Smart hospital system

Abstract

The invention relates to the technical field of hospital intelligent management, in particular to an intelligent hospital system, which comprises an information collection module, an information analysis module, a data preparation module and an optimal scheduling module, and is characterized in that a hospital guide consultation record of each user is acquired and then input into a preset examination department prediction model; and then, according to the registered users of the target consulting room and the examination department prediction vector of each registered user, optimization scheduling is carried out on the hospital according to the registered users of the target consulting room and the examination department prediction vector of each registered user. Compared with the prior art, the method has the advantages that the association between the vocabularies in the hospital guide record of the patient and the examination departments is analyzed through an attention mechanism in natural language processing, so that the use condition of each examination department is predicted; the possible load condition of each consulting room in a hospital can be known in advance before actual registration and medical treatment, so that optimal scheduling can be performed in advance, and the problem of long waiting time of patients is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of hospital intelligent management, and in particular to a smart hospital system. Background Art

[0002] With the rapid development of information technology, hospital information systems have achieved deep digital transformation. The current hospital system can collect, store, process, extract and exchange patient medical information and administrative information to meet the functional requirements of authorized users. Patients can use mobile terminals to conduct intelligent medical guidance in advance, accurately locate the required departments and experts, and then complete targeted online registration and examination appointment operations, thereby significantly improving the efficiency of the medical process.

[0003] However, in the medical process, patients still inevitably face a long waiting time. In addition to the time waiting for the initial diagnosis after registration, the waiting time in the examination link is particularly prominent, such as blood tests, computed tomography (CT) and other items. And after completing the examination, the patient still needs to wait for the generation of the test report. After the report is issued, the patient needs to return to the clinic again for the doctor to professionally interpret and analyze the report. This process not only prolongs the patient's medical treatment time, but also has a further chain effect on the subsequent patients waiting for the initial diagnosis, resulting in a further increase in the overall waiting time for medical treatment.

[0004] Therefore, people need a smart hospital system that can reduce patients' waiting time. Summary of the invention

[0005] Therefore, the present invention provides a smart hospital system to solve the problem of long waiting time for patients to see a doctor in the prior art.

[0006] The present invention provides a smart hospital system, comprising:

[0007] Information collection module, used to obtain each user's consultation record;

[0008] An information analysis module is used to input each user's medical consultation record into a preset examination department prediction model to obtain a prediction vector of each user's examination department, wherein the preset examination department prediction model is used to calculate the association between the words in the medical consultation record and the name of the hospital's examination department based on the attention mechanism, and output the examination department prediction vector based on the association, and the examination department prediction vector is used to represent the probability that the user will go to each examination department for examination;

[0009] The data preparation module is used to obtain the registered users of the target clinic and the examination department prediction vector of each registered user;

[0010] The optimization scheduling module is used to optimize the hospital scheduling based on the registered users of the target clinic and the predicted vector of the examination department for each registered user.

[0011] In a preferred embodiment: the preset examination department prediction model includes a first coding layer and a prediction layer, wherein:

[0012] The first coding layer is used to obtain first context data according to the medical consultation record of the target user;

[0013] The prediction layer is used to obtain a prediction vector of the examination department of the target user according to the first context data.

[0014] In a preferred embodiment: the first context data is a vector; the prediction layer includes an input layer, at least one fully connected layer and an output layer connected in sequence, wherein:

[0015] The input layer is used to input first context data;

[0016] The fully connected layer is used to analyze the association between the first context data and the name of the hospital's examination department;

[0017] The output layer is used to output the examination department prediction vector.

[0018] In a preferred embodiment: the preset inspection department prediction model further includes at least one second coding layer, the second coding layer has the same structure as the first coding layer, and the output end of the second coding layer is connected to the prediction layer, wherein:

[0019] The second coding layer is used to obtain the second context data according to the name of the target inspection department;

[0020] The prediction layer is used to obtain a test department prediction vector of the target user corresponding to the target test department according to the first context data and the second context data.

[0021] In a preferred embodiment: the preset inspection department prediction model further includes a text normalization layer, and the first encoding layer includes a word segmentation layer, a word embedding layer, a position embedding layer and a plurality of self-attention analysis layers connected in sequence, wherein:

[0022] The text standardization layer is used to convert the consultation records into standard format text, which includes the names of all the examination departments in the hospital;

[0023] The self-attention analysis layer is used to output the attention matrix of the standard text as the first context data

[0024] The prediction layer is used to extract the attention vector of each examination department from the first context data, and aggregate the attention vectors of multiple examination departments to obtain the examination department prediction vector.

[0025] In a preferred embodiment, the hospital is optimized and scheduled according to the registered users of the target clinic and the predicted vector of the examination department of each registered user, including:

[0026] Obtain the prediction vector of the registered users of the target clinic in the target time period and the corresponding examination department;

[0027] According to the prediction vector of the registered users and the corresponding examination departments, the load prediction value of the target clinic in the target time period is obtained;

[0028] Optimize hospital scheduling based on load forecast values.

[0029] In a preferred embodiment: the registered users include consulted users and non-consulted users, the consulted users have corresponding examination department prediction vectors, and the non-consulted users do not have corresponding examination department prediction vectors; according to the registered users and the corresponding examination department prediction vectors, the load prediction value of the target clinic in the target time period is obtained, including:

[0030] Obtaining a first load value according to a prediction vector of the examination department corresponding to the consulted user;

[0031] According to the first load value and the number of non-consulting users, a load forecast value of the target consulting room in the target time period is obtained.

[0032] In a preferred embodiment, obtaining a first load value according to a prediction vector of an examination department corresponding to the consulted user includes:

[0033] Obtain the relevant examination clinics of the target clinic;

[0034] The sum of the element values ​​corresponding to all relevant examination clinics in the examination department prediction vector corresponding to the consulted user is counted to obtain a first load value.

[0035] In a preferred embodiment, obtaining a load forecast value of a target clinic in a target time period according to the first load value and the number of users who have not consulted includes:

[0036] Obtaining the examination department prediction vectors of all users other than the consulted users collected within a preset neighborhood time period of the target time period;

[0037] A second load value is obtained according to the vector of the user predicted by the examination department of all non-consulted users;

[0038] The first load value, the number of non-consulting users, and the second load value are weightedly summed to obtain a load forecast value of the target clinic in the target time period.

[0039] In a preferred embodiment, optimizing the scheduling of hospitals according to the load forecast value includes:

[0040] According to the load forecast value, the number of appointments issued in the target clinic after the target time period, the staff scheduling and resource allocation of the target clinic during the target time period are optimized.

[0041] The beneficial effects of adopting the above embodiment are:

[0042] The present invention provides a smart hospital system, which includes an information collection module, an information analysis module, a data preparation module and an optimization scheduling module. First, the guidance consultation record of each user is obtained, and then the guidance consultation record of each user is input into a preset examination department prediction model to obtain the examination department prediction vector of each user, and then the registered users of the target clinic and the examination department prediction vector of each registered user are obtained. Finally, according to the registered users of the target clinic and the examination department prediction vector of each registered user, the hospital is optimized and scheduled. The preset examination department prediction model in the present invention is used to calculate the association between the vocabulary in the guidance consultation record and the name of the examination department of the hospital based on the attention mechanism, and output the examination department prediction vector according to the association, and the examination department prediction vector is used to characterize the probability that the user will go to each examination department for examination. Compared with the prior art, the present invention analyzes the association between words and examination departments in patient guidance records through the attention mechanism in natural language processing, thereby predicting the usage of each examination department, and utilizing the patient's pre-guidance consultation records to know in advance the possible load conditions of each consulting room in the hospital before the actual registration and medical treatment begins, so that optimization scheduling can be carried out in advance, solving the problem of long waiting time for patients and greatly improving the medical experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A system structure diagram of an embodiment of a smart hospital system provided by the present invention;

[0044] Figure 2 A schematic diagram of the structure of a preset inspection department prediction model in one embodiment of the present invention;

[0045] Figure 3 It is a schematic diagram of the structure of a preset inspection department prediction model in another embodiment of the present invention;

[0046] Figure 4 It is a schematic diagram of the structure of a preset inspection department prediction model in another embodiment of the present invention;

[0047] Figure 5 This is a diagram of specific steps executed by the optimization scheduling module in the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Combination Figure 1 As shown, a specific embodiment of the present invention discloses a smart hospital system, including:

[0050] Information collection module 110, used to obtain the medical consultation record of each user;

[0051] The information analysis module 120 is used to input the medical consultation record of each user into the preset examination department prediction model to obtain the examination department prediction vector of each user, wherein the preset examination department prediction model is used to calculate the association between the words in the medical consultation record and the name of the examination department of the hospital based on the attention mechanism, and output the examination department prediction vector according to the association, and the examination department prediction vector is used to represent the probability that the user will go to each examination department for examination;

[0052] The data preparation module 130 is used to obtain the registered users of the target clinic and the examination department prediction vector of each registered user;

[0053] The optimization scheduling module 140 is used to optimize the scheduling of the hospital according to the registered users of the target clinic and the examination department prediction vector of each registered user.

[0054] In the above process, the guidance function refers to the guidance function in the existing intelligent hospital system, which uses advanced information technology to provide patients with efficient medical guidance. One of its core functions is intelligent triage. After the patient enters the symptoms, the system uses natural language processing technology to understand the symptoms and combines the medical knowledge base to analyze the type of disease and accurately recommend the department for treatment. For example, if you enter "headache, fever", the neurology department or the infectious department will be recommended. At the same time, this function also supports AI-assisted self-examination, recommends suspected illnesses and severity based on symptoms and other factors, matches the registration department, provides patients with accurate medical advice, effectively reduces the workload of the guidance desk, and improves medical efficiency. The present invention is based on the guidance function of the prior art and is improved. The guidance consultation record is the text record generated by the user during the guidance. It can be understood that in the present invention, the guidance step is completed before the user actually registers and sees the doctor.

[0055] The examination department refers to the relevant department in the hospital that conducts examinations, such as blood sampling, electrocardiogram, carbon 13, ultrasound examination, and any other existing department used to examine the patient's physical condition. The preset examination department prediction model is specifically a model that calculates the association between the vocabulary in the consultation record and the name of the hospital's examination department based on the attention mechanism. The attention mechanism is a neural network architecture that simulates human attention. It aims to enable the model to focus on the most important part of the input data for the current task, thereby improving the performance and efficiency of the model. It is currently mainly used in natural language processing. The attention mechanism is a prior art, so this article will not explain it in detail. The specific structure of the preset examination department prediction model will be described in detail later. In this embodiment, the target clinic is the clinic currently to be optimized and analyzed.

[0056] Compared with the prior art, the present invention analyzes the association between words and examination departments in patient guidance records through the attention mechanism in natural language processing, thereby predicting the usage of each examination department, and utilizing the patient's pre-guidance consultation records to know in advance the possible load conditions of each consulting room in the hospital before the actual registration and medical treatment begins, so that optimization scheduling can be carried out in advance, solving the problem of long waiting time for patients and greatly improving the medical experience.

[0057] Specifically, in a preferred embodiment, the preset examination department prediction model includes a first coding layer and a prediction layer, wherein:

[0058] The first coding layer is used to obtain first context data according to the medical consultation record of the target user;

[0059] The prediction layer is used to obtain a prediction vector of the examination department of the target user according to the first context data.

[0060] Existing guidance technologies generally use natural language processing models combined with attention mechanisms. The most commonly used model is the transformer model. The transformer model is a neural network architecture based on the attention mechanism, which was proposed by Vaswani et al. in their 2017 paper "Attention Is All You Need". It completely abandons the traditional recurrent neural network (RNN) structure and relies entirely on the self-attention mechanism to process sequence data. It can process all elements in the input sequence in parallel, greatly improving the training efficiency. The core components of the transformer model include the encoder and the decoder. The encoder converts the input sequence into a contextual representation through multiple stacked self-attention layers and feedforward neural network layers; the decoder uses the encoder's output and its own self-attention mechanism to gradually generate the output sequence. The transformer model has achieved remarkable results in the field of natural language processing (NLP), such as machine translation, text generation and other tasks. Its parallel processing capability and powerful context modeling capabilities make it one of the preferred architectures for modern NLP tasks. Simply put, in existing medical guidance technology, the encoder in the transformer is usually used to analyze the association between words in the text to obtain the context vector of the order and association information between words. Then, through the subsequent decoder, other linear layers, feedforward neural network modules, etc., intelligent response of the text is achieved based on the context vector, such as registration department, doctor's recommendation, etc.

[0061] This embodiment is an improvement on the existing transformer. In this embodiment, the first encoding layer can be directly implemented by the encoder in the existing transformer model to output the first context data containing the vocabulary associations in the patient guidance text, while the prediction layer is where the present invention makes the main improvement. Different from the prior art, compared with the subsequent modules such as the decoding layer in the existing transformer model, the prediction layer in the present invention pays more attention to the analysis of the patient's potential department, and is not responsible for the response of the dialogue.

[0062] It is understandable that the present invention can be directly improved in the existing model. It only needs to directly connect the encoder of the existing transformer model to the newly added prediction layer, so as to simultaneously complete the guidance response and department prediction, which has good practicality.

[0063] Specifically, in a preferred embodiment, the first context data is a vector; the prediction layer includes an input layer, at least one fully connected layer and an output layer connected in sequence, wherein:

[0064] The input layer is used to input first context data;

[0065] The fully connected layer is used to analyze the association between the first context data and the name of the hospital's examination department;

[0066] The output layer is used to output the examination department prediction vector.

[0067] This embodiment directly uses multiple fully connected layers to form a feedforward neural network as a way to implement the prediction layer. The prediction layer in this embodiment needs to be pre-trained, and the training process can be:

[0068] Obtain the guidance record and medical history information of existing patients, including the patient's examination department in the medical record. At this time, a training set can be established, and the patient's guidance record can be input into the transformer model used for guidance, and the context vector output by the encoding layer can be obtained as context data. Then, the context data is used as the input of the prediction layer, and the department where the patient is examined is encoded and used as the output of the prediction layer. The prediction layer is then back-propagated using gradient descent to complete the training of the prediction layer. It can be seen that this embodiment has little improvement over the existing system, and only requires training and adding a prediction layer, which has good practicality. The specific structure of the preset examination department prediction model in this embodiment is as follows: Figure 2 shown.

[0069] Obviously, the prediction layer in the above embodiment has good versatility, but the training cost and scalability of the prediction layer are poor. Therefore, further, the present invention also provides a preferred embodiment. In this embodiment, the preset inspection department prediction model also includes at least one second coding layer, the second coding layer has the same structure as the first coding layer, and the output end of the second coding layer is connected to the prediction layer, wherein:

[0070] The second coding layer is used to obtain the second context data according to the name of the target inspection department;

[0071] The prediction layer is used to obtain a test department prediction vector of the target user corresponding to the target test department according to the first context data and the second context data.

[0072] In the above process, the target examination department name is the name of the examination department associated with the target clinic. This embodiment further reuses the coding layer in the transformer model, and analyzes the target examination department name through the second coding layer with the same structure as the first coding layer. In this way, the guidance consultation record and the target examination department name are analyzed simultaneously through the same vocabulary understanding method, and the first context data representing the meaning and association of the guidance vocabulary and the second context data representing the meaning of the target examination department name are obtained at the same time. The prediction layer is then used to combine the first context data and the second context data to obtain the examination department prediction vector.

[0073] Compared to the previous embodiment, on the one hand, this embodiment analyzes the name of the inspection department while actually establishing the inspection department prediction vector, which makes it possible for the inspection departments that need to be considered to be flexibly adjusted according to specific needs during actual use. For example, when a hospital is small in scale and does not have certain inspection departments, or some inspection departments cannot be used, there is no need to enter the names of these invalid inspection departments in the second coding layer, thereby improving the pertinence and accuracy of the analysis of the preset inspection department prediction model and making it more flexible. On the other hand, because the second coding layer itself is a reuse of the coding layer in the transformer model in the existing guidance function, and the function of analyzing the inspection clinic name in this embodiment is implemented by the second coding layer, the prediction layer only needs to complete the analysis of the clinic association, and the prediction layer can be more lightweight, with lower training costs, faster development speed, and faster computing speed. The specific structure of the preset inspection department prediction model in this embodiment is as follows: Figure 3 shown.

[0074] It can be imagined that the prediction layers in the above two embodiments need to be pre-trained, and the present invention also provides a more preferred implementation method to omit the training of the prediction layer and further improve the practicality of the preset examination department prediction model.

[0075] Specifically, in a preferred embodiment, the preset inspection department prediction model further includes a text normalization layer, and the first encoding layer includes a word segmentation layer, a word embedding layer, a position embedding layer and a plurality of self-attention analysis layers connected in sequence, wherein:

[0076] The text standardization layer is used to convert the consultation records into standard format text, which includes the names of all the examination departments in the hospital;

[0077] The self-attention analysis layer is used to output the attention matrix of the standard text as the first context data

[0078] The prediction layer is used to extract the attention vector of each examination department from the first context data, and aggregate the attention vectors of multiple examination departments to obtain the examination department prediction vector.

[0079] In the above content, the standard format text refers to the preset standardized format text including the consultation record and the name of the examination department that needs to be analyzed and associated, which is used to analyze the department association by the first coding layer. For example, a patient has the following consultation record:

[0080] "Hello, I have had a headache and fever for three days. I don't know which department to go to. The headache is persistent, accompanied by nausea, but no vomiting. I have a low-grade fever and feel a little weak." The hospital has the following examination departments: laboratory, radiology, ultrasound, electrocardiogram, pulmonary function, electroencephalogram, gastroscopy, colonoscopy, bronchoscopy, nuclear medicine, pathology, and functional examination. Then, the generated standardized format text can be:

[0081] "The patient said: Hello, I have had a headache and fever recently, which has lasted for 3 days. I don't know which department to go to? The headache is persistent, accompanied by nausea, but no vomiting. I have a low-grade fever and feel a little weak. According to the above information, the clinics that the patient may go to include the laboratory, radiology, ultrasound, electrocardiogram, pulmonary function, electroencephalogram, gastroscopy, colonoscopy, bronchoscopy, nuclear medicine, pathology, and functional examination."

[0082] The purpose of constructing a standard format text is to simulate a reasonable text including consultation records and department names, and use reasonable word order relationships and the self-attention analysis capability of the encoding layer in the existing transformer model to analyze the relationship between departments and vocabulary and obtain the attention matrix.

[0083] The word segmentation layer, word embedding layer, position embedding layer and multiple self-attention analysis layers in the first coding layer are all structures that already exist in the existing transformer model, so this article will not introduce them in detail. However, it should be noted that the self-attention analysis layer in the prior art is used to output the attention matrix. The attention matrix is ​​the core output of the self-attention mechanism, which is used to represent the relationship and importance between the elements in the input sequence. In this embodiment, the attention matrix includes information representing the relationship between the various words in the consultation record. For example, under the premise of reasonable word embedding technology, in the attention matrix obtained by the above-mentioned standard format text, the vocabulary "low fever" and "radiology department" are highly correlated, but the correlation with "gastroscopy room" and "colonoscopy room" is relatively low. In the prior art, the attention matrix needs to be processed by other structures such as feedforward neural networks before the context vector output by the encoder in the transformer model can be obtained.

[0084] In this embodiment, there is no need to perform subsequent processing steps, and the attention matrix is ​​directly used as context data. Because the lexical position of the department name in the standard format text is determined, the position of the expected related attention vector in the attention matrix is ​​also determined. Therefore, the prediction layer only needs to extract the attention vector of each inspection department, and then summarize it to obtain the inspection department prediction vector without the need for machine learning analysis.

[0085] This embodiment makes full use of the analysis capability of the first coding layer, so that the prediction layer does not need to be analyzed and trained, and can be used directly after being simply built. The same is true for the text standardization layer. This further improves the usability of the preset inspection department prediction model. The specific structure of the preset inspection department prediction model in this embodiment is as follows: Figure 4 shown.

[0086] Furthermore, in a preferred embodiment, the inspection department prediction vector is calculated by the following formula:

[0087] u i =||v i ||;

[0088] Among them, u represents the prediction vector of the examination department, u i represents the element corresponding to the i-th examination department in the examination department prediction vector, v i represents the attention vector of the i-th examination department, and |||| represents the calculation of the L2 norm of a vector.

[0089] Further, combined with Figure 5 As shown, in a preferred embodiment, the steps performed by the above-mentioned optimization scheduling module are: optimizing the hospital scheduling according to the registered users of the target clinic and the examination department prediction vector of each registered user, specifically including:

[0090] S501, obtaining the registered users of the target clinic in the target time period and the corresponding examination department prediction vector;

[0091] S502, obtaining a load forecast value of a target clinic in a target time period according to a forecast vector of a registered user and a corresponding examination department;

[0092] S503. Optimize the scheduling of hospitals based on load forecast values.

[0093] In the above process, the target time period is the time period for optimized scheduling. For example, the registered users of the target clinic are currently counted, and the target time period may be tomorrow, the day after tomorrow, or even a certain time of the next week. The optimized scheduling is performed by calculating the load forecast value, making the process controllable.

[0094] This embodiment assumes that all registered users will receive a medical consultation before the face-to-face consultation. However, it is understandable that in practice not all registered users will receive a consultation. There are also users who register directly without receiving a prior medical consultation. Therefore, in a preferred embodiment, registered users include users who have consulted and users who have not consulted. Users who have consulted have corresponding examination department prediction vectors, while users who have not consulted do not have corresponding examination department prediction vectors.

[0095] The above step S502, based on the prediction vector of the registered user and the corresponding examination department, obtains the load prediction value of the target clinic in the target time period, specifically including:

[0096] Obtaining a first load value according to a prediction vector of the examination department corresponding to the consulted user;

[0097] According to the first load value and the number of non-consulting users, a load forecast value of the target consulting room in the target time period is obtained.

[0098] In the above process, for users who have consulted, the prediction vector of the examination department can be directly called and the first load value can be calculated. For users who have not consulted, the number of users can be used to match the first load value to calculate the load prediction value. For example, the probability of the target clinic going to a certain examination department for examination can be calculated based on experience or statistics, and then the probability can be used as the weight coefficient of the number of users who have not consulted. The load prediction value is obtained by weighted summation of the first load value and the number of users who have not consulted. This makes the load prediction value obtained in this embodiment more scientific and reasonable.

[0099] Specifically, the above steps obtain the first load value according to the examination department prediction vector corresponding to the consulted user, and specifically include:

[0100] Obtain the relevant examination clinics of the target clinic;

[0101] The sum of the element values ​​corresponding to all relevant examination clinics in the examination department prediction vector corresponding to the consulted user is counted to obtain a first load value.

[0102] The examination department prediction vector is likely to represent all departments of the hospital. It is understandable that not every clinic will use all departments, so the above process screens the relevant examination clinics related to the target clinic, and then analyzes the examination department prediction vector (that is, the sum of the element values ​​corresponding to all relevant examination clinics in the examination department prediction vector) to obtain a more accurate first load value.

[0103] Furthermore, in practice, not all consulting users will register after the consultation, and the present invention is a prediction based on the consultation data, and the target time period of the target clinic is not the time period for counting the number of registered users, which may lead to a deviation between the actual situation and the predicted situation, so this error should also be taken into consideration. In a preferred embodiment, the above step: according to the first load value and the number of non-consulting users, obtaining the load prediction value of the target clinic in the target time period specifically includes:

[0104] Obtaining the examination department prediction vectors of all users other than the consulted users collected within a preset neighborhood time period of the target time period;

[0105] Obtaining a second load value according to the examination department prediction vectors of all users who are not the consulted users;

[0106] The first load value, the number of non-consulting users, and the second load value are weightedly summed to obtain a load forecast value of the target clinic in the target time period.

[0107] In the above process, the preset domain time period can be any artificially defined time period such as the week before the target time period, the first three days, etc. It is used to collect the number of people who have received consultation during the time period, and then to a certain extent indicate the use of the examination department. The above error is expressed by the second load value to obtain a more accurate and reasonable load prediction value. The process of obtaining the second load value can be the same as the process of obtaining the first load value, or it can be obtained by using a linear model or any existing technology based on actual experience or situation.

[0108] Furthermore, in a preferred embodiment, the steps performed by the above optimization scheduling module include: optimizing the scheduling of hospitals according to the load forecast value, specifically including:

[0109] According to the load forecast value, the number of appointments issued in the target clinic after the target time period, the staff scheduling and resource allocation of the target clinic during the target time period are optimized.

[0110] It is understandable that how to perform optimized scheduling can be flexibly set according to specific circumstances.

[0111] The present invention provides a smart hospital system, which includes an information collection module, an information analysis module, a data preparation module and an optimization scheduling module. First, the guidance consultation record of each user is obtained, and then the guidance consultation record of each user is input into a preset examination department prediction model to obtain the examination department prediction vector of each user, and then the registered users of the target clinic and the examination department prediction vector of each registered user are obtained. Finally, according to the registered users of the target clinic and the examination department prediction vector of each registered user, the hospital is optimized and scheduled. The preset examination department prediction model in the present invention is used to calculate the association between the vocabulary in the guidance consultation record and the name of the examination department of the hospital based on the attention mechanism, and output the examination department prediction vector according to the association, and the examination department prediction vector is used to characterize the probability that the user will go to each examination department for examination. Compared with the prior art, the present invention analyzes the association between words and examination departments in patient guidance records through the attention mechanism in natural language processing, thereby predicting the usage of each examination department, and utilizing the patient's pre-guidance consultation records to know in advance the possible load conditions of each consulting room in the hospital before the actual registration and medical treatment begins, so that optimization scheduling can be carried out in advance, solving the problem of long waiting time for patients and greatly improving the medical experience.

[0112] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0113] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A smart hospital system, characterized in that: include: Information collection module, used to obtain each user's consultation record; An information analysis module is used to input each user's medical consultation record into a preset examination department prediction model to obtain a prediction vector of each user's examination department, wherein the preset examination department prediction model is used to calculate the association between the words in the medical consultation record and the name of the hospital's examination department based on the attention mechanism, and output the examination department prediction vector based on the association, and the examination department prediction vector is used to represent the probability that the user will go to each examination department for examination; The data preparation module is used to obtain the registered users of the target clinic and the examination department prediction vector of each registered user; The optimization scheduling module is used to optimize the hospital scheduling based on the registered users of the target clinic and the predicted vector of the examination department for each registered user.

2. The smart hospital system according to claim 1, characterized in that: The preset inspection department prediction model includes the first coding layer and the prediction layer, where: The first coding layer is used to obtain first context data according to the medical consultation record of the target user; The prediction layer is used to obtain a prediction vector of the examination department of the target user according to the first context data.

3. The smart hospital system according to claim 2, characterized in that: The first context data is a vector; the prediction layer includes an input layer, at least one fully connected layer and an output layer connected in sequence, wherein: The input layer is used to input first context data; The fully connected layer is used to analyze the association between the first context data and the name of the hospital's examination department; The output layer is used to output the examination department prediction vector.

4. The smart hospital system according to claim 2, characterized in that: The preset inspection department prediction model also includes at least one second coding layer, the second coding layer has the same structure as the first coding layer, and the output end of the second coding layer is connected to the prediction layer, wherein: The second coding layer is used to obtain the second context data according to the name of the target inspection department; The prediction layer is used to obtain a test department prediction vector of the target user corresponding to the target test department according to the first context data and the second context data.

5. The smart hospital system according to claim 2, characterized in that: The preset inspection department prediction model also includes a text normalization layer, and the first encoding layer includes a word segmentation layer, a word embedding layer, a position embedding layer and multiple self-attention analysis layers connected in sequence, wherein: The text standardization layer is used to convert the consultation records into standard format text, which includes the names of all the examination departments in the hospital; The self-attention analysis layer is used to output the attention matrix of the standard text as the first context data The prediction layer is used to extract the attention vector of each examination department from the first context data, and aggregate the attention vectors of multiple examination departments to obtain the examination department prediction vector.

6. The smart hospital system according to claim 1, characterized in that: According to the target clinic's registered users and the predicted vector of each registered user's examination department, the hospital is optimized and dispatched, including: Obtain the prediction vector of the registered users of the target clinic in the target time period and the corresponding examination department; According to the prediction vector of the registered users and the corresponding examination departments, the load prediction value of the target clinic in the target time period is obtained; Optimize hospital scheduling based on load forecast values.

7. The smart hospital system according to claim 6, characterized in that: The registered users include users who have consulted and users who have not consulted. Users who have consulted have corresponding examination department prediction vectors, while users who have not consulted do not have corresponding examination department prediction vectors. According to the prediction vector of the registered users and the corresponding examination departments, the load prediction value of the target clinic in the target time period is obtained, including: Obtaining a first load value according to a prediction vector of the examination department corresponding to the consulted user; According to the first load value and the number of non-consulting users, a load forecast value of the target consulting room in the target time period is obtained.

8. The smart hospital system according to claim 7, characterized in that: According to the prediction vector of the examination department corresponding to the consulted user, a first load value is obtained, including: Obtain the relevant examination clinics of the target clinic; The sum of the element values ​​corresponding to all relevant examination clinics in the examination department prediction vector corresponding to the consulted user is counted to obtain a first load value.

9. The smart hospital system according to claim 7, characterized in that: According to the first load value and the number of users who have not consulted, the load forecast value of the target clinic in the target time period is obtained, including: Obtaining the examination department prediction vectors of all users other than the consulted users collected within a preset neighborhood time period of the target time period; Obtaining a second load value according to the prediction vectors of the examination departments of all non-consulted users; The first load value, the number of non-consulting users, and the second load value are weightedly summed to obtain a load forecast value of the target clinic in the target time period.

10. The smart hospital system according to claim 6, characterized in that: According to the load forecast value, the hospital is optimized for scheduling, including: According to the load forecast value, the number of appointments issued in the target clinic after the target time period, the staff scheduling and resource allocation of the target clinic during the target time period are optimized.

Images