An intelligent online reservation management method

Abstract

The present application discloses an intelligent online reservation management method, which relates to the field of online reservation technology, including: establishing a conference room information database to store static information of the conference room; collecting historical reservation data sets; using data analysis algorithms to establish and train a conference room usage demand prediction model, the prediction model is used to predict the usage demand of the target conference room within a preset time period; collecting real-time status data; associating the collected real-time status data of each conference room with the static information in the database to generate dynamic data for each conference room; receiving reservation requests submitted by users; performing conflict detection on the reserved conference room and the reserved time period through the prediction model and dynamic data to determine whether the user's reservation request conflicts; when there is a conflict, generating a conflict-free alternative reservation plan through the prediction model and dynamic data. In view of the problem of inaccurate conflict detection of online reserved conference rooms in the prior art, the present application improves the accuracy of conflict detection.

Description

Technical Field

[0001] The present application relates to the technical field of online appointment, and in particular to an intelligent online appointment management method. Background Art

[0002] With the rapid development of information technology, more and more medical institutions have begun to use online reservation systems to manage conference room resources. Medical meetings are frequent and important, including multidisciplinary consultations, discussions on difficult cases, academic seminars, etc. The rational and efficient use of limited conference room resources is crucial to the smooth development of medical work. However, the existing online reservation system still has many shortcomings in conference room management.

[0003] At present, the conference room reservation system used by most medical institutions usually only provides simple reservation and query functions, and lacks intelligent demand prediction and conflict detection capabilities. On the one hand, the system cannot accurately predict the demand for conference room use in different time periods, resulting in unreasonable allocation of conference room resources, sometimes idle resources, and sometimes insufficient supply; on the other hand, the system lacks an effective conflict detection mechanism and cannot promptly detect and handle reservation conflicts for the same conference room in the same time period by multiple departments or individuals, resulting in frequent reservation conflicts, which seriously affects the normal arrangement of medical work. Although some hospitals are aware of these problems, they are limited by technical means and can only rely on manual coordination to resolve conflicts, which is inefficient and error-prone.

[0004] In the related art, for example, Chinese patent document CN111768015B provides an online reservation method and related equipment, which are applied to electronic devices, and the method includes: obtaining conference room reservation information, the conference room reservation information is used to reserve a conference room; determining the target conference room based on the conference room reservation information and the first building information model BIM of the target building; outputting the information of the target conference room; updating the first BIM to the second BIM based on the information of the target conference room; and displaying the second BIM. However, this solution mainly relies on the static information of BIM, so the real-time conflict detection accuracy of this solution needs to be further improved. Summary of the invention

[0005] In view of the high frequency of conflicts in online conference room reservations in the prior art, the present application provides an intelligent online reservation management method, which uses time series analysis and Gaussian mixture models to achieve accurate conflict detection and reduce the frequency of conflicts.

[0006] The purpose of this application is achieved through the following technical solutions.

[0007] This specification provides an intelligent online reservation management method, including: Step 1, establishing a conference room information database. The system collects static information of each conference room of a medical institution, including basic attributes such as conference room name and location, and stores it in the conference room information database. The database provides the necessary basic data support for subsequent reservation management. Step 2: Collect historical reservation data sets. The system extracts historical reservation records within a certain time range from the existing reservation system. These records contain key fields such as reservation time, conference room information and actual usage time to form a complete historical reservation data set. This data set is an important input for building a demand forecasting model. Step 3: Build a conference room usage demand forecasting model. The system uses a time series analysis algorithm, such as an ARIMA model, to model and train the historical reservation data set. By analyzing the usage of each conference room in different time periods, the prediction model can accurately estimate the usage demand of the target conference room in a specific time period in the future. The prediction result provides a decision-making basis for conflict detection. Step 4: Collect real-time status data of the conference room. The system deploys status sensors in each conference room to collect the usage status data of the conference room in real time. The sensor can identify whether the conference room is currently idle and report the status information to the data processing module in real time. The introduction of real-time data can dynamically reflect the actual usage of the conference room. Step 5: Generate dynamic data of conference rooms. The system associates the collected real-time status data with the static information in the conference room information database to generate a dynamic data view of each conference room. The dynamic data combines the static attributes and real-time usage status of the conference room, providing a more comprehensive and accurate basis for subsequent conflict detection. Step 6: Receive user reservation requests. The user submits a reservation request through the client interface, which contains key elements such as user information, reserved conference room, and reserved time period. After receiving the reservation request, the system extracts the key information and passes it to the reservation conflict detection module. Step 7: Reservation conflict detection. The system uses the prediction model to estimate the demand for the conference room and time period reserved by the user, and combines the dynamic data of the conference room to determine whether the reservation request conflicts with other existing reservations. Through accurate conflict detection, the probability of conference room reservation conflicts can be greatly reduced. Step 8: Generate alternative reservation plans. When the system detects a reservation conflict, it automatically calls the alternative plan generation module. This module uses the prediction model and dynamic data to search for available alternative conference rooms and time periods, and evaluates the rationality of the alternative plan. Finally, the system recommends the preferred non-conflicting alternative reservation plan to the user for selection and confirmation.

[0008] Furthermore, the time series analysis algorithm is used to accurately predict the usage demand of different conference rooms in different time periods through in-depth mining and modeling of historical reservation data. First, the collected historical reservation data set is cleaned to remove incomplete, inconsistent or obviously abnormal data records to improve data quality. Then, according to the reservation time field, the data set is divided into several time intervals, such as daily, weekly, monthly, etc., for subsequent time series analysis. At the same time, the system will also extract key time features in the data set, such as holidays and weekends, which may have an important impact on the usage demand of conference rooms. Construct an ARIMA model. For each conference room and each time period, the system will use the ARIMA (autoregressive moving average) algorithm to build an independent ARIMA model. ARIMA is a classic time series prediction method. By modeling the autoregression, difference and moving average of historical data, it can well fit and predict future time series values. The system will adjust the order (p, d, q) parameters of the ARIMA model according to the characteristics of the data to obtain the best fitting effect.

[0009] In order to further optimize the ARIMA model, the system introduces the Gaussian mixture algorithm to evaluate the model parameters. The data of the Gaussian mixture algorithm obeys a mixture of multiple Gaussian distributions, and the mean, variance and weight parameters of each Gaussian component are estimated iteratively. The system takes the residuals of ARIMA models of different orders as input and uses the Gaussian mixture algorithm to fit the distribution of the residuals. By comparing the parameter distributions of different orders, it can be found which order of models can better capture the inherent laws of the data. After obtaining the parameter distributions of ARIMA models of different orders, the system uses AIC (Akaike Information Criterion) with MAPE (mean absolute percentage error) to evaluate the prediction accuracy of the model. The AIC criterion comprehensively considers the fit and complexity of the model, and can balance the overfitting and underfitting risks of the model. On the basis of AIC, the system adds the MAPE indicator, which is the average percentage error between the predicted value and the actual value. MAPE can intuitively measure the prediction accuracy of the model. The smaller the value, the better the prediction effect. By comparing the AIC-MAPE scores of models of different orders, the model with the best prediction accuracy can be selected. For each time period of each conference room, the system will eventually select the ARIMA model with the lowest AIC-MAPE score as the corresponding prediction model. This means that for different prediction scenarios, the system will adaptively select the optimal model structure and parameters. In actual applications, when a user submits a reservation request, the system will call the corresponding prediction model, combine the current time period and conference room, and predict the demand for that time period, thereby providing a key basis for conflict detection and alternative solution generation.

[0010] Furthermore, the Gaussian mixture algorithm is used to evaluate the parameter distribution of ARIMA models of different orders. The system uses the expectation maximization method (EM) to estimate the key parameters of the Gaussian mixture model, including the mean, variance, and weight. For each ARIMA model, the system will establish a corresponding Gaussian mixture distribution model to fit the distribution of its parameters. The key parameters of the ARIMA model include the autoregressive order p and the moving average order q, which reflect the autocorrelation and lag effects of the time series. The parameters of the Gaussian mixture distribution model obey the weighted mixture of K Gaussian distributions, each of which has its own independent mean, variance, and weight coefficient. Before starting the EM algorithm, the system will randomly initialize the parameters of each Gaussian mixture distribution model. The mean parameter usually selects K random sample points of the data, the variance parameter can be set to the overall variance of the data, and the weight parameter is usually set to 1 / K, that is, each Gaussian component contributes equally to the data at the beginning. These initial values ​​provide a starting point for the EM algorithm, but do not affect the final convergence result of the algorithm. In each iteration of the EM algorithm, the first step is to calculate the posterior probability of each parameter on each Gaussian component. The posterior probability represents the probability that a data point belongs to each Gaussian component under the current model parameters. According to Bayes' theorem, the posterior probability is equal to the prior probability (i.e., weight) of the Gaussian component multiplied by the conditional probability of the data point on the component, divided by the sum of the weighted conditional probabilities of all components. This step can be regarded as soft classification of the data under the current model. Based on the posterior probability calculated in step E, the system updates the model parameters in step M to maximize the likelihood function of the entire data set. The likelihood function measures the probability of the data appearing under the current parameters. By taking the derivative and setting the derivative to zero, the update formula for the mean, variance, and weight parameters can be obtained. The new mean is equal to the weighted average of each data point, and the weight is its posterior probability; the new variance is equal to the weighted average of the square of the difference between each data point and the mean; the new weight is equal to the sum of the posterior probabilities of each Gaussian component divided by the total number of data points. The system repeats steps E and M to continuously update the model parameters until the parameters converge or the preset maximum number of iterations is reached. The convergence criterion can be that the parameter change is less than a certain threshold, or the increase in the likelihood function is less than a certain threshold. When the algorithm converges, it means that under the current Gaussian mixture distribution model, the likelihood function of the data has reached a local maximum, and the model has well fitted the distribution of the parameters. For each ARIMA model, the Gaussian mixture algorithm will eventually output a converged Gaussian mixture distribution model as the distribution of its parameters p and q. This model contains the mean, variance, and weight parameters of K Gaussian components, which can describe the probability of occurrence of combinations of different orders. The system will use these parameter distribution models, combined with the AIC-MAPE criterion, to evaluate and select the optimal ARIMA model for the prediction of conference room usage demand.

[0011] Furthermore, the AIC criterion with MAPE is used to evaluate and screen the prediction accuracy of ARIMA models of different orders. This criterion takes into account the complexity and prediction error of the model, and obtains a comprehensive score by weighted summation to balance the fitting ability and generalization ability of the model. Based on the parameter distribution of the ARIMA model obtained by the Gaussian mixture algorithm, the system first calculates the model complexity corresponding to each order combination (p, d, q). Model complexity reflects the number of parameters and structural complexity of the model. For the ARIMA (p, d, q) model, its complexity can be measured by p+q, which is the sum of the autoregressive order and the moving average order. The higher the order, the more parameters the model contains, and the greater the complexity. The system will calculate the complexity of each order combination based on the parameter distribution. In order to evaluate the prediction accuracy of different order combinations, the system uses the MAPE (mean absolute percentage error) indicator. MAPE measures the average percentage error between the predicted value and the true value. The smaller the value, the better the prediction effect. For each order combination, the system will train the corresponding ARIMA model and calculate its MAPE value on the validation set. This step can reveal the performance of models of different complexity in actual prediction tasks. After obtaining the model complexity and MAPE, the system uses a weighted summation method to calculate the weighted AIC value of each order combination. The traditional AIC criterion only considers the complexity and likelihood function value of the model, and the present application introduces MAPE on this basis to more comprehensively evaluate the predictive ability of the model. The system sets the weight of MAPE to 30% to 50% of the total AIC weight. This weight range is based on experience and experimental results, and can better balance the influence of complexity and accuracy. The calculation formula of the weighted AIC value is as follows: weighted AIC = complexity + (1-MAPE weight) × (-2 × log-likelihood function) + MAPE weight × MAPE. According to the weighted AIC value, the system can compare the comprehensive performance of different order combinations and select the order combination with the smallest weighted AIC value as the optimal ARIMA model order. The smaller the weighted AIC value, the better the balance between complexity and prediction accuracy. The model is neither too simple nor overfitted. After determining the optimal order combination, the system will also select the specific ARIMA model with the highest prediction accuracy (i.e., the smallest MAPE) under this order as the final model for conference room demand forecasting. Through the above steps, the system finally obtains an ARIMA model that achieves the best balance between complexity and accuracy and has the smallest prediction error. This model can fit historical data well and make reliable predictions about future conference room usage needs. In the actual appointment management process, when a user submits an appointment request, the system will call the optimal model, combine the current time period and conference room, predict the corresponding demand, and perform conflict detection and recommend alternative solutions based on this.

[0012] Furthermore, the Gaussian mixture algorithm is used to estimate the parameter distribution of the ARIMA model, and the joint distribution of the p-order and q-order is constructed based on the estimated parameter distribution, thereby generating candidate ARIMA model order combinations. For each ARIMA model, the system will establish a corresponding Gaussian mixture distribution model for its key parameters. The key parameters here include the autoregressive order p and the moving average order q. Each parameter of the Gaussian mixture distribution model obeys a weighted mixture of K Gaussian distributions, and each Gaussian component has its own independent mean, variance and weight parameters. In this way, the system can flexibly fit the true distribution of the parameters and capture the multi-peak or asymmetric characteristics that may exist therein. Before applying the Gaussian mixture algorithm, the system will randomly initialize the parameters of each Gaussian mixture distribution model. The mean parameter can select K random sample points of the data, the variance parameter can be set to the overall variance of the data, and the weight parameter is usually set to 1 / K, that is, each Gaussian component contributes equally to the data at the beginning. These initial values ​​provide a starting point for the iterative optimization of the algorithm. In each iteration of the Gaussian mixture algorithm, the system first calculates the posterior probability of each parameter on each Gaussian component. The posterior probability represents the probability that a data point belongs to each Gaussian component under the current model parameters. This step can be regarded as soft classification of data under the current model. The calculation of the posterior probability is based on Bayes' theorem, which comprehensively considers the prior probability (i.e., weight) of the Gaussian component and the conditional probability of the data point on each component. According to the posterior probability calculated in step E, the system updates the model parameters in step M to maximize the likelihood function of the entire data set. By taking the derivative and setting the derivative to zero, the update formula for the mean, variance, and weight parameters can be obtained. The new mean is equal to the weighted average of each data point, the new variance is equal to the weighted average of the square of the difference between each data point and the mean, and the new weight is equal to the sum of the posterior probabilities of each Gaussian component divided by the total number of data points. This step can be regarded as adjusting the model parameters to better fit the data under the condition of fixed data soft classification. The system repeats step E and step M, and continuously updates the model parameters until the preset convergence conditions are met. The convergence condition can be that the parameter change is less than a certain threshold, or the increase in the likelihood function is less than a certain threshold, or the maximum number of iterations can be set. When the algorithm converges, the Gaussian mixture distribution model has well fitted the distribution of parameters. For each ARIMA model, the Gaussian mixture algorithm will eventually output a converged Gaussian mixture distribution model as the distribution of its autoregressive order p and moving average order q. This model contains the mean, variance, and weight parameters of K Gaussian components, which can fully describe the probability distribution of p and q. Based on the p-order and q-order distribution models obtained by the Gaussian mixture algorithm, the system further constructs their joint distribution. The joint distribution describes the probability that p and q take different values ​​at the same time, reflecting the correlation between the two orders. In practice, there is a certain degree of independence between the p-order and the q-order, so that the joint distribution can be approximated as the product of two marginal distributions.In order to obtain candidate ARIMA model order combinations, the system uses the Monte Carlo algorithm to sample from the joint distribution of the p-order and q-order. The Monte Carlo algorithm generates a large number of order combination samples based on the probability density function of the distribution by random sampling. These samples cover different areas of the order space and reflect the characteristics of the parameter distribution. The system can set a sampling number and select a part of the sampled order combinations as candidates for subsequent model evaluation and screening.

[0013] Furthermore, in order to evaluate the complexity of different ARIMA model order combinations, the system introduces a complexity measurement method based on information entropy. By calculating the parameter entropy value of each order combination, the amount of information contained in the model can be quantitatively measured, thereby reflecting its complexity. For the ARIMA model corresponding to each order combination, the system first calculates the probability distribution of its parameters in the Gaussian mixture distribution model. Based on the parameter distribution model obtained by the Gaussian mixture algorithm, the probability density of each parameter under different values ​​can be obtained. In order to facilitate the subsequent information entropy calculation, the system discretizes the continuous probability density function to obtain the discrete probability distribution of the parameter. The discretization method can be equal interval binning or quantile-based binning, and the number of bins can be set according to the characteristics of the data and computing performance. The discrete probability distribution of each parameter is regarded as a discrete random variable, and the system calculates its corresponding information entropy. Information entropy measures the uncertainty of random variables and reflects the amount of information contained in the variable. For a discrete random variable X, the information entropy H(X) is calculated as: H(X) = -∑[p(x)×log2p(x)], where p(x) represents the probability that the random variable X takes the value x, and the summation covers all possible values ​​of X. The unit of information entropy is bit, and the base of the logarithm is 2. The larger the information entropy, the higher the uncertainty of the random variable and the greater the amount of information contained. For each order combination, the system sums the information entropy of all its parameters to obtain the parameter entropy value of the combination. The parameter entropy value reflects the amount of information contained in the entire order combination, taking into account the uncertainty of each parameter. The order combination (p, q) contains k parameters, and the parameter entropy value E(p, q) is calculated as: E(p, q) = ∑[H(X i )], i = 1, 2, ..., k, where Xi represents the discrete random variable corresponding to the i-th parameter, H(X i) represents its information entropy. The larger the parameter entropy value, the richer the information contained in the order combination, and the higher the complexity of the model. The system uses the parameter entropy value of each order combination as the complexity measure of its corresponding model. The parameter entropy value provides a quantitative, information-theoretic-based complexity evaluation method, avoiding simple measurements that rely solely on the number of parameters. By comparing the parameter entropy values ​​of different order combinations, it is possible to intuitively determine which combinations correspond to higher model complexity and contain more information. When evaluating and screening models, complexity measures can be combined with other indicators (such as prediction accuracy) to balance the model's fitting ability and generalization ability.

[0014] Furthermore, a conflict detection mechanism based on the state matrix is ​​adopted, and the demand forecast output by the prediction model and the real-time status data of the conference room are comprehensively used to calculate the idle probability of each conference room in each time period, so as to determine whether there is a conflict between the reservation requests submitted by the user. The system first obtains the demand forecast of the conference room in different time periods from the prediction model. These forecasts reflect the potential demand for the conference room in each time period. At the same time, the system obtains real-time status data from the state sensors deployed in the conference room, such as the occupancy status and environmental parameters of the conference room. These status data provide the actual current usage of the conference room. In order to establish the correlation between the demand forecast and the real-time status data, the system uses the mutual information algorithm to calculate their correlation. Mutual information measures the mutual dependence between two random variables. The larger the value, the stronger the correlation. Under the Bayesian framework, the system takes the demand forecast as the observed quantity and the real-time status data as the random variable. By calculating their joint probability distribution and marginal probability distribution, the mutual information value is obtained as the correlation. Here, the demand forecast plays the role of prior knowledge and imposes certain constraints on the distribution of the status data. According to the calculated correlation, the system constructs an association matrix to represent the correlation between the demand forecast and the real-time status data. The rows of the association matrix correspond to different demand forecasts, the columns correspond to different real-time status data, and the values ​​of the matrix elements are the corresponding associations. The association matrix provides a compact representation that facilitates subsequent calculations and queries. In order to combine the real-time status data with the association, the system first normalizes the status data and maps it to the interval of 0 to 1. Then, the system multiplies the mapped status data with the corresponding association in the association matrix to obtain the idle probability of each conference room in each time period. The higher the idle probability, the more likely the conference room is to be idle in that time period, and vice versa, the conference room may be occupied. Based on the calculated idle probability, the system constructs a state matrix to represent the idle status of different conference rooms in different time periods. The rows of the state matrix correspond to different time periods, the columns correspond to different conference rooms, and the values ​​of the matrix elements are the idle probabilities of the corresponding conference rooms in the corresponding time periods. The state matrix provides an intuitive representation that facilitates conflict detection and query. When a user submits a reservation request, the system receives and parses the request to obtain the reserved conference room and reservation time period specified by the user. This information will be used for conflict detection in the state matrix. According to the user's reserved conference room and reservation time period, the system searches for the corresponding matrix element in the state matrix to obtain the idle probability of the conference room in the time period. Then, the system compares the idle probability with the preset threshold. If the idle probability is less than the threshold, it means that the conference room is likely to be occupied in the time period and there is a reservation conflict; conversely, if the idle probability is greater than or equal to the threshold, it means that the conference room is likely to be idle in the time period and the reservation can be successful.The threshold setting can be adjusted according to actual needs and experience to balance the reservation success rate and conference room utilization rate. Based on the results of conflict detection, the system returns corresponding information to the user. If a conflict is detected, the system will prompt the user that the reservation request cannot be met and that the user can consider changing the reservation time or conference room; if no conflict is detected, the system will confirm that the reservation is successful and update the reservation status of the conference room.

[0015] Furthermore, the mutual information algorithm is used to calculate the correlation between the demand forecast and the real-time status data. Through data sampling, binning, probability distribution calculation and other steps, the mutual information value is obtained as a quantitative indicator of the correlation. In order to calculate the mutual information between the demand forecast and the real-time status data, the system first samples the two data sources so that they have the same time granularity. According to the preset time interval (such as every 5 minutes), the system extracts the values ​​of the corresponding time from the demand forecast and the real-time status data to form two data sequences of equal length. This step ensures the alignment of the two variables in the time dimension and lays the foundation for subsequent analysis. Since the calculation of mutual information is based on discrete variables, the system needs to bin the sampled data sequence to convert the continuous variable into a discrete variable. The purpose of binning is to divide the value range of each variable into several intervals, each interval corresponding to a discrete value. The binning method can be selected according to the distribution characteristics of the data and business needs. Common methods include equal-width binning, equal-frequency binning, cluster binning, etc. After binning, both the demand forecast and the real-time status data are converted into discrete variables, and the value set of each variable is limited. On the binned data, the system calculates the joint probability distribution and marginal probability distribution of the demand forecast and real-time status data. The joint probability distribution describes the probability that two variables take specific values ​​at the same time, that is, P(X=x, Y=y). The system obtains an estimate of the joint probability distribution by counting the frequency of the two variables in each value combination and dividing it by the total number of samples. The marginal probability distribution describes the probability that each variable takes a specific value alone, that is, P(X=x) and P(Y=y). The system obtains an estimate of the marginal probability distribution by counting the frequency of each variable in each value and dividing it by the total number of samples. Based on the calculated joint probability distribution and marginal probability distribution, the system uses the mutual information formula to calculate the mutual information value I(X;Y) of the demand forecast X and the real-time status data Y. The mutual information formula is as follows: Among them, p(x,y) represents the joint probability distribution of X and Y, p(x) and p(y) represent the marginal probability distribution of X and Y respectively, and the summation range covers all combinations of values ​​of X and Y. The mutual information value I(X;Y) measures the correlation between X and Y, that is, how much information about the value of another variable can be provided by knowing the value of one variable. The larger I(X;Y), the higher the correlation between X and Y, and knowing the value of one variable can better predict the value of the other variable; the smaller I(X;Y), the lower the correlation between X and Y, and the two variables tend to be independent. In order to facilitate the comparison and interpretation between different variables, the system normalizes the calculated mutual information value. Common normalization methods include dividing by the minimum value of the entropy of the two variables, dividing by the square root of the entropy of the two variables, etc. The normalized mutual information value ranges from 0 to 1, and the larger the value, the higher the correlation. The normalized mutual information value can be used as the final quantitative indicator of the correlation between the demand forecast and the real-time status data.

[0016] Further, the joint probability distribution and marginal probability distribution of the demand forecast amount and the real-time status data are calculated by the kernel density estimation method, and the probability density function estimation of the continuous variable is obtained by frequency distribution statistics and kernel function fitting. According to the demand forecast amount X and the real-time status data Y after binning, the system calculates the number of samples in each interval to obtain the frequency distribution. The frequency distribution reflects the distribution of samples in each interval and provides basic data for subsequent probability density estimation. For the demand forecast amount X and the real-time status data Y, the system calculates their frequency distribution in their respective intervals, that is, the number of samples in each interval. This step can be achieved by a simple counting operation. After obtaining the frequency distribution of the demand forecast amount X and the real-time status data Y, the system estimates their marginal probability distribution using the kernel density estimation method. Kernel density estimation is a non-parametric estimation method that places a kernel function around the sample point and superimposes all kernel functions to obtain an estimate of the probability density function. Commonly used kernel functions include Gaussian kernel, Epanechnikov kernel, triangular kernel, etc. In this application, the system selects the Epanechnikov kernel function for estimation. For the demand forecast X, the system estimates the probability density function f(x) based on its frequency distribution using the Epanechnikov kernel function, and uses f(x) as the marginal probability distribution of X. Similarly, for the real-time status data Y, the system estimates the probability density function g(y) based on its frequency distribution using the Epanechnikov kernel function, and uses g(y) as the marginal probability distribution of Y. In order to estimate the joint probability distribution of the demand forecast X and the real-time status data Y, the system first calculates their two-dimensional frequency distribution. The two-dimensional frequency distribution reflects the distribution of samples in each interval combination of X and Y. According to the binned X and Y, the system counts the number of samples in each interval combination to obtain a two-dimensional frequency matrix. This step can be achieved through nested loops or efficient matrix operations. After obtaining the two-dimensional frequency distribution of the demand forecast X and the real-time status data Y, the system estimates their joint probability distribution using the kernel density estimation method. Similar to the estimation of the marginal probability distribution, the estimation of the joint probability distribution also uses the kernel function for fitting, except that the kernel function becomes a two-dimensional form. In this application, the system selects a two-dimensional Gaussian kernel function for estimation. According to the two-dimensional frequency distribution, the system uses the two-dimensional Gaussian kernel function to estimate the joint probability density function h(x,y) of X and Y, and uses h(x,y) as the joint probability distribution of X and Y. The two-dimensional Gaussian kernel function can well capture the correlation and nonlinear relationship between variables.

[0017] Furthermore, in order to adapt to changes in data distribution and improve the accuracy of estimation, the system uses a dynamic binning method to bin the sampled data sequence. Dynamic binning adaptively adjusts the interval division according to the distribution characteristics of the data, so that the distribution of samples in each interval is as uniform as possible. Common dynamic binning methods include quantile-based binning, cluster-based binning, etc. Through dynamic binning, the system can better capture the local characteristics of the data and improve the reliability of probability distribution estimation.

[0018] Compared with the prior art, the advantages of this application are:

[0019] Through time series analysis and Gaussian mixture model, this solution builds an accurate conference room usage demand prediction model. This model can accurately predict the conference room usage demand in different time periods, provide a basis for the reasonable scheduling of conference room resources, and maximize the utilization rate of conference rooms while ensuring the normal convening of meetings and avoiding resource waste.

[0020] The ARIMA model in time series analysis is used to predict the demand for conference room use. The Gaussian mixture model is introduced to evaluate the distribution characteristics of the ARIMA model parameters, and the AIC criterion with MAPE is used to select the best prediction model, which improves the accuracy of demand prediction.

[0021] The real-time status data of the conference room is introduced into the appointment management. The correlation matrix between the demand forecast and the real-time status data is constructed through the mutual information theory. On this basis, the dynamic idle probability of the conference room is calculated, and the accurate detection of appointment conflicts is achieved. This method can timely discover the appointment conflicts of multiple departments or individuals for the same conference room at the same time. Through accurate conflict detection and intelligent recommendation of alternative solutions, the frequency of meeting appointment conflicts in medical institutions can be greatly reduced.

[0022] This solution automates appointment management through intelligent algorithms, including demand forecasting, conflict detection, and generation of alternative solutions, minimizing human intervention and significantly improving appointment management efficiency.

[0023] In the mutual information calculation process, an adaptive dynamic binning method is proposed, which can perform adaptive binning according to the data distribution characteristics and construct probability distribution by combining kernel density estimation, which improves the rationality and accuracy of mutual information calculation and strengthens data security and privacy protection. Dynamic binning can avoid direct transmission and storage of original data, prevent sensitive information leakage to a certain extent, and protect the data privacy of hospitals and patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0025] Figure 1 is an exemplary flow chart of an intelligent online appointment management method according to some embodiments of this specification;

[0026] Figure 2 is an exemplary flow chart for obtaining an optimal ARIMA model according to some embodiments of this specification;

[0027] Figure 3 is an exemplary flow chart of obtaining an association matrix according to some embodiments of this specification;

[0028] Figure 4 This is an exemplary flow chart of obtaining the conference room vacancy probability according to some embodiments of this specification. DETAILED DESCRIPTION

[0029] The method and system provided in the embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0030] Figure 1 This is an exemplary flow chart of an intelligent online reservation management method shown in some embodiments of the present specification, establishing a conference room information database to store static information of the conference room, the static information including the name and location of the conference room; collecting historical reservation data sets, the historical reservation data sets including reservation time, conference room information and usage time; based on the collected historical reservation data sets, using a time series analysis algorithm to establish and train a conference room usage demand prediction model, the prediction model is used to predict the usage demand of a target conference room within a preset time period; setting a status sensor in the conference room to collect real-time status data, the status data including the usage status of the conference room, the usage status indicating whether the conference room is idle; associating the collected real-time status data of each conference room with the static information in the database to generate dynamic data for each conference room; receiving a reservation request submitted by a user, the reservation request including user information, a reserved conference room and a reservation time period; performing conflict detection on the reserved conference room and the reservation time period through the prediction model and dynamic data to determine whether the user's reservation request conflicts; when there is a conflict, generating a conflict-free alternative reservation plan through the prediction model and dynamic data.

[0031] Collect static information of each conference room in the medical institution, including conference room name, location, capacity, facilities and equipment, etc., build a conference room information database, and provide basic data for subsequent appointment management. Extract historical appointment data within a certain time range from the existing appointment system of the medical institution, including appointment time, appointment department / personnel, meeting theme, actual usage time, etc., to form a complete historical appointment data set.

[0032] Figure 2 This is an exemplary flow chart for obtaining the optimal ARIMA model according to some embodiments of this specification. The historical reservation data set D contains the reservation records of conference rooms R1 to R10 in each time period T1 to T6 in the past year, with a sample size of 10,000. The data fields include reservation time, conference room number, reservation personnel, etc. The historical reservation data set D1 contains the reservation records of conference room R1 in time period T1 for the past 100 days. The number of reservations per day is as follows: [23, 19, 26, 17, 20, 28, 24, 22, 25, 21, ..., 18, 27, 23, 19, 25]. The system cleans the historical reservation data set D, removes outliers and missing values, and obtains a clean data set D1. Then, the system divides the data set D1 according to the preset time interval (such as 6 time periods T1 to T6 per day) and extracts time features (such as week, month, etc.). The divided data set D2 contains the usage records of each conference room in each time period and the corresponding time features. For each conference room (such as R1) and each time period (such as T1), the system uses the ARIMA algorithm to build multiple ARIMA models of different orders. ARIMA (AutoRegressiveIntegratedMoving Average) is a classic time series prediction model that characterizes the characteristics of time series through three parts: autoregression (AR), difference (I), and moving average (MA). For the data of conference room R1 in time period T1, the system built a total of 12 ARIMA models with order p ranging from 1 to 3 and order q ranging from 0 to 2. These models form a candidate model set M. For the data of conference room R1 in time period T1, the system constructed the following 12 ARIMA models: ARIMA(1,0,0), ARIMA(1,0,1), ARIMA(1,0,2), ARIMA(1,1,0), ARIMA(1,1,1), ARIMA(1,1,2), ARIMA(2,0,0), ARIMA(2,0,1), ARIMA(2,0,2), ARIMA(2,1,0), ARIMA(2,1,1), ARIMA(2,1,2). The order p of these models ranges from 1 to 2, and the order q ranges from 0 to 2, forming a candidate model set M.

[0033] In order to evaluate the parameter distribution of ARIMA models of different orders, the system uses the Gaussian mixture algorithm to model the model set M. The Gaussian mixture model is a commonly used probability density estimation method that fits complex distribution forms through a linear combination of multiple Gaussian distributions. For each parameter of the ARIMA model (such as order p and order q), the system establishes a corresponding Gaussian mixture distribution model. For the ARIMA (1, 1, 1) model of conference room R1 in time period T1, its parameters p=1 and q=1 correspond to a two-component Gaussian mixture distribution respectively. The system uses the EM algorithm to iteratively optimize the parameters (mean, variance, weight) of the Gaussian mixture model until convergence. Finally, the system obtains the parameter distribution of each ARIMA model for subsequent model evaluation. For each ARIMA model in the model set M, the system uses the Gaussian mixture algorithm to establish a parameter distribution model. Taking ARIMA(1,1,1) as an example, its parameters p=1 and q=1 correspond to a Gaussian mixture distribution with two components respectively: Gaussian mixture distribution of p: first component: mean=0.5, variance=0.1, weight=0.6, second component: mean=0.8, variance=0.2, weight=0.4; Gaussian mixture distribution of q: first component: mean=0.3, variance=0.1, weight=0.7, second component: mean=0.6, variance=0.15, weight=0.3. The system performs EM algorithm to optimize the parameter distribution of each ARIMA model and obtains the converged Gaussian mixture distribution model.

[0034] After obtaining the parameter distribution of the ARIMA model, the system uses the AIC criterion with MAPE added to evaluate the prediction accuracy of ARIMA models of different orders. AIC (Akaike Information Criterion) is a commonly used model selection criterion that selects the optimal model by balancing model complexity and goodness of fit. The system first calculates the model complexity corresponding to each order combination (such as p=1, q=1) based on the Gaussian mixture distribution. Model complexity is measured by the information entropy of the parameters. The larger the information entropy, the more information the model contains and the higher the complexity. Then, the system uses MAPE (Mean Absolute Percentage Error) to evaluate the prediction accuracy of each order combination. MAPE measures the average percentage error between the predicted value and the true value and is a commonly used prediction accuracy indicator. Finally, the system weightedly sums the model complexity and MAPE to obtain the weighted AIC value of each order combination. The smaller the AIC value, the better the balance between complexity and accuracy. For each ARIMA model in the model set M, the system uses the Gaussian mixture algorithm to establish a parameter distribution model. Taking ARIMA(1,1,1) as an example, its parameters p=1 and q=1 correspond to a Gaussian mixture distribution with two components respectively: Gaussian mixture distribution of p: first component: mean=0.5, variance=0.1, weight=0.6, second component: mean=0.8, variance=0.2, weight=0.4; Gaussian mixture distribution of q: first component: mean=0.3, variance=0.1, weight=0.7, second component: mean=0.6, variance=0.15, weight=0.3. The system performs EM algorithm to optimize the parameter distribution of each ARIMA model and obtains the converged Gaussian mixture distribution model.

[0035] Figure 3This is an exemplary flow chart for obtaining an association matrix according to some embodiments of this specification. According to the Gaussian mixture distribution model, the system calculates the model complexity and prediction accuracy of each order combination. Take ARIMA (1, 1, 1) as an example: Model complexity: Information entropy of p = -0.6log (0.6) -0.4log (0.4) = 0.673, Information entropy of q = -0.7log (0.7) -0.3log (0.3) = 0.611, Model complexity = 0.673 + 0.611 = 1.284. Prediction accuracy: Use ARIMA (1, 1, 1) to predict the data set D1, obtain the predicted value sequence P1, and calculate MAPE = mean (abs ((P1-D1) / D1)) = 0.0852. The system repeats the above calculations for each order combination to obtain their model complexity and MAPE. Then, the model complexity and MAPE are weighted and summed (MAPE weight is 0.4), and the weighted AIC values ​​are obtained: ARIMA (1, 0, 0): AIC = 1.052, ARIMA (1, 0, 1): AIC = 1.185, ..., ARIMA (2, 1, 1): AIC = 1.326. According to the weighted AIC value, the system selects the model ARIMA (1, 1, 0) with the smallest AIC value from the candidate model set M as the optimal demand forecasting model for conference room R1 in time period T1. Output data: Demand forecasting model for conference room R1 in time period T1: ARIMA (1, 1, 0). Model parameters: Gaussian mixture distribution with p = 1: mean = 0.6, variance = 0.15, weight = 1.0, q = 0. Input data: conference room information database D_room, including static attributes such as conference room ID, name, capacity, location, etc. Room_ID|Name|Capacity|Location, R001|Large Conference Room A|20|3rd Floor, Building B, R002|Medium Conference Room B|10|5th Floor, Building A, R003|Small Conference Room C|5|2nd Floor, Building C. The status sensor collects the conference room usage status data D_status in real time, including the conference room ID, collection time, and usage status (0 means idle, 1 means occupied). Room_ID|Time|Status, R001|2023-05-1509:00:00|0, R002|2023-05-1509:00:05|1, R003|2023-05-1509:00:10|0. Processing process: The status sensor collects the usage status of each conference room at a frequency of 5 seconds and generates a real-time status data stream D_status. The system associates D_status with the conference room information database D_room to form a dynamically updated conference room status database D_room_status. The association condition is "D_room.Room_ID=D_status.Room_ID".

[0036] Room_ID|Name|Capacity|Location|Time|Status, R001|Conference Room A|20|Block B, 3rd Floor

[0037] |2023-05-1509:00:00|0, R002|Medium Meeting Room B|10|5th Floor, Building A|2023-05-1509:00:05|1, R003|Small Meeting Room C|5|2nd Floor, Building C|2023-05-1509:00:10|0. Output data: Dynamically updated meeting room status database D_room_status, which reflects the usage status of each meeting room in real time. When new status data arrives, the system automatically updates the latest status and time of the corresponding meeting room in D_room_status.

[0038] Figure 4 This is an exemplary flowchart for obtaining the probability of conference room vacancy as shown in some embodiments of this specification. A reservation request is submitted. The reservation request data D_request submitted by the user through the client includes information such as user ID, reservation time period, conference room ID, number of participants, etc.

[0039] User_ID|Start_Time|End_Time|Room_ID|Attendees, U001|2023-05-1614:00:00|2023-05-1616:00:00|R001|15, U002|2023-05-1615:30:00|2023-05-1617:00:00|R002|8. The user fills in the appointment form through the client and submits the appointment request. The key fields in the form include the start time, end time, conference room, number of participants, etc. After receiving the appointment request data D_request, the system extracts the key information and passes it to the appointment conflict detection module.

[0040] User_ID|Start_Time|End_Time|Room_ID|Attendees, U001|2023-05-1614:00:00|2023-05-1616:00:00|R001|15. Output data: key information data D_request_info of appointment request, which is passed to the appointment conflict detection module for subsequent processing.

[0041] User_ID|Start_Time|End_Time|Room_ID|Attendees, U001|2023-05-1614:00:00|2023-05-1616:00:00|R001|15.

[0042] Reservation conflict detection. The demand forecast model M_predict outputs the demand forecast D_predict for conference room R1 in time periods T1, T2, and T3. Time|Room_ID|Demand, T1|R1|8, T2|R1|12, T3|R1|6. The real-time status data D_status of conference room R1 in time periods T1, T2, and T3 in the conference room status database D_room_status. Time|Room_ID|Status, T1|R1|0.8; T2|R1|0.3; T3|R1|0.9. The reservation request data D_request submitted by user U1 reserves conference room R1 for time period T2.

[0043] User_ID|Room_ID|Start_Time|End_Time, U1|R1|T2|T2. Calculate the mutual information between the demand forecast and the real-time status data, sample the demand forecast D_predict and the real-time status data D_status according to the time period (T1, T2, T3), and obtain data sequences of equal length. D_predict: 8, 12, 6, D_status: 0.8, 0.3, 0.9. Perform dynamic binning on the sampled data, divide the demand forecast into 3 intervals: L1 (0-7), L2 (8-14), L3 (15-20), and divide the real-time status data into 3 intervals: S1 (0-0.4), S2 (0.4-0.7), S3 (0.7-1). Calculate the frequency distribution of the demand forecast X and the real-time status data Y after binning. X: L1(1), L2(2), L3(0), Y: S1(1), S2(0), S3(2), use Epanechnikov kernel function to estimate marginal probability density functions f(x) and g(y). f(x): L1(0.2), L2(0.8), L3(0), g(y): S1(0.3), S2(0), S3(0.7). Calculate the two-dimensional frequency distribution, and use the two-dimensional Gaussian kernel function to estimate the joint probability density function h(x, y). h(x, y): S1S2S3, L10.100.1, L20.200.6, L3000. Use the mutual information formula to calculate the mutual information value I(X; Y). I(X;Y)=0.1log(0.1 / (0.20.3))+0.6log(0.6 / (0.80.7))=0.485, and the association matrix A is constructed according to the mutual information value. T1, T2, T3, R1, 0.485, 0.485, 0.485.

[0044] Calculate the status matrix of the conference room and map the real-time status data D_status to the interval of 0-1. D_status_norm: 0.8, 0.3, 0.9, 2.2 Multiply the association matrix A by D_status_norm to get the idle probability matrix P. P: T1T2T3, R10.3880.1460.437. Determine whether there is a conflict in the reservation request. Extract the reservation conference room R1 and the reservation time period T2 from the reservation request D_request. Query the value of the corresponding element P[R1, T2] in the matrix P and get the idle probability of 0.146. Set the reservation threshold to 0.3 and compare the idle probability with the threshold. 0.146<0.3, it is determined that the reservation request D_request conflicts with the existing reservation.

[0045] Reservation conflict detection result: Request_ID|Conflict, U1_R1_T2|Yes, Through the above steps, the system uses the output of the demand prediction model and the real-time status data of the conference room to construct the association matrix and the status matrix. When the user submits a reservation request, the system calculates the corresponding conference room idle probability and compares it with the preset threshold to determine whether there is a reservation conflict. Generate alternatives, input data: Reservation conflict detection result, Request_ID|Conflict, U1_R1_T2|Yes, the demand forecast D_predict of each conference room in time period T2 output by the demand prediction model M_predict. Time|Room_ID|Demand, T2|R1|12, T2|R2|5, T2|R3|8. The real-time status data D_status of each conference room in time period T2 in the conference room status database D_room_status. Time|Room_ID|Status, T2|R1|0.3, T2|R2|0.9, T2|R3|0.7. Extract the conflicting reservation request U1_R1_T2 from the reservation conflict detection result. Filter out the data with time period T2 in the demand forecast D_predict and real-time status data D_status. Sort the filtered data from high to low according to the idle probability. Sorting result: Time|Room_ID|Status, T2|R2|0.9, T2|R3|0.7, T2|R1|0.3. Select the conference room with the highest idle probability and that meets user needs (such as capacity, equipment, etc.) as an alternative, where R2 meets the conditions. Generate alternative data D_alternative.

[0046] Request_ID|Room_ID|Start_Time|End_Time, U1_R1_T2|R2|T2|T2, output data: alternative data D_alternative, provided for user selection. Request_ID|Room_ID|Start_Time|End_Time, U1_R1_T2|R2|T2|T2. Original reservation request data D_request.

[0047] Request_ID|User_ID|Room_ID|Start_Time|End_Time, U1_R1_T2|U1|R1|T2|T2, appointment conflict detection result, Request_ID|Conflict, U1_R1_T2|Yes, alternative plan data D_alternative (if there is a conflict). Request_ID|Room_ID|Start_Time|End_Time, U1_R1_T2|R2|T2|T2. Processing process: The system checks the conflict detection result of appointment request U1_R1_T2 and finds that there is a conflict. The system extracts the corresponding alternative plan from the alternative plan data D_alternative. The system pushes the appointment result notification to user U1 through the client, including the original appointment information, conflict prompts and alternative plans. User U1 chooses to accept alternative plan R2 on the client, and the system records the user's selection result.

[0048] Final reservation result data D_result. Request_ID|User_ID|Room_ID|Start_Time|End_Time

[0049] U1_R1_T2|U1|R2|T2|T2. Update the database and the final reservation result data D_result.

[0050] Request_ID|User_ID|Room_ID|Start_Time|End_Time; U1_R1_T2|U1|R2|T2|T2. The system extracts key fields from D_result, including conference room ID, reservation start time, end time, etc. The system updates the extracted data fields to the corresponding records in the conference room information database D_room. The system adds D_result to the historical reservation dataset D_history as a new sample. Output data: Updated conference room information database D_room. Room_ID|Name|Capacity|Location|Date|Start_Time|End_Time|Status, R2|Medium Conference Room B|10|Building A, 5th Floor|2023-05-16|T2|T2|Reserved. Updated historical reservation dataset D_history.

[0051] Request_ID|User_ID|Room_ID|Date|Start_Time|End_Time,

[0052] U1_R1_T2|U1|R2|2023-05-16|T2|T2.

Claims

1. An intelligent online appointment management method, comprising: Step 1: Establish a conference room information database to store static information of the conference room, which includes the name and location of the conference room; Step 2: Collect historical reservation data sets, which include reservation time, conference room information, and usage time; Step 3: Based on the collected historical reservation data set, a time series analysis algorithm is used to establish and train a conference room usage demand prediction model, which is used to predict the usage demand of the target conference room within a preset time period; Step 4: Set up a status sensor in the conference room to collect real-time status data. The status data includes the usage status of the conference room. The usage status refers to whether the conference room is idle. Step 5: Associating the collected real-time status data of each conference room with the static information in the database to generate dynamic data of each conference room; Step 6: receiving a reservation request submitted by the user, the reservation request including user information, the reservation meeting room and the reservation time period; Step 7: Conflict detection is performed on the reserved conference rooms and reserved time periods through the prediction model and dynamic data to determine whether the user's reservation request conflicts; Step 8: When there is a conflict, generate a conflict-free alternative reservation plan through the prediction model and dynamic data; Conflict detection, using the state matrix composed of the real-time state data of the conference room for calculation; Obtain demand forecasts for different time periods output by the prediction model, and obtain real-time status data of the conference room from the status sensor; The mutual information algorithm is used to calculate the correlation between the demand forecast and the real-time status data. Under the Bayesian framework, the demand forecast is used as the observation and the real-time status data is used as the random variable to establish the correlation between the demand forecast and the real-time status data. The demand forecast is used as the prior knowledge constraint status data. According to the calculated correlation, a correlation matrix is ​​constructed, in which the rows represent the demand forecast, the columns represent the real-time status data, and the element value is the correlation; Map the collected real-time status data to the interval of 0 to 1, multiply the mapped real-time status data by the correlation degree to obtain the idle probability of each conference room in each time period; According to the idle probability, a state matrix is ​​constructed; wherein the rows in the state matrix represent time periods, the columns represent conference rooms, and the elements represent the idle probability of each conference room; Receive reservation requests submitted by users, and obtain the reserved meeting room and reservation time period; In the state matrix, the idle probability of the corresponding matrix element is calculated according to the reserved conference room and the reserved time period; The calculated idle probability is compared with a preset threshold. When the idle probability is less than the threshold, it is determined that there is a conflict, otherwise there is no conflict; The mutual information algorithm is used to calculate the correlation between demand forecast and real-time status data, including: The demand forecast and real-time status data are sampled at preset time intervals to obtain data sequences of equal length; The sampled data sequence is binned to discretize the continuous variables. After discretization, the value range of each variable is divided into several intervals. Calculate the joint probability distribution and marginal probability distribution of the demand forecast and real-time status data after binning; the joint probability distribution represents the probability that two variables take specific values ​​at the same time, and the marginal probability distribution represents the probability that each variable takes a specific value; According to the calculated joint probability distribution and marginal probability distribution, the mutual information value of demand forecast X and real-time status data Y is calculated using the following formula: : ； in, represents the joint probability distribution of X and Y, and Represents the marginal probability distribution of X and Y, and the mutual information value represents the correlation between X and Y, The larger the value, the higher the correlation.

2. The intelligent online appointment management method according to claim 1, characterized in that: Use time series analysis algorithms to build and train a conference room usage demand forecasting model, including: Preprocess the collected historical reservation data set by cleaning, dividing time intervals, and extracting time features; For each conference room and each time period, the ARIMA algorithm is used to build an ARIMA model; The Gaussian mixture algorithm is used to evaluate the parameter distribution of ARIMA models of different orders; The AIC criterion with MAPE added is used to evaluate the forecasting accuracy of ARIMA models of different orders based on parameter distribution. The ARIMA model with the highest prediction accuracy is selected as the prediction model for the corresponding time period of the corresponding conference room. The prediction model is used to predict the demand forecast amount in different time periods.

3. The intelligent online appointment management method according to claim 2, characterized in that: The Gaussian mixture algorithm uses the expectation maximization method EM to estimate model parameters; in: For each parameter of the ARIMA model, a Gaussian mixture distribution model is established, and the parameters include order p and order q; Randomly initialize the mean, variance, and weight parameters of each Gaussian mixture distribution model; Calculate the posterior probability of each parameter in each Gaussian mixture distribution model; Maximize the likelihood function of the parameters according to the posterior probability; Repeat the above steps until the parameters converge; Output the converged Gaussian mixture distribution model as the parameter distribution of the ARIMA model.

4. The intelligent online appointment management method according to claim 3, characterized in that: Using the AIC criterion with MAPE added, the forecast accuracy of ARIMA models of different orders is evaluated based on parameter distribution, including: Based on the ARIMA model parameter distribution obtained by the Gaussian mixture algorithm, the model complexity corresponding to each order combination is calculated; MAPE is used to evaluate the prediction accuracy corresponding to each order combination, and the weight of MAPE accounts for 30% to 50% of the total weight of AIC; The weighted sum of model complexity and MAPE is taken to obtain the weighted AIC value of each order combination; The order combination with the smallest weighted AIC value is selected as the order of the optimal ARIMA model, and the ARIMA model with the highest prediction accuracy under the corresponding order is selected as the final prediction model.

5. The intelligent online appointment management method according to claim 4, characterized in that: The ARIMA model parameter distribution obtained based on the Gaussian mixture algorithm includes: A Gaussian mixture distribution model is established for the parameters of each ARIMA model, and the parameters include order p and order q; Randomly initialize the mean, variance, and weight parameters of each Gaussian mixture distribution model; Calculate the posterior probability of each parameter in each Gaussian mixture distribution model; Maximize the likelihood function of the parameters according to the posterior probability; Repeat the above steps until the parameters reach the preset convergence conditions; The converged Gaussian mixture distribution model is used as the parameter distribution of the ARIMA model; According to the ARIMA model parameter distribution obtained by the Gaussian mixture algorithm, the joint distribution of the p-order and q-order is constructed; The Monte Carlo algorithm is used to sample from the joint distribution of the p-order and q-order combinations to generate multiple p-order and q-order combinations as candidate ARIMA model order combinations.

6. The intelligent online appointment management method according to claim 5, characterized in that: Calculate the model complexity corresponding to each order combination, including: Calculate the probability distribution of each parameter in the Gaussian mixture distribution model; Treat the probability distribution of each parameter as a discrete random variable and calculate the corresponding information entropy; Sum the information entropy of all parameters to obtain the parameter entropy value of the corresponding order combination; The parameter entropy value of each order combination is used as the model complexity. The larger the parameter entropy value, the more information the model contains and the higher the model complexity.

7. The intelligent online appointment management method according to claim 1, characterized in that: Calculate the joint probability distribution and marginal probability distribution of the demand forecast and real-time status data after binning, including: According to the demand forecast X and real-time status data Y after binning, the number of samples in each interval is calculated to obtain the frequency distribution; According to the demand forecast X and real-time status data Y after binning, the number of samples in each interval is calculated to obtain the frequency distribution of the demand forecast X and real-time status data Y; According to the frequency distribution of the demand forecast quantity X, the Epanechnikov kernel function is used to estimate the probability density function f(x) of the demand forecast quantity X, and the probability density function f(x) is used as the marginal probability distribution of the demand forecast quantity X; According to the frequency distribution of the real-time status data Y, the Epanechnikov kernel function is used to estimate the probability density function g(y) of the real-time status data Y, and the probability density function g(y) is used as the marginal probability distribution of the real-time status data Y; According to the demand forecast X and real-time status data Y after binning, the number of samples in each interval combination is calculated to obtain a two-dimensional frequency distribution; According to the two-dimensional frequency distribution, the joint probability density function h(x, y) of the demand forecast X and the real-time status data Y is estimated by the two-dimensional Gaussian kernel function, and the joint probability density function h(x, y) is used as the joint probability distribution of the demand forecast X and the real-time status data Y.

8. The intelligent online appointment management method according to claim 7, characterized in that: The dynamic binning method is used to bin the sampled data sequence.

Images