, which is urgently needed for the operating room. Therefore, the robot immediately puts the blood into its top priority delivery position.

[0148] Next, the robot will continue to scan and identify other blood products. If its remaining storage space allows, it will load the blood needed by other departments in sequence according to the preset departmental blood product reserve priority order. For example, if the pediatric department has a high demand for type A blood, the robot will prioritize the type A blood needed by the pediatric department after loading the emergency blood.

[0149] Step S320, start the blood distribution robot to sort the distribution order of departments from high to low according to the urgency of different departments and distribute blood products. The distribution order of the remaining departments is determined by the shortest path planning based on the last distribution location of the blood distribution robot.

[0150] In step S320, the blood delivery robot will develop an optimized delivery route based on the information collected and analyzed in the previous steps. First, the robot will identify and sort the departments with urgency information to ensure that the first department to be delivered is the department with the most urgent need. The urgency information may come from the blood product label identified in step S310, or the emergency demand data updated in real time by the hospital system.

[0151] The robot will sort the departments according to their urgency from high to low, giving priority to departments with high urgency. After completing the emergency delivery task, the robot will re-evaluate the needs of the remaining departments and use the shortest path planning algorithm to determine the delivery order of the remaining departments, starting from its last delivery location. This can minimize the robot's moving distance and time, and improve delivery efficiency.

[0152] Assume that a hospital has five departments that need blood distribution, namely: Department A (high urgency), Department B (medium urgency), Department C (low urgency), Department D (medium urgency), and Department E (low urgency). In step S320, the blood distribution robot will first identify that Department A has the highest urgency, so it will put Department A at the top of the distribution route. Next, the robot will evaluate that Department B and Department D have the same urgency, but because Department D is closer to the current location of the robot, Department D will be ranked before Department B. Finally, Department C and Department E have the lowest urgency, and they will be ranked at the end of the distribution route. Based on the above analysis, the robot will develop the following distribution route: Department A → Department D → Department B → Department C / Department E (select the order according to the actual situation). In this way, the robot can ensure that the needs of the department with the highest urgency are met first, and at the same time complete the distribution tasks of the remaining departments with the best path.

[0153] Among them, starting the blood distribution robot to sort the distribution order of departments from high to low according to the urgency of different departments and distribute blood products includes: analyzing whether the distributed blood products are within the preset expiration time range; if yes, the urgency of the blood products within the preset expiration time range will be increased, and the urgency of the department is defined as high, medium and low. The blood demand of the high-urgency department has the highest priority; if no, analyze whether the distributed blood products have temperature restrictions; if yes, distribute the temperature-restricted blood products to the location area that meets the conditions for storing the corresponding blood products. For example, blood products that need to be refrigerated are centrally distributed to the department area with cold storage to reduce the risk of temperature changes.

[0154] In addition, to supplement, a plan is also included that is carried out in parallel with increasing the urgency of blood products within a preset expiration time range, as follows: obtaining the location of the department where the blood products within the preset expiration time range are used, and sending the location of the blood distribution robot to the terminal held by the corresponding personnel of the corresponding department, and based on the location of the corresponding personnel of the corresponding department and the current route of the blood distribution robot, planning a planned route for the corresponding personnel of the corresponding department to encounter the blood distribution robot.

[0155] The planning of the route for the corresponding personnel of the corresponding department to encounter the blood distribution robot includes the following steps, which are as follows: Step 1: Obtain the current location information of the corresponding personnel of the corresponding department, including the floor, area, specific room number or coordinates; at the same time, obtain the real-time location, direction of travel and the remaining preset distribution routes of the blood distribution robot; Step 2: Analyze the internal structure and traffic rules of the hospital, including the distribution and use restrictions of corridors, stairs, and elevators, and determine the feasible movement paths of the corresponding personnel and robots of the corresponding departments; Step 3: For each possible intersection in the hospital, such as the intersection of corridors, elevator entrances, etc., calculate the time required for the corresponding personnel of the corresponding department to reach the intersection from the current position, and the time required for the robot to reach the same intersection from the current position, and consider the normal walking speed of the corresponding personnel of the corresponding department and the moving speed of the robot when calculating. Step 4: Compare the arrival time of both parties at all possible intersections, and select the intersection with the smallest sum of the arrival time of both parties as the best meeting point. Step 5: Plan the shortest and most traffic-compliant route for the corresponding personnel in the corresponding department from their current location to the optimal meeting point, and adjust the robot's preset delivery route so that it can meet the corresponding personnel in the corresponding department at the optimal meeting point.

[0156] For example, suppose that the medical staff is in Ward 502 on the 5th floor of the hospital, and the blood delivery robot is in the elevator from the 4th floor to the 6th floor and is about to reach the 5th floor. Through calculation, it takes 2 minutes for the medical staff to reach the elevator entrance from Ward 502 through the corridor, and it takes 1 minute for the robot to reach the elevator entrance on the 5th floor from its current position. The elevator entrance on the 5th floor is selected as the best meeting point, and a route is planned for the medical staff from Ward 502 through the corridor to the elevator entrance. The robot stops at the elevator entrance on the 5th floor to wait for the medical staff.

[0157] For example, if the medical staff is at the south end of the hospital corridor and the robot is at the north end of the corridor, calculate the time it takes for both parties to reach the meeting point in the middle of the corridor. If it takes 3 minutes for the medical staff to arrive and 2 minutes for the robot to arrive, then choose the middle of the corridor as the best meeting point. Plan a route for the medical staff to walk north along the corridor, and the robot continues to move north to the meeting point.

[0158] Based on the same inventive concept, an embodiment of the present invention provides an intelligent blood management system based on multi-dimensional data, including a memory and a processor, wherein the memory stores data that can be executed on the processor to implement the following Figures 1 to 4 Procedure for either method.

[0159] The embodiments of this specific implementation method are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Therefore, all equivalent changes made based on the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. An intelligent blood product management method based on multi-dimensional data, characterized in that: include: Obtain blood demand information of each department in the hospital, including blood type categories and demand for different blood types; According to the blood demand information of each department of the hospital, the blood packaging robot is started to package the blood into blood products according to the blood demand information, and a barcode or RFID tag is attached to the blood products; Start the blood distribution robot to store blood products and distribute blood to various departments of the hospital along the preset distribution route; Obtain blood demand information for each department of the hospital, including: Get the preset time steps of each department in the hospital; Input the preset time step of the same department of the hospital into the trained ARIMA model of the corresponding department to generate the predicted blood demand information of the corresponding department of the hospital; Obtaining the ARIMA model of the corresponding department that has completed training includes: Collect the historical blood usage data of the corresponding departments and perform data preprocessing, wherein the historical blood usage data of each department includes but is not limited to the historical blood usage volume, blood usage type, and blood usage time of each department, and the data preprocessing includes but is not limited to filling missing values, processing outliers, and smoothing noise; Parameter estimation based on information criteria to determine the optimal order of the ARIMA model; Using the selected parameters, the historical data is fitted by maximum likelihood estimation to train the coefficients of the ARIMA model; Compare the model prediction results with historical data and calculate the mean square error. If and only if the mean square error is less than the preset error value, the ARIMA model training of the corresponding department is determined to have completed.

2. The intelligent blood product management method based on multi-dimensional data according to claim 1, characterized in that: Obtain blood demand information for each department of the hospital, including: Analyze whether the difference between the blood demand predicted by the ARIMA model and the actual blood demand of each department of the hospital within the preset time period exceeds the preset difference; If not, maintain the original setting; If yes, then obtain the historical forecast data of the ARIMA model, the actual blood demand data, and the patient's diagnosis and treatment dynamics information, including but not limited to the patient's condition changes, surgery arrangements, and treatment plan adjustment information; The historical forecast data of the ARIMA model, the actual blood demand data, and the patient diagnosis and treatment dynamic information are integrated, and features closely related to blood demand are extracted through feature engineering, and the extracted features are converted into numerical forms suitable for machine learning models; Use the processed data set to train the gradient boosting tree model and use cross-validation techniques to select the best model parameters; Using the preset weighted average method, the ARIMA model and the gradient boosting tree model are integrated to form a new prediction model; According to the new prediction model, the blood demand information of each department in the hospital is obtained.

3. The intelligent blood product management method based on multi-dimensional data according to claim 2 is characterized in that: The new prediction model includes: Use the validation set to evaluate the ARIMA model and the gradient boosting tree model, calculate the mean square error of different models, and use the inverse of the mean square error of each model as the weight coefficient of the corresponding model; According to the weight coefficients of the corresponding models obtained through analysis and the weighted average method, the prediction results of the ARIMA model and the gradient boosting tree model are fused to form a new prediction model.

4. The intelligent blood product management method based on multi-dimensional data according to claim 3 is characterized in that: According to the blood demand information of each department of the hospital, the blood packaging robot is started to package the blood according to the blood demand information to form blood products including: Analyze whether there is any special blood product demand information proactively issued by various departments of the hospital; If yes, then according to the special blood product demand information, the blood packaging robot is started to package the blood to form special blood products first according to the special blood product demand information, and then continues to package the blood to form blood products according to the blood demand information; If no, remain unchanged.

5. The intelligent blood product management method based on multi-dimensional data according to claim 4 is characterized in that: Special blood demand information proactively issued by various departments of the hospital include: Obtain patient disease-related information uploaded to the system by various departments of the hospital, including specific diseases, personal blood transfusion allergy history, age and physical condition information; Analyze whether the patient disease-related information uploaded to the system by various departments of the hospital contains specific diseases or personal blood transfusion allergy history, and whether there is allergy to some blood components; If yes, the patient's disease-related information is used as input information to the trained special blood product prediction model, and the special blood demand information is output as the special blood demand information proactively issued by various departments of the hospital; If no, remain unchanged.

6. The intelligent blood product management method based on multi-dimensional data according to claim 5, characterized in that: Barcode or RFID labeling of blood products includes: Analyze whether blood products are special blood products; If yes, then the patient's disease-related information is used as input information to the urgency analysis model, and the urgency of the patient served by the special blood product is output; Attach barcodes or RFID tags to blood products and add urgency information, as well as special identification patterns; If not, keep the original setting.

7. The intelligent blood product management method based on multi-dimensional data according to any one of claims 1 to 6, characterized in that: The blood distribution robot is activated to store blood products and distribute blood to various departments of the hospital along the preset distribution route, including: Start the blood distribution robot to identify the barcode or RFID tag on the packaged blood products, give priority to assembling the blood products with urgency information, and reserve the remaining blood products in the preset department blood product reserve priority order according to the remaining reserve space of the blood distribution robot; The blood distribution robot is started to sort the distribution order of departments from high to low according to the urgency of different departments and distribute blood products. The distribution order of the remaining departments is determined by the shortest path planning based on the last distribution location of the blood distribution robot.

8. An intelligent blood management system based on multi-dimensional data, characterized in that: It comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein the program can be loaded and executed by the processor to implement an intelligent blood product management method based on multi-dimensional data as described in any one of claims 1 to 7.

Images