Hospital specialized diagnosis and treatment flow prediction method based on big data

Abstract

The invention relates to the technical field of diagnosis and treatment flow prediction, and discloses a hospital specialized diagnosis and treatment flow prediction method based on big data, and the method comprises the steps: carrying out the aggregation of hospital specialized flow data through employing a time window sliding mode, and obtaining the hospital specialized flow aggregation data of each time period; calculating the dynamic information gain of the flow index, and extracting the multi-layer diagnosis and treatment flow features by using the dynamic information gain of the flow index; and predicting the hospital specialized diagnosis and treatment flow by using the hospital specialized diagnosis and treatment flow prediction model. According to the method, the periodic window sizes of different flow indexes are generated in combination with the data distribution characteristics and the fluctuation degree, then the adaptive time window is generated, data aggregation is performed by adopting a time window sliding mode of the adaptive window size, the obvious degree of the periodic characteristics is improved, and the data aggregation efficiency is improved. The multi-layer diagnosis and treatment flow features representing the multi-layer features of the flow indexes are extracted, multi-layer time sequence analysis and prediction are carried out, and the flow prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of diagnosis and treatment flow prediction, and particularly to a method for predicting the diagnosis and treatment flow of hospital specialties based on big data. Background Art

[0002] With the rapid development of information technology and the application of big data technology, the intelligent level of the medical industry has been significantly improved. Especially in hospital management and operation, the decision support system based on big data has gradually become an important means for hospitals to improve operation efficiency, optimize resource allocation, and enhance the patient experience. As a key issue in hospital operation management, the prediction of hospital specialty diagnosis and treatment flow has an important impact on the reasonable allocation of hospital resources, the load balance of departments, the patient visit experience, and the quality of medical services. Traditional hospital flow prediction methods mainly rely on manual experience, statistical analysis, and simple linear regression models. Since the diagnosis and treatment flow is affected by multiple factors, such as seasonal changes, epidemic trends, weather, holidays, etc., traditional models are difficult to capture these non-linear relationships, and hospital diagnosis and treatment flow data usually has obvious time correlations (such as seasonal fluctuations, weekend effects, etc.). Traditional methods lack effective modeling capabilities for such complex time series characteristics. For example, the impact of external factors such as meteorological conditions, traffic conditions, and social activities on hospital flow is difficult to quantify through simple regression analysis, resulting in limited prediction accuracy. Summary of the Invention

[0003] In view of this, the present invention proposes a method for predicting the diagnosis and treatment flow of hospital specialties based on big data. By combining multi-source data such as hospital electronic medical records, registration systems, diagnosis and treatment records, patient demographic information, and environmental factors, it can more accurately capture the changing rules of the diagnosis and treatment flow and improve the prediction accuracy.

[0004] To achieve the above object, a method for predicting the diagnosis and treatment flow of hospital specialties based on big data provided by the present invention includes the following steps: S1: Collect hospital specialty diagnosis and treatment data and external influence data as hospital specialty flow data based on flow indicators, and aggregate the hospital specialty flow data by using the time window sliding method to obtain the hospital specialty flow aggregation data for each time period.

[0005] S2: Calculate the dynamic information gain of the flow indicators based on the hospital specialty flow aggregation data, and extract multi-layer diagnosis and treatment flow characteristics by using the dynamic information gain of the flow indicators.

[0006] S3: Construct a hospital specialty diagnosis and treatment flow prediction model.

[0007] S4: Use the hospital specialty diagnosis and treatment flow prediction model to predict the hospital specialty diagnosis and treatment flow.

[0008] As a further improvement method of the present invention: Optionally, the internal hospital traffic impact indicators include the registration information, diagnosis information, surgery information, hospitalization information of patients, and the load situation of departments, and the external traffic impact indicators include meteorological conditions, epidemic trends, traffic conditions, and event information.

[0009] The registration information describes the number of registrations in hospital specialties.

[0010] The diagnosis information describes the number of critically ill patients, the surgery information describes the number of patients undergoing surgery, the hospitalization information describes the number of inpatients, and the load situation of departments describes the current load of different departments, where the load represents the number of patients seeking medical treatment.

[0011] The collected data under the meteorological condition indicator includes temperature and PM2.5. The temperature and PM2.5 under the collected meteorological condition indicator are respectively normalized, and the results of the normalization are multiplied as the collected data under the meteorological condition indicator. The collected data under the epidemic trend indicator includes the presence or absence of an epidemic. The collected data under the traffic condition indicator is the degree of congestion at the intersections around the hospital. The collected data under the event information indicator is whether there is a large event in the area where the hospital is located.

[0012] Optionally, the hospital specialty traffic data is aggregated in a time window sliding manner, including: The hospital specialty traffic data is divided into hourly granularity to obtain the hourly hospital specialty traffic data, and the hourly traffic data of each traffic indicator is extracted. The hourly traffic data of the nth traffic indicator is: .

[0013] Where: represents the hourly traffic data of the nth traffic indicator, and the , and the 1st - 9th traffic indicators are registration information, diagnosis information, surgery information, hospitalization information, load situation of departments, meteorological conditions, epidemic trends, traffic conditions, and event information in sequence; represents the collected data of the nth traffic indicator at the mth hour, , and M represents the time series length.

[0014] The fluctuation degree of the hourly traffic data of different traffic indicators is calculated, and the fluctuation degree and the data distribution characteristics of the hourly traffic data are integrated to generate the window weight of the traffic indicator, where the data distribution characteristics include skewness and kurtosis.

[0015] Calculate the time window size required for each traffic metric, perform weighted processing using window weights to obtain the length of the time window, and construct the time window.

[0016] Use the time window to perform time window sliding processing on the hospital specialty traffic data to obtain the hospital specialty traffic data under different time windows, and the hospital specialty traffic data under adjacent time windows do not overlap. Take the hospital specialty traffic data under each time window as the hospital specialty traffic aggregation data for the corresponding time period. The number of the hospital specialty traffic aggregation data is 。

[0017] Optionally, calculate the dynamic information gain of the traffic metric based on the hospital specialty traffic aggregation data, including. The calculation process of the dynamic information gain of the nth traffic metric is as follows: Calculate the external influence conditional entropy and information entropy of different traffic metrics in each hospital specialty traffic aggregation data, and take the difference between the external influence conditional entropy and the information entropy as the information gain of the traffic metric in the hospital specialty traffic aggregation data. The information entropy of the nth traffic metric in the e-th hospital specialty traffic aggregation data is : 。

[0018] Where: Represents the standard value of the preset nth traffic metric; Represents the data acquisition result of the nth traffic metric at the i-th hour in the e-th hospital specialty traffic aggregation data, , Represents the number of the hospital specialty traffic aggregation data.

[0019] The external influence conditional entropy of the nth traffic metric in the e-th hospital specialty traffic aggregation data is : 。

[0020] Where: Represents the mean value of the data acquisition results of the nth traffic metric at the moments when the meteorological conditions, epidemic trends, traffic conditions, and activity information in the e-th hospital specialty traffic aggregation data are all lower than the corresponding influence index thresholds, where the unit of each moment is hour.

[0021] The information gain of the nth traffic metric in the e-th hospital specialty traffic aggregation data is 。

[0022] Calculate the dynamic change value of the information gain of the traffic indicator in the aggregated traffic data of adjacent hospital specialties, and standardize the dynamic change value of the information gain as the dynamic information gain of the traffic indicator in different time periods.

[0023] The dynamic information gain of the nth traffic indicator in the time period corresponding to the aggregated traffic data of the e-th hospital specialty is: 。

[0024] Where: represents the dynamic information gain of the nth traffic indicator in the time period corresponding to the aggregated traffic data of the e-th hospital specialty.

[0025] Optionally, use the dynamic information gain of the traffic indicator to extract multi-layer diagnosis and treatment traffic characteristics, including: Extract the traffic data sequence corresponding to the traffic indicator from the aggregated traffic data of the hospital specialty based on the dynamic information gain, and use the traffic data sequence as the first-layer diagnosis and treatment traffic characteristic of the traffic indicator.

[0026] Calculate the statistical characteristics of the traffic data sequence as the second-layer diagnosis and treatment traffic characteristic of the traffic indicator, where the statistical characteristics include the mean, variance, standard deviation, and skewness of the traffic data sequence.

[0027] Use the dynamic information gain as the weight to weight the index data to obtain the weighted time series of each traffic indicator. Use the smoothing filtering method to extract the long-term trend term of the weighted time series, and use the Fourier transform method to extract the periodic term of the weighted time series. Use the long-term trend term and the periodic term as the third-layer diagnosis and treatment traffic characteristic of the traffic indicator.

[0028] Obtain the traffic data sequence corresponding to the external traffic impact indicator, and calculate the impact perturbation sequence of the external traffic impact indicator as the fourth-layer diagnosis and treatment traffic characteristic.

[0029] Use the first-layer to fourth-layer diagnosis and treatment traffic characteristics as the multi-layer diagnosis and treatment traffic characteristics.

[0030] Optionally, calculate the impact perturbation sequence of the external traffic impact indicator, including: The form of the impact perturbation sequence is: ; 。

[0031] Where: represents the impact perturbation sequence, represents the e-th impact perturbation value in the impact perturbation sequence g, To affect the disturbance coefficient, They are the mean values of the data acquisition results of the 6th - 9th traffic indicators in the aggregated traffic data of the e - th hospital specialty, where the 6th - 9th traffic indicators are meteorological conditions, epidemic trends, traffic conditions, and activity information in sequence.

[0032] Optionally, the hospital specialty diagnosis and treatment traffic prediction model includes an input layer, a hidden state extraction layer, and a fully - connected layer. The hospital specialty diagnosis and treatment traffic prediction model is a long short - term memory neural network structure. The input layer is used to receive multi - layer diagnosis and treatment traffic features. The hidden layer state extraction layer includes a forgetting gate, an input gate, a memory unit, and an output gate. The fully - connected layer is used to map the output result of the output gate to obtain the prediction result of the hospital specialty diagnosis and treatment traffic. The loss function of the model parameters in the hospital specialty diagnosis and treatment traffic prediction model is the mean square error loss.

[0033] The training process of the hospital specialty diagnosis and treatment traffic prediction model is as follows: Initialize the model parameters to be optimized , set the current iteration number of the model parameters as t, and the maximum iteration number as Max. Then the t - th iteration result of the model parameters is ; Construct the loss function of the model parameters : ; ; .

[0034] Among them: represents the loss function of the model parameters , represents the L2 norm, represents the true value of the r - th group of training samples at subsequent times, represents the predicted value of the r - th group of training samples at subsequent times predicted based on the model parameters ; represents the variance in the r - th group of training samples. Each group of training samples is a group of multi - layer diagnosis and treatment traffic features, represents the regularization parameter, represents the variance influence control coefficient, represents the error influence control coefficient, represents the prediction error of predicting the r - th group of samples using the model parameters , represents the sample weight of the r - th group of training samples for the model parameters .

[0035] Iterate the model parameters based on the loss function until the maximum number of iterations is reached, where the model parameters have the following iteration formula: ; ; ; ; ; ; .

[0036] Where: represents the learning rate during the t-th iteration, represents the iteration coefficient, represents the initial coefficient, represents the model parameters of the first moment, represents the model parameters of the second moment, represents the iteration control parameter; represents the model parameters of the gradient, represents the gradient direction control coefficient, represents the first-order gradient decay value, represents the second-order gradient decay value, represents the gradient descent direction.

[0037] Optionally, use the hospital specialist diagnosis and treatment traffic prediction model to predict the hospital specialist diagnosis and treatment traffic, including: The input layer receives multi-layer diagnosis and treatment traffic features and inputs the multi-layer diagnosis and treatment traffic features into the hidden layer state extraction layer.

[0038] The hidden layer state extraction layer inputs the multi-layer diagnosis and treatment traffic features into the forget gate and sequentially performs operations of input gate convolution, update memory unit, and output gate convolution to obtain the hidden information of the multi-layer diagnosis and treatment traffic features.

[0039] The fully connected layer uses weight parameters and bias parameters to map the hidden information of the multi-layer diagnosis and treatment traffic features to obtain the prediction result of the hospital specialist diagnosis and treatment traffic at subsequent times.

[0040] To solve the above problems, the present invention provides an electronic device, where the electronic device includes: A memory that stores at least one instruction; A communication interface that enables communication of the electronic device; and A processor that executes instructions stored in the memory to implement the above-mentioned big-data-based hospital specialty diagnosis and treatment traffic prediction method.

[0041] To solve the above problems, the present invention also provides a computer-readable storage medium storing at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned big-data-based hospital specialty diagnosis and treatment traffic prediction method.

[0042] Compared with the prior art, the present invention proposes a big-data-based hospital specialty diagnosis and treatment traffic prediction method, which has the following advantages: First, the present solution proposes a quantitative analysis method for traffic indicators. Based on traffic indicators, multi-source data affecting diagnosis and treatment traffic is collected, and periodic window sizes for different traffic indicators are generated by combining data distribution characteristics and fluctuation degrees. Then, an adaptive time window is generated, and a time window sliding method with an adaptive window size is used for data aggregation to reduce data dimensions and improve the obviousness of periodic characteristics. By combining the information gain of traffic indicators in different time periods, the dynamic information gain of traffic indicators is characterized. The higher the dynamic information gain, the higher the influence of traffic indicators on environmental factors, and the more obvious the characteristics of long-term trend terms and periodic terms.

[0043] At the same time, the present invention proposes a model parameter training method, extracts multi-layer diagnosis and treatment traffic characteristics representing traffic indicators at multiple levels, performs multi-faceted time series analysis to achieve traffic prediction. During the model parameter optimization process, L2 regularization is used to avoid model overfitting, and as training progresses, the regularization parameter will gradually decrease, allowing the model to fit the data more freely in the later stage, further reducing the risk of overfitting. Considering past prediction errors, the sample weights of each sample are dynamically adjusted to increase the model's attention to samples with prediction errors. The first-order gradient decay value is combined with the gradient direction for gradient descent. For cases where the gradient direction changes greatly, the first moment is reduced to reduce the dependence on historical gradients. For cases where the gradient direction is stable, the first moment is increased to maintain the optimization direction and avoid over-reliance on historical information when the gradient direction changes greatly. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic flowchart of a big-data-based hospital specialty diagnosis and treatment traffic prediction method provided by an embodiment of the present invention.

[0045] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0047] An embodiment of the present application provides a method for predicting the flow of hospital specialist diagnosis and treatment based on big data. The execution subject of the method for predicting the flow of hospital specialist diagnosis and treatment based on big data includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for predicting the flow of hospital specialist diagnosis and treatment based on big data can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0048] Refer to Figure 1 , Embodiment 1 of the present invention is: A method for predicting the flow of hospital specialist diagnosis and treatment based on big data, comprising the following steps: S1: Collect hospital specialist diagnosis and treatment data and external influence data as hospital specialist flow data based on flow indicators, and aggregate the hospital specialist flow data in a time window sliding manner to obtain the hospital specialist flow aggregation data for each time period.

[0049] The internal hospital flow influence indicators include the registration information, diagnosis information, surgery information, hospitalization information of patients, and the load situation of departments. The external flow influence indicators include meteorological conditions, epidemic trends, traffic conditions, and activity information.

[0050] The registration information describes the number of registrations in a hospital specialist department.

[0051] The diagnosis information describes the number of critically ill patients, the surgery information describes the number of patients undergoing surgery, the hospitalization information describes the number of inpatients, and the load situation of the department describes the current load of different departments. The load represents the number of patients seeking medical treatment.

[0052] The collected data under the meteorological condition index includes temperature and PM2.5. The temperature and PM2.5 under the collected meteorological condition index are respectively normalized, and the normalized results are multiplied as the collected data under the meteorological condition index. The collected data under the epidemic trend index includes the existence of an epidemic. The collected data under the traffic condition index is the degree of congestion at the intersections around the hospital. The collected data under the activity information index is whether there is a large-scale activity in the area where the hospital is located. As an embodiment of the present invention, the degree of congestion at the intersection is the average time required to pass through the intersection after the vehicle enters the lane associated with the intersection. Large-scale activities include marathons and conferences, etc.

[0053] The collection result of the hospital specialist flow data is : .

[0054] Wherein: represents hospital specialist diagnosis and treatment data, represents external influence data.

[0055] The hospital specialist traffic data is aggregated by means of time window sliding, including: The hospital specialist traffic data is divided by hour granularity to obtain the hospital specialist traffic data per hour, and the hourly traffic data of each traffic indicator is extracted. The hourly traffic data of the nth traffic indicator is: .

[0056] Wherein: represents the hourly traffic data of the nth traffic indicator, and the , and the 1st - 9th traffic indicators are registration information, diagnosis information, surgery information, hospitalization information, load situation of the department, meteorological conditions, epidemic trend, traffic conditions, and activity information in sequence; represents the collected data of the nth traffic indicator in the mth hour, , and M represents the time series length;

[0057] The fluctuation degree of the hourly traffic data of different traffic indicators is calculated, and the fluctuation degree and the data distribution characteristics of the hourly traffic data are integrated to generate the window weight of the traffic indicator, where the data distribution characteristics include skewness and kurtosis.

[0058] The fluctuation degree of the hourly traffic data is : .

[0059] Wherein: represents the variance of the hourly traffic data , represents the mean value of the hourly traffic data ; specifically, by using the standardized variance as the fluctuation degree, the volatility of the data source is adjusted to a relatively balanced level. Especially for data sources with a large difference in numerical ranges, the influence brought by units and dimensions can be eliminated after standardization.

[0060] The window weight of the nth traffic indicator is : .

[0061] Wherein: Represents the hourly traffic data skewness of, Represents the hourly traffic data kurtosis of, respectively represent the skewness control factor and the kurtosis control factor.

[0062] Calculate the time window size required for each traffic metric, perform weighted processing using window weights to obtain the length of the time window, and construct the time window; where the time window size required for the nth traffic metric is : ; .

[0063] Where: c represents the standard adjustment factor, represents the adjustment factor of the nth traffic metric.

[0064] The length of the time window is: .

[0065] Use the time window to perform time window sliding processing on the hospital specialty traffic data to obtain the hospital specialty traffic data under different time windows, and the hospital specialty traffic data under adjacent time windows do not overlap. Take the hospital specialty traffic data under each time window as the hospital specialty traffic aggregation data for the corresponding time period. The number of the hospital specialty traffic aggregation data is .

[0066] S2: Calculate the dynamic information gain of the traffic metrics based on the hospital specialty traffic aggregation data, and extract multi-layer diagnosis and treatment traffic features using the dynamic information gain of the traffic metrics.

[0067] Calculate the dynamic information gain of the traffic metrics based on the hospital specialty traffic aggregation data, including that the calculation process of the dynamic information gain of the nth traffic metric is: Calculate the external influence conditional entropy and information entropy of different traffic metrics in each hospital specialty traffic aggregation data, and take the difference between the external influence conditional entropy and the information entropy as the information gain of the traffic metric in the hospital specialty traffic aggregation data; the information entropy of the nth traffic metric in the e-th hospital specialty traffic aggregation data is : .

[0068] Where: represents the preset standard value of the nth traffic metric; Represents the data collection result of the nth traffic indicator at the ith hour in the traffic aggregation data of the e-th hospital specialty.

[0069] The external influence conditional entropy of the nth traffic indicator in the traffic aggregation data of the e-th hospital specialty is : .

[0070] Where: Represents the average value of the data collection results of the nth traffic indicator at the moments when the meteorological conditions, epidemic trends, traffic conditions, and activity information in the traffic aggregation data of the e-th hospital specialty are all lower than the corresponding impact indicator thresholds, where the unit of each moment is hour.

[0071] The information gain of the nth traffic indicator in the traffic aggregation data of the e-th hospital specialty is .

[0072] Calculate the dynamic change value of the information gain of the traffic indicator in adjacent hospital specialty traffic aggregation data, and standardize the dynamic change value of the information gain as the dynamic information gain of the traffic indicator in different time periods.

[0073] The dynamic information gain of the nth traffic indicator in the corresponding time period of the traffic aggregation data of the e-th hospital specialty is: .

[0074] Where: Represents the dynamic information gain of the nth traffic indicator in the corresponding time period of the traffic aggregation data of the e-th hospital specialty.

[0075] Extract multi-layer diagnosis and treatment traffic features using the dynamic information gain of the traffic indicator, including: Extract the traffic data sequence corresponding to the traffic indicator from the hospital specialty traffic aggregation data based on the dynamic information gain, and use the traffic data sequence as the first-layer diagnosis and treatment traffic feature of the traffic indicator, where the traffic data sequence corresponding to the nth traffic indicator is : .

[0076] Where: Represents the traffic data of the nth traffic indicator in traffic aggregation data of the hospital specialty, Represents the traffic data of the nth traffic indicator in the traffic aggregation data of the e-th hospital specialty, where the traffic data Including the dynamic gain information of the nth traffic indicator in the traffic aggregation data of the e-th hospital specialty And the indicator data The indicator data is the data collection result of the nth traffic indicator under the time period corresponding to the traffic aggregation data of the e-th hospital specialty.

[0077] Calculate the statistical features of the traffic data sequence as the second-level diagnosis and treatment traffic features of the traffic indicator, where the statistical features include the mean, variance, standard deviation, and skewness of the traffic data sequence.

[0078] Taking the dynamic information gain as the weight, weight the indicator data to obtain the weighted time series of each traffic indicator. Use the smoothing filtering method to extract the long-term trend term of the weighted time series, and use the Fourier transform method to extract the periodic term of the weighted time series. Take the long-term trend term and the periodic term as the third-level diagnosis and treatment traffic features of the traffic indicator.

[0079] Obtain the traffic data sequence corresponding to the external traffic impact indicator, and calculate the impact perturbation sequence of the external traffic impact indicator as the fourth-level diagnosis and treatment traffic feature.

[0080] Take the first-level to fourth-level diagnosis and treatment traffic features as multi-level diagnosis and treatment traffic features.

[0081] Calculate the impact perturbation sequence of the external traffic impact indicator, including: The form of the impact perturbation sequence is: ; .

[0082] Wherein: represents the impact perturbation sequence,[[]] represents the e-th impact perturbation value in the impact perturbation sequence g,[[]] is the impact perturbation coefficient,[[]] are the mean values of the data collection results of the 6th - 9th traffic indicators in the traffic aggregation data of the e-th hospital specialty in sequence. The 6th - 9th traffic indicators are meteorological conditions, epidemic trends, traffic conditions, and activity information in sequence.

[0083] S3: Construct a hospital specialty diagnosis and treatment traffic prediction model.

[0084] The hospital's specialist diagnosis and treatment traffic prediction model includes an input layer, a hidden state extraction layer, and a fully connected layer. The hospital's specialist diagnosis and treatment traffic prediction model is a long short-term memory neural network structure. The input layer is used to receive multi-layer diagnosis and treatment traffic features. The hidden layer state extraction layer includes a forgetting gate, an input gate, a memory unit, and an output gate. The fully connected layer is used to map the output result of the output gate to obtain the prediction result of the hospital's specialist diagnosis and treatment traffic. The loss function of the model parameters in the hospital's specialist diagnosis and treatment traffic prediction model is the mean square error loss.

[0085] The training process of the hospital's specialist diagnosis and treatment traffic prediction model is as follows: Initialize the model parameters to be optimized , set the current iteration number of the model parameters to t, and the maximum iteration number to Max. Then, the t-th iteration result of the model parameters is ; Construct the loss function of the model parameters : ; ; .

[0086] Among them: represents the loss function of the model parameters , represents the L2 norm, represents the true value of the r-th group of training samples at subsequent times, represents the predicted value of the r-th group of training samples at subsequent times predicted based on the model parameters ; represents the variance in the r-th group of training samples. Each group of training samples is a group of multi-layer diagnosis and treatment traffic features, represents the regularization parameter, represents the variance influence control coefficient, represents the error influence control coefficient, represents the prediction error of using the model parameters to predict the r-th group of samples, represents the sample weight of the r-th group of training samples for the model parameters .

[0087] Specifically, L2 regularization is used to avoid model overfitting. As the training progresses, the regularization parameter gradually decreases, allowing the model to fit the data more freely in the later stage and further reducing the risk of overfitting. Considering the past prediction errors, the sample weights of each sample are dynamically adjusted to increase the importance of the model for samples with prediction errors.

[0088] Iterate the model parameters based on the loss function until the maximum number of iterations is reached. The model parameters have the following iteration formula: ; ; ; ; ; ; .

[0089] Where: represents the learning rate during the t-th iteration process, represents the iteration coefficient, represents the initial coefficient, represents the model parameters of the first moment, represents the model parameters of the second moment, represents the iteration regulation parameter; represents the model parameters of the gradient, represents the gradient direction control coefficient, represents the first-order gradient attenuation value, represents the second-order gradient attenuation value, represents the gradient descent direction.

[0090] Use the first-order gradient attenuation value combined with the gradient direction for gradient descent. For the case where the gradient direction changes greatly, reduce the first moment and reduce the dependence on historical gradients; for the case where the gradient direction is stable, increase the first moment, maintain the optimization direction, and avoid over-relying on historical information when the gradient direction changes greatly.

[0091] S4: Use the hospital specialty diagnosis and treatment traffic prediction model to predict the hospital specialty diagnosis and treatment traffic.

[0092] Using the hospital specialty diagnosis and treatment traffic prediction model to predict the hospital specialty diagnosis and treatment traffic includes: The input layer receives multi-layer diagnosis and treatment traffic features and inputs the multi-layer diagnosis and treatment traffic features into the hidden layer state extraction layer.

[0093] The hidden layer state extraction layer inputs the multi-layer diagnosis and treatment traffic features into the forget gate and sequentially performs operations of input gate convolution, updating the memory unit, and output gate convolution to obtain the hidden information of the multi-layer diagnosis and treatment traffic features.

[0094] The fully connected layer uses weight parameters and bias parameters to map the hidden information of the multi-layer diagnosis and treatment traffic characteristics, and obtains the prediction result of the hospital specialty diagnosis and treatment traffic at subsequent moments.

[0095] As a preferred embodiment of the present invention, based on the prediction result of the hospital specialty diagnosis and treatment traffic at subsequent moments, a hospital specialty operation random forest prediction model is constructed to predict the operation service ability of the hospital specialty. The hospital specialty operation random forest prediction data takes the existing operation status and the prediction result of the hospital specialty diagnosis and treatment traffic as inputs. The operation status includes the human resource status and equipment resource status of the department, specifically including the doctor diagnosis and treatment time allocation and equipment usage time allocation of the hospital specialty. The operation service ability includes the operation efficiency and medical service efficiency of the hospital specialty. Among them, the operation efficiency includes the bed utilization rate, operating room utilization rate, doctor diagnosis and treatment time utilization rate, etc., and the medical service efficiency includes the patient waiting time and treatment satisfaction, etc. The hospital management makes real-time decisions according to the prediction result of the model, such as adjusting resource allocation, optimizing patient treatment arrangements, etc. The actual operation data of the hospital will be continuously fed back to the model as new training data, so that the model can gradually optimize and improve the prediction accuracy. The random forest adopts ensemble learning, enabling the model to integrate the results of multiple decision trees, thereby improving the robustness and accuracy of the prediction. The random forest model randomly selects subsets from the samples by the bootstrap method, making the prediction results of each tree more diverse, and finally obtaining a more stable and accurate prediction result.

[0096] As an embodiment of the present invention, the time series prediction algorithm is the LSTM model algorithm.

[0097] It should be understood that the above embodiments are only for illustration purposes and are not limited by this structure in the scope of the patent application.

[0098] It should be noted that the above serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the term "including" or "comprising" or any other variant thereof in this article is intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, device, article or method. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, device, article or method including the element.

[0099] Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0100] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A method for predicting hospital specialty diagnosis and treatment flow based on big data, characterized in that: The method comprises: S1: Based on the flow index, hospital specialty diagnosis and treatment data and external influence data are collected as hospital specialty flow data, and hospital specialty flow data are aggregated by sliding the time window to obtain the hospital specialty flow aggregation data for each time period; The flow index includes an internal hospital flow impact index and an external flow impact index, and hospital specialist diagnosis and treatment data are collected based on the internal hospital flow impact index, and external impact data are collected based on the external flow impact index; S2: Calculate the dynamic information gain of the flow index based on the hospital specialty flow aggregation data, and use the dynamic information gain of the flow index to extract multi-layer diagnosis and treatment flow characteristics; S3: construct a hospital specialist diagnosis and treatment flow prediction model, wherein the hospital specialist diagnosis and treatment flow prediction model takes multi-layer diagnosis and treatment flow characteristics as input and takes flow prediction results as output; S4: Use the hospital specialist diagnosis and treatment flow prediction model to predict the hospital specialist diagnosis and treatment flow.

2. A method for predicting hospital specialty diagnosis and treatment flow based on big data as claimed in claim 1, characterized in that: The hospital internal traffic influencing indicators include patient registration information, diagnosis information, surgery information, hospitalization information and department load conditions, and the external traffic influencing indicators include meteorological conditions, epidemic trends, traffic conditions and activity information; The registration information describes the number of registrations of hospital specialties; The diagnosis information describes the number of critically ill patients, the surgery information describes the number of patients undergoing surgery, the hospitalization information describes the number of hospitalized patients, and the department load describes the current load of different departments, where the load represents the number of patients visiting the hospital; The collected data under the meteorological condition indicators include temperature and PM2.5, and the temperature and PM2.5 collected under the meteorological condition indicators are normalized respectively, and the normalized results are multiplied as the collected data under the meteorological condition indicators. The collected data under the epidemic trend indicators include whether there is an epidemic. The collected data under the traffic condition indicators are the degree of congestion at the intersections around the hospital. The collected data under the activity information indicators are whether there are large-scale activities in the area where the hospital is located.

3. A method for predicting hospital specialty diagnosis and treatment flow based on big data as claimed in claim 2, characterized in that: The hospital specialist flow data is aggregated by using a time window sliding method, including: The hospital specialist flow data is divided into hourly granularity to obtain hourly hospital specialist flow data, and the hourly flow data of each flow indicator is extracted, wherein the hourly flow data of the nth flow indicator is: ； in: Indicates the hourly traffic data of the nth traffic indicator. ,The 1st to 9th traffic indicators are registration information, diagnosis information, ,operation information, hospitalization information, department load, meteorological conditions, ,epidemic trends, traffic conditions, and activity information; It represents the collection data of the nth traffic indicator at the mth hour. , M represents the time series length; Calculate the fluctuation degree of hourly traffic data of different traffic indicators, integrate the fluctuation degree and data distribution characteristics of hourly traffic data, and generate the window weight of traffic indicators, where the data distribution characteristics include skewness and kurtosis; Calculate the time window size required for each traffic indicator, use the window weight for weighted processing, obtain the length L of the time window, and construct the time window; The hospital specialist flow data is processed by sliding the time window using the time window to obtain the hospital specialist flow data under different time windows, and the hospital specialist flow data under adjacent time windows do not overlap each other. The hospital specialist flow data under each time window is used as the hospital specialist flow aggregate data of the corresponding time period. The number of hospital specialist flow aggregate data is .

4. A method for predicting hospital specialty diagnosis and treatment flow based on big data as claimed in claim 3, characterized in that: The dynamic information gain of the flow index is calculated based on the hospital specialist flow aggregation data, including the calculation process of the dynamic information gain of the nth flow index as follows: Calculate the external influence condition entropy and information entropy of different traffic indicators in each hospital specialty traffic aggregation data, and use the difference between the external influence condition entropy and the information entropy as the information gain of the traffic indicator in the hospital specialty traffic aggregation data; the information entropy of the nth traffic indicator in the eth hospital specialty traffic aggregation data is : ； in: Indicates the standard value of the preset nth flow index; It represents the data collection result of the nth traffic indicator in the i-th hour of the traffic aggregation data of the e-th hospital specialty. , Indicates the number of aggregated data of specialty traffic in the hospital; The external influence condition entropy of the nth flow index in the aggregated flow data of the eth hospital specialty is: : ； in: It represents the mean of the data collection results of the nth type of flow index in the aggregated data of the specialty flow of the eth hospital when the meteorological conditions, epidemic trends, traffic conditions, and activity information are all lower than the corresponding influencing index threshold; The information gain of the nth flow indicator in the aggregated flow data of the eth hospital specialty is: ; Calculate the dynamic change value of information gain of the flow index in the aggregated data of specialized flow of adjacent hospitals, and standardize the dynamic change value of information gain as the dynamic information gain of the flow index in different time periods; The dynamic information gain of the nth flow indicator in the time period corresponding to the aggregated flow data of the eth hospital specialty is: ； in: It represents the dynamic information gain of the nth flow indicator in the time period corresponding to the aggregated flow data of the eth hospital specialty.

5. The method for predicting hospital specialty diagnosis and treatment flow based on big data as claimed in claim 1, characterized in that: The dynamic information gain of the flow index is used to extract multi-layer diagnosis and treatment flow characteristics, including: Extracting the flow data sequence corresponding to the flow index from the hospital specialist flow aggregation data based on dynamic information gain, and using the flow data sequence as the first-level diagnosis and treatment flow feature of the flow index; Calculate and obtain the statistical characteristics of the flow data sequence as the second-level diagnosis and treatment flow characteristics of the flow indicator, wherein the statistical characteristics include the mean, variance, standard deviation and skewness of the flow data sequence; The dynamic information gain is used as a weight to weight the indicator data to obtain a weighted time series sequence of each flow indicator, a smoothing filter method is used to extract the long-term trend item of the weighted time series sequence, and a Fourier transform method is used to extract the periodic item of the weighted time series sequence, and the long-term trend item and the periodic item are used as the third-layer diagnosis and treatment flow characteristics of the flow indicator; Obtain the flow data sequence corresponding to the external flow influencing index, and calculate the influence disturbance sequence of the external flow influencing index as the fourth-layer diagnosis and treatment flow feature; The first-layer medical treatment flow characteristics to the fourth-layer medical treatment flow characteristics are used as multi-layer medical treatment flow characteristics.

6. A method for predicting hospital specialty diagnosis and treatment flow based on big data as claimed in claim 5, characterized in that: The influence disturbance sequence of external flow influencing indicators is calculated, including: The form of the influencing disturbance sequence is: ； ； in: represents the impact disturbance sequence, represents the e-th impact disturbance value in the impact disturbance sequence g, To influence the disturbance coefficient, They are the means of the data collection results of the 6th to 9th flow indicators in the aggregated data of the specialty flow of the eth hospital, respectively. The 6th to 9th flow indicators are meteorological conditions, epidemic trends, traffic conditions, and activity information, respectively.

7. A method for predicting hospital specialty diagnosis and treatment flow based on big data as claimed in claim 1, characterized in that: The hospital specialized diagnosis and treatment flow prediction model includes an input layer, a hidden state extraction layer and a fully connected layer. The hospital specialized diagnosis and treatment flow prediction model is a long short-term memory neural network structure. The input layer is used to receive multi-layer diagnosis and treatment flow characteristics. The hidden layer state extraction layer includes a forgetting gate, an input gate, a memory unit and an output gate. The fully connected layer is used to map the output result of the output gate to obtain the prediction result of the hospital specialized diagnosis and treatment flow. The loss function of the model parameters in the hospital specialized diagnosis and treatment flow prediction model is the mean square error loss.

8. A method for predicting hospital specialty diagnosis and treatment flow based on big data as claimed in claim 7, characterized in that: The hospital specialist diagnosis and treatment flow prediction model is used to predict the hospital specialist diagnosis and treatment flow, including: The input layer receives the multi-layer diagnosis and treatment flow characteristics and inputs the multi-layer diagnosis and treatment flow characteristics into the hidden layer state extraction layer; The hidden layer state extraction layer inputs the multi-layer diagnosis and treatment flow characteristics into the forget gate, and sequentially performs input gate convolution, updates the memory unit, and output gate convolution operations to obtain hidden information of the multi-layer diagnosis and treatment flow characteristics; The fully connected layer uses weight parameters and bias parameters to map the hidden information of multi-layer diagnosis and treatment flow characteristics to obtain the prediction results of hospital specialty diagnosis and treatment flow at subsequent times.

Images