A method and system for outpatient resource prediction based on multi-objective tasks

Abstract

The present invention belongs to the field of information processing technology, and discloses an outpatient resource prediction method and system based on multi-objective tasks, including: collecting and organizing outpatient medical treatment data, performing data preprocessing on the data to generate an initial outpatient treatment information data set; performing data analysis and feature extraction on the initial outpatient treatment information data set, constructing and training an outpatient resource prediction model based on multi-objective task learning; and continuously iterating and evaluating the outpatient resource prediction model; inputting time range conditions and custom target conditions, applying the outpatient resource prediction model to predict the input conditions and output the results, and formulating an outpatient scheduling plan. The present invention integrates cutting-edge time series analysis algorithms and multi-objective task deep learning technology, and can accurately predict the outpatient resource demand in each time period, optimize the outpatient scheduling plan in real time, realize the scientific configuration and reasonable allocation of resources, effectively reduce resource idleness and waste, and thus significantly improve the utilization efficiency of medical resources.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a method and system for outpatient resource prediction based on multi-objective tasks. Background Art

[0002] In the modern medical system, outpatient resource forecasting is an extremely complex and challenging task. Traditional outpatient resource forecasting mainly relies on historical single registration data. This single data source analysis method cannot effectively respond to the dynamic changes in outpatient demand. Some use information technology to comprehensively consider the impact of seasonal changes, public emergencies, and regional epidemic characteristics, but outpatient resources are also closely related to a variety of actual in-hospital medical factors such as hospital space layout, patient disease characteristics, patient treatment habits, clinical pathways for disease diagnosis, and physical time during the diagnosis and treatment process. Due to the frequent changes and complex interactions of these factors, traditional forecasting methods based on historical registration data often find it difficult to capture the fluctuations in actual demand, resulting in inaccurate resource allocation, which further affects the utilization efficiency and service quality of medical resources. In addition, there are significant differences in outpatient resources in different scenarios, such as the number of daily visits to different departments, the proportion of patients undergoing examinations and tests, and the distribution of doctors' workload. There is a close relationship between these target tasks, but traditional forecasting methods usually find it difficult to effectively integrate multi-dimensional information, so there are obvious deficiencies in the accuracy of the overall outpatient resource forecast results and the practicality of resource allocation. Furthermore, the complexity and diversity of outpatient data make resource forecasting more challenging. Outpatient data includes not only basic patient information and medical records, but also emergency data, regional disease data, patient satisfaction survey results, etc. These data come from a wide range of sources and are of various types. Traditional single-task prediction models have limited performance in processing these complex and changeable data, making it difficult to achieve comprehensive and accurate predictions of outpatient resources.

[0003] In summary, the existing outpatient resource prediction methods cannot fully meet the needs of accurate prediction and resource optimization in modern medical services. These traditional methods lack the ability to comprehensively analyze multidimensional data and the comprehensive consideration of the correlation between prediction tasks, resulting in the lack of accuracy and practical application of the prediction results. To this end, a new outpatient resource prediction method is needed that can comprehensively process multiple data sources, dynamically adapt to changes, and simultaneously solve the prediction needs of multi-objective tasks, thereby providing more accurate and reliable outpatient resource prediction results, and providing strong support for the efficient management and optimal allocation of medical resources. Summary of the invention

[0004] In view of this, the present invention provides an outpatient resource prediction method and system based on multi-objective tasks to solve the above problems.

[0005] In order to solve the above technical problems, the present invention provides an outpatient resource prediction method based on multi-objective tasks, comprising:

[0006] Collect and organize outpatient medical visit data, and perform data preprocessing to generate an initial outpatient visit information dataset;

[0007] Perform data analysis and feature extraction on the initial outpatient visit information data set, build and train an outpatient resource prediction model based on multi-objective task learning; and continuously iterate and evaluate the outpatient resource prediction model until it meets the preset training completion conditions;

[0008] Input the time range conditions and custom target conditions, apply the outpatient resource prediction model to predict the input conditions and output the results, and formulate an outpatient scheduling plan.

[0009] As an option, outpatient medical visit data include:

[0010] Historical outpatient registration and treatment data, historical patient treatment data, patient treatment process data, patient satisfaction survey data, medical staff satisfaction survey data, emergency event data, and regional and local disease data.

[0011] As an optional method, the outpatient resource prediction model includes an input layer, a shared layer, a specific layer and an output layer; the output end of the input layer is connected to the input end of the shared layer, and the shared layer is used for data feature extraction; the output end of the shared layer is connected to the input end of the specific layer, and the specific layer is used for accurate prediction of multi-objective tasks; the output end of the specific layer is connected to the output layer.

[0012] As an optional approach, data with temporal and spatial dependencies are processed in a shared layer to perform general feature extraction, which includes:

[0013] Convert the input data into tensor form;

[0014] The data updates are fused into the memory unit of the outpatient resource prediction model using input gate control regulation;

[0015] Use the forget gate to control the proportion of memory information from the previous time step that is retained in the memory unit of the current time step;

[0016] Use candidate memory cells and memory cell updates to capture long-term dependencies in the time series data of medical visit data and output intermediate candidate memory cell information.

[0017] As an optional method, convert the input data into tensor form, which includes:

[0018] Convert the input data into a tensor form X=(T,H,W,C), where T represents the number of steps or time points in the time dimension; H represents the data height, indicating the categories of various outpatient resources; W represents the data width, indicating the number of outpatient resources under various categories; C represents the number of feature channels; and use ConvLSTM to extract the temporal and spatial features of the data.

[0019] As an optional method, the input gate control regulation is used to fuse the data update into the memory unit of the outpatient resource prediction model, including:

[0020] ,in is the degree to which the output of the input gate controls the data introduced, represents the sigmoid function, and Represent the weight matrices of the current input and the previous hidden state, respectively. represents the hidden state of the previous step, is the bias term.

[0021] As an optional method, a forget gate is used to control the proportion of memory information of the previous time step retained in the memory unit of the current time step, including:

[0022] ,in and Represent the weight matrices of the current input and the previous hidden state of the forget gate, respectively. represents the sigmoid function, is the bias term.

[0023] As an optional method, candidate memory cells and memory cell updates are used to capture long-term dependencies in the time series data of medical visit data, and intermediate candidate memory cell information is output, including:

[0024] Output intermediate candidate memory unit information ,in represents the weight matrix of the candidate memory unit, It represents the concatenation of the hidden state of the previous time step and the input of the current time step. Represents the bias term of the candidate memory unit;

[0025] Continue to update candidate memory cells ,in Represents the state of the memory cell at the previous time step; uses the output gate value Controls the hidden state of the current time step Updates, including represents the weight matrix of the output gate, It represents the concatenation of the hidden state of the previous time step and the input of the current time step. Represents the bias term of the output gate;

[0026] Output the new data feature tensor Y=(T,H,W,F), where F represents the number of feature channels. The new tensor at this time is a set of high-level training feature data processed by the model.

[0027] As an optional method, in a specific layer, a multi-objective task prediction model is established to optimize the prediction results of multiple prediction tasks at the same time, including:

[0028] Set N target prediction result item tasks, 0<i<N, i represents any target prediction result item;

[0029] Setting output , calculate the loss error , evaluate the gap between the predicted results and the actual values, and optimize the model to minimize the error;

[0030] in represents the number of samples in the i-th prediction task, represents the predicted value of the jth sample, represents the actual value of the jth sample;

[0031] Weighted summation of the loss errors of all prediction tasks , used to balance the impact of different tasks on resource prediction, where represents the weight coefficient of the i-th task;

[0032] Further Use The optimization model is iterated, where Indicates that the learning rate is used to control the step size of model parameter updates. Represents L for model parameters gradient.

[0033] On the other hand, the present invention also provides an outpatient resource prediction system based on multi-objective tasks, comprising:

[0034] A data processing module, used to collect and organize outpatient medical treatment data, and perform data preprocessing on it to generate an initial outpatient treatment information data set;

[0035] The model training module is used to perform data analysis and feature extraction on the initial outpatient visit information data set, build and train an outpatient resource prediction model based on multi-objective task learning, and continuously iterate and evaluate the outpatient resource prediction model until it meets the preset training completion conditions;

[0036] The resource prediction module is used to input time range conditions and custom target conditions, apply the outpatient resource prediction model to predict the input conditions and output the results, and formulate an outpatient scheduling plan.

[0037] The beneficial effects of the present invention are:

[0038] The present invention proposes an outpatient resource prediction method based on multi-objective task learning, which integrates cutting-edge time series analysis algorithms and multi-objective task deep learning technology to construct a comprehensive and sophisticated data analysis system. By extracting common features from historical outpatient registration data and dynamic medical process data, the system can accurately predict the outpatient resource demand in each time period, including department reception load, doctor workload, resource consumption of medical technology departments, and outpatient medical resource consumption. Based on these prediction results, the system can optimize the outpatient scheduling plan in real time, realize the scientific configuration and reasonable allocation of resources, effectively reduce idle resources and waste, and thus significantly improve the utilization efficiency of medical resources. In addition, by reducing patients' waiting time and optimizing the medical process, this method can comprehensively improve the quality of medical services and patient satisfaction, and provide strong support for medical institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of the outpatient resource prediction process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with specific implementation methods.

[0041] This embodiment aims to provide an outpatient resource prediction method based on multi-objective tasks, using a multi-objective task deep reinforcement learning algorithm to comprehensively analyze historical outpatient data, patient visit characteristics, and spatiotemporal distribution characteristics, and accurately predict the outpatient resource situation of various departments, specialties, doctors, and other links in different time periods. Through this method, the outpatient scheduling plan and registration number source configuration can be dynamically optimized to achieve a scientific and reasonable allocation of the working hours and work intensity of departments and doctors, thereby significantly improving the utilization efficiency of medical resources, reducing patients' waiting time, and improving the overall quality of medical services and patient satisfaction.

[0042] To this end, this embodiment adopts the following method to achieve:

[0043] See also Figure 1 ,This embodiment provides an outpatient resource prediction method based on multi-objective tasks, which can be divided into three steps:

[0044] 1. Collect and organize outpatient medical treatment data, and pre-process it to generate an initial outpatient treatment information data set;

[0045] 2. Perform data analysis and feature extraction on the initial outpatient visit information data set, build and train an outpatient resource prediction model based on multi-objective task learning; and continuously iterate and evaluate the outpatient resource prediction model until it meets the preset training completion conditions;

[0046] 3. Input the time range conditions and custom target conditions, apply the outpatient resource prediction model to predict the input conditions and output the results, and formulate an outpatient scheduling plan.

[0047] Specifically, collect and organize data related to outpatient medical visits, including historical outpatient registration data, historical patient visit data, patient visit process data, patient satisfaction survey data, medical staff satisfaction survey data, emergency event data, regional and local disease data, and other multi-source medical data to ensure that all key information required for outpatient resource forecasting is covered.

[0048] Among them, outpatient visit data is a comprehensive multi-dimensional data generated in the process of hospital outpatient services. The value and significance of each data are precious, which can fully understand all aspects of outpatient services. Comprehensive data can help identify and analyze complex problems in outpatient services and help further study patients' needs and behavior patterns. General information technology processing methods only process historical outpatient registration data, and do not take into account the impact of various complex scenarios, process details, each link, link operation and other aspects of outpatient services on the overall outpatient service. This is also where hospital outpatient management should conduct in-depth research and focus on problem optimization. This solution design explores ways to optimize hospital business resource management from comprehensive and rich data, thereby improving the utilization efficiency of hospital outpatient business resources. Medical visit data includes historical outpatient registration visit data, historical patient visit data, patient visit process data, patient satisfaction survey data, medical staff satisfaction survey data, emergency data, regional regional disease data and other medical related data.

[0049] In one scenario, the historical outpatient registration and treatment data includes: the total number of historical outpatient registrations, the total number of historical registrations of each specialty, and the total number of historical registrations of each doctor - the total number of registrations for each item is further subdivided multiple times, including by registration category (such as general outpatient clinic / specialist clinic / emergency clinic / special outpatient clinic / health clinic / physical examination, etc.), by time period (such as year / quarter / month / week / day / shift), by registration method (such as on-site registration / appointment registration / WeChat registration / Alipay registration / website registration / self-service machine registration, etc.), by professional doctor title (such as chief physician / deputy chief physician / attending physician, etc.), by gender, by age, by occupation, by marital status, by disease diagnosis type, by medical The total number of registration data can be subdivided multiple times by insurance type, socioeconomic status, residential area, etc.; the number of people who visited the outpatient clinic after historical registration, and the number of people who did not visit the outpatient clinic after historical registration - these two items of data are further differentiated multiple times, including by registration category (such as general outpatient clinic / specialist outpatient clinic / emergency / special outpatient clinic / health clinic / physical examination, etc.), by time period (such as year / quarter / month / week / day / shift), by registration method (such as on-site registration / appointment registration / WeChat registration / Alipay registration / website registration / self-service machine registration, etc.), by professional doctor title (such as chief physician / deputy chief physician / attending physician, etc.), etc.

[0050] Historical patient visit data: All data related to patient characteristics of historical outpatient visits are collected based on historical outpatient registration visit data. Including visit type (initial visit / return visit), gender, age, occupation, marital status, patient destination (such as returning home to rest / admission to hospital / outpatient diagnosis and treatment / multidisciplinary consultation / consultation / transfer to hospital / emergency observation / outpatient observation / others, etc.), patient region, patient diagnosed disease type, patient drug treatment (including whether medication is prescribed / the number of medications prescribed, etc.), patient examination and treatment (including whether examination and treatment / items for examination and treatment / the number of examination and treatment items), patient test and treatment (including whether examination and treatment / items for examination and treatment / the number of test and treatment items), patient surgical treatment (including whether surgical treatment / items for surgical treatment / the number of surgical treatment items), patient general treatment (including whether general treatment / items for general treatment / the number of general treatment items), test results, examination results, etc.

[0051] Patient treatment process data: Based on historical outpatient registration and outpatient patient treatment data, all data during the patient's treatment process are collected. The following are examples, including the time a patient reports to the doctor, the time from the time the patient reports to the first time the doctor calls the patient, the time the doctor first sees the patient, the time the doctor finishes treating the patient for the first time, the time the patient reports to the examination department, the time the patient reports to the laboratory department, the time the examination report is issued, the time the laboratory report is issued, the time the patient reports to the second queue, the time the doctor calls the patient for the second queue, the time the doctor sees the patient for the second time, the time the doctor finishes treating the patient for the second time, the time the patient's treatment is officially over, etc., including no limit to the number of times such data is included.

[0052] The collected outpatient medical treatment data is subjected to comprehensive data preprocessing, including data cleaning, data transformation, data normalization, data classification and other operations, to generate a structured initial outpatient treatment information data set. This embodiment does not impose any limitation on this.

[0053] In step 2 of this embodiment, an outpatient resource prediction model based on multi-objective task learning is constructed, including an input layer, a shared layer, a specific layer, and an output layer. The input layer serves as the input end of the model, and its output end is connected to the input end of the shared layer; the shared layer is used for data feature extraction, and its output end is connected to the input end of the specific layer; the specific layer is used for accurate prediction of multi-objective tasks, and its output end is connected to the input end of the output layer; the output layer serves as the output end of the model.

[0054] In the shared layer, the data with time dependency and space dependency are processed to extract common features. The input data is converted into a tensor form X=(T,H,W,C). T represents the number of steps or time points in the time dimension; H is the data height, which represents the categories of various outpatient resources; W is the data width, which represents the number of outpatient resources under various categories; C represents the number of feature channels. ConvLSTM is used to extract the time and space features of the data. Traditional outpatient resource prediction generally uses a single LSTM long-term and short-term prediction method, which cannot handle multi-dimensional data involving spatial features.

[0055] The input gate control regulation is used to integrate data updates into the memory unit of the outpatient resource prediction model. ,in is the degree to which the output of the input gate controls the data introduced, represents the sigmoid function, and Represent the weight matrices of the current input and the previous hidden state, respectively. represents the hidden state of the previous step, Is the bias term. The output value of the input gate close to 1 indicates that the input data of the current time step has a greater impact on the update of the memory unit, which means that this type of data is very important for outpatient resource prediction and needs to be fully considered; the output value close to 0 indicates that the input data of the current time step has a smaller impact on the update of the memory unit, which means that this type of data has a small impact on outpatient resource prediction and the model can ignore this data.

[0056] Use the forget gate to control the proportion of memory information of the previous time step to be retained in the memory unit of the current time step. ,in and Represent the weight matrices of the current input and the previous hidden state of the forget gate, respectively. represents the sigmoid function, is a bias term. The output value of the forget gate close to 1 indicates that the proportion of memory information of the previous time step retained in the memory unit of the current time step is high, which has a greater impact on the outpatient resource prediction model; the output value close to 0 indicates that the proportion of memory information of the previous time step retained in the memory unit of the current time step is small, which has a smaller impact on the outpatient resource prediction model.

[0057] Use candidate memory cells and memory cell updates to capture long-term dependencies in the time series data of medical visits, improve the model's ability to handle dynamic outpatient resource data, and output intermediate candidate memory cell information ,in represents the weight matrix of the candidate memory unit, It represents the concatenation of the hidden state of the previous time step and the input of the current time step. Represents the bias term of the candidate memory unit. Continue to update the candidate memory unit ,in Represents the state of the memory cell at the previous time step. Use the output gate value Controls the hidden state of the current time step Updates, including represents the weight matrix of the output gate, It represents the concatenation of the hidden state of the previous time step and the input of the current time step. Represents the bias term of the output gate. Finally, the new data feature tensor Y=(T,H,W,F) is output, where F represents the number of feature channels. The new tensor at this time is a set of advanced training feature data processed by the model, which deeply refines the dimensions of outpatient medical data. Compared with ordinary LSTM, using memory unit updates helps to enhance data granularity and capture long-term dependencies between features, and has better applicability and robustness in the face of complex outpatient medical data.

[0058] In a specific layer, a multi-objective task prediction model is established to optimize the prediction results of multiple prediction tasks at the same time. Assume that N target prediction result items are currently set, 0<i<N, i represents any target prediction result item. Set the output , calculate the loss error , evaluate the gap between the predicted results and the actual values, promote further optimization of the model to minimize the error and improve the accuracy of the resource prediction model. represents the number of samples in the i-th prediction task, represents the predicted value of the jth sample, Represents the actual value of the jth sample. The loss error of all prediction tasks is weighted summed. , used to balance the impact of different tasks on resource prediction, where Represents the weight coefficient of the i-th task. Further use Optimization model, where Indicates that the learning rate is used to control the step size of model parameter updates. Represents L for model parameters The gradient of the resource prediction model is accelerated.

[0059] After that, enter the time range conditions and custom target conditions, apply the trained outpatient resource prediction model to predict the outpatient resources under the input conditions and output the results. In an optional scenario, the output results of the outpatient resource prediction are all outpatient medical resource-related prediction results items covering departments, doctors, medical technology departments, the entire hospital, resource consumption, and scheduling, etc., according to the conditions set according to the actual business goals.

[0060] This embodiment is described by the following examples:

[0061] Enter the time range condition and the custom target condition such as [0: "June 2024", 1: "Gastroenterology", 2: "00102_Dr. Li*", 3: "Number of patients", 4: "Number of patients undergoing colon cancer surgery", 5: "Number of capecitabine tablets"}, and the prediction output result is {"Number of patients": "54", "Number of patients undergoing colon cancer surgery": "3", "Number of capecitabine tablets": "10"}. In this way, an outpatient scheduling plan can be formulated based on the prediction results of the outpatient resource model, and the scheduling plan and outpatient resource prediction results can be released to reasonably allocate medical resources, improve outpatient operation efficiency, and enhance patient experience.

[0062] Through the above scheme, this embodiment integrates cutting-edge time series analysis algorithms and multi-objective task deep learning technology to build a comprehensive and sophisticated data analysis system. By extracting common features from historical outpatient registration data and dynamic medical process data, the system can accurately predict the outpatient resource demand in each time period, including department reception load, doctor workload, resource consumption of medical technology departments, and outpatient medical resource consumption. Based on these prediction results, the system can optimize the outpatient scheduling plan in real time, realize the scientific configuration and reasonable allocation of resources, effectively reduce idle resources and waste, and thus significantly improve the utilization efficiency of medical resources. In addition, by reducing patients' waiting time and optimizing the medical process, the quality of medical services and patient satisfaction can be comprehensively improved, providing strong support for medical institutions.

[0063] The above are only preferred embodiments of the present invention. It should be noted that the above preferred embodiments should not be regarded as limiting the present invention, and the protection scope of the present invention should be based on the scope defined by the claims. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the spirit and scope of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A multi-objective task-based outpatient resource prediction method, characterized in that: include: Collect and organize outpatient medical visit data, and perform data preprocessing to generate an initial outpatient visit information dataset; The outpatient medical treatment data includes: historical outpatient registration treatment data, historical patient treatment data, patient treatment process data, patient satisfaction survey data, medical staff satisfaction survey data, emergency data, and regional disease data; The initial outpatient medical information data set is subjected to data analysis and feature extraction, and an outpatient resource prediction model based on multi-objective task learning is constructed and trained; the outpatient resource prediction model comprises an input layer, a shared layer, a specific layer and an output layer; the output end of the input layer is connected to the input end of the shared layer, and the shared layer is used for data feature extraction; the output end of the shared layer is connected to the input end of the specific layer, and the specific layer is used for multi-objective task accurate prediction; the output end of the specific layer is connected to the output layer; wherein, In the shared layer, data with temporal and spatial dependencies are processed to perform general feature extraction, which includes: Convert the input data into tensor form; use the input gate to control and adjust the data update to be integrated into the memory unit of the outpatient resource prediction model; use the forget gate to control the proportion of the memory information of the previous time step to be retained in the memory unit of the current time step; use candidate memory units and memory unit updates to capture the long-term dependencies in the time series data of medical treatment data, and output the intermediate candidate memory unit information; The step of converting the input data into a tensor form includes: Convert the input data into a tensor form X=(T,H,W,C), where T represents the number of steps or time points in the time dimension; H represents the data height, which represents the categories of various outpatient resources; W represents the data width, which represents the number of outpatient resources under various categories; C represents the number of feature channels; use ConvLSTM to extract the temporal and spatial features of the data; The outpatient resource prediction model is continuously iterated and evaluated until it meets the preset training completion conditions, the time range conditions and custom target specific conditions are input, the outpatient resource prediction model is applied to predict the input specific conditions and output the results, and an outpatient scheduling plan is formulated.

2. The outpatient resource prediction method based on multi-objective tasks according to claim 1 is characterized in that: The method of using input gate control to adjust and fuse data update into the memory unit of the outpatient resource prediction model includes: ,in is the degree to which the output of the input gate controls the data introduced, represents the sigmoid function, and Represent the weight matrices of the current input and the previous hidden state, respectively. represents the hidden state of the previous step, is the bias term.

3. The outpatient resource prediction method based on multi-objective tasks according to claim 2 is characterized in that: The use of the forget gate to control the proportion of memory information of the previous time step retained in the memory unit of the current time step includes: ,in and Represent the weight matrices of the current input and the previous hidden state of the forget gate, respectively. represents the sigmoid function, is the bias term.

4. The outpatient resource prediction method based on multi-objective tasks according to claim 3 is characterized in that: The method uses candidate memory units and memory unit updates to capture long-term dependencies in the medical visit data time series data and outputs intermediate candidate memory unit information, including: Output intermediate candidate memory unit information ,in represents the weight matrix of the candidate memory unit, It represents the concatenation of the hidden state of the previous time step and the input of the current time step. Represents the bias term of the candidate memory unit; Continue to update candidate memory cells ,in Represents the state of the memory cell at the previous time step; uses the output gate value Controls the hidden state of the current time step Updates, including represents the weight matrix of the output gate, It represents the concatenation of the hidden state of the previous time step and the input of the current time step. Represents the bias term of the output gate; Output the new data feature tensor Y=(T,H,W,F), where F represents the number of feature channels. The new tensor at this time is a set of high-level training feature data processed by the model.

5. The outpatient resource prediction method based on multi-objective tasks according to claim 1 is characterized in that: In a specific layer, a multi-objective task prediction model is established to optimize the prediction results of multiple prediction tasks at the same time, including: Set N target prediction result item tasks, 0<i<N, i represents any target prediction result item; Setting output , calculate the loss error , evaluate the gap between the predicted results and the actual values, and optimize the model to minimize the error; in represents the number of samples in the i-th prediction task, represents the predicted value of the jth sample, represents the actual value of the jth sample; Weighted summation of the loss errors of all prediction tasks , used to balance the impact of different tasks on resource prediction, where Represents the weight coefficient of the i-th task; Further Use The optimization model is iterated, where Indicates that the learning rate is used to control the step size of model parameter updates. Represents L for model parameters gradient.

6. An outpatient resource prediction system based on multi-objective tasks, characterized in that: include: A data processing module, used to collect and organize outpatient medical treatment data, and perform data preprocessing on it to generate an initial outpatient treatment information data set; The outpatient medical treatment data includes: historical outpatient registration treatment data, historical patient treatment data, patient treatment process data, patient satisfaction survey data, medical staff satisfaction survey data, emergency data, and regional disease data; The model training module is used to perform data analysis and feature extraction on the initial outpatient medical information data set, and to construct and train an outpatient resource prediction model based on multi-objective task learning; the outpatient resource prediction model includes an input layer, a shared layer, a specific layer and an output layer; the output end of the input layer is connected to the input end of the shared layer, and the shared layer is used for data feature extraction; the output end of the shared layer is connected to the input end of the specific layer, and the specific layer is used for multi-objective task accurate prediction; the output end of the specific layer is connected to the output layer; wherein, Processing data with time dependency and space dependency in a shared layer and performing general feature extraction includes: converting input data into a tensor form; using an input gate to control and adjust the data update to be integrated into the memory unit of the outpatient resource prediction model; using a forget gate to control the proportion of memory information of the previous time step to be retained in the memory unit of the current time step; using candidate memory units and memory unit updates to capture long-term dependencies in the time series data of medical treatment data, and outputting intermediate candidate memory unit information; converting the input data into a tensor form includes: converting the input data into a tensor form X=(T,H,W,C), where T represents the number of steps or time points in the time dimension; H is the data height, representing the categories of various outpatient resources; W is the data width, representing the number of outpatient resources under various categories; C represents the number of feature channels; using ConvLSTM to extract the time and space features of the data; and continuously iterating and evaluating the outpatient resource prediction model until the preset training completion conditions are met; The resource prediction module is used to input time range conditions and custom target conditions, apply the outpatient resource prediction model to predict the input conditions and output the results, and formulate an outpatient scheduling plan.

Images