An intelligent scheduling method

Abstract

The application provides a kind of intelligent scheduling method, comprising: extracting the nursing staff characteristics of nursing staff and the nursing staff characteristics of nursing staff;Nursing staff characteristics and / or nursing staff characteristics are pretreated;According to the nursing staff characteristics and nursing staff characteristics, construct BP neural network, the number of hidden layers is one layer, the output layer is single neuron, and Sigmoid function is used as activation function in hidden layer and output layer;The historical data of nursing staff characteristics and nursing staff characteristics are put into training to obtain the recommended score of nursing staff characteristics and nursing staff characteristics two two pairing, wherein the recommended score represents the degree of corresponding nursing staff to corresponding nursing staff;According to the recommended score, the score of nursing staff is ranked, and recommended to staff;And staff selects the highest score nursing staff to schedule.To solve the unreasonable allocation of existing nursing staff work, make the influence caused by nursing staff shortage more serious.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent scheduling method. Background Technology

[0002] The existing home-based elderly care market is large in scale, geographically widespread, and services are scattered, making it impossible to achieve the unified care and management of elderly care institutions. On the other hand, according to relevant statistics, China's elderly care industry needs 13 million caregivers, but the actual number of practitioners is less than 300,000, which further restricts the development of the elderly care industry and leaves the needs of many elderly people unmet.

[0003] Furthermore, China faces a significant shortage of caregivers, and the training process is lengthy. Caregivers must undergo extensive training and pass the "Elderly Caregiver Professional Qualification Examination" to obtain a caregiver certificate before they can be employed. Moreover, the limited manpower leads to unreasonable work assignments, resulting in a large number of caregivers being unfairly distributed, which further exacerbates the shortage. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent scheduling method to address the problem of unreasonable work allocation for caregivers, which exacerbates the impact of caregiver shortages.

[0005] To address the aforementioned technical problems, this invention provides an intelligent scheduling method, comprising:

[0006] Extract caregiver characteristics from caregivers and caregiver characteristics from caregivers;

[0007] Preprocessing of caregiver characteristics and / or caregiver characteristics;

[0008] Based on the characteristics of caregivers and caregivers, a backpropagation neural network is constructed with one hidden layer and a single neuron in the output layer. The sigmoid function is used as the activation function for both the hidden and output layers.

[0009] Historical data on caregiver and careee characteristics are fed into the training to derive recommendation scores for pairwise pairings of careee and caregiver characteristics, wherein the recommendation scores represent the degree to which a corresponding caregiver is assigned to a corresponding careee.

[0010] Caregivers are ranked and recommended to staff based on the recommended scores; and

[0011] Staff selected the caregiver with the highest score for scheduling.

[0012] Furthermore, historical data is collected and fed into a neural network for learning to obtain a final model for subsequent scheduling. After the staff schedule is completed, new data is fed into the original model for training to continuously optimize the model.

[0013] When staff need to assign caregivers to caregivers, they can query for caregivers who meet the criteria. The data of the caregivers who meet the criteria and the caregivers who are being cared for are combined into a vector and used as the input to the model. The output is sorted and recommended to the staff, who then select the first caregiver to schedule the caregiver.

[0014] The input features of the care recipient are preprocessed, and the weights calculated by the initialization module are obtained by the expert system based on experience.

[0015] After collecting features of caregivers and caregivers, a backpropagation (BP) neural network was constructed with one hidden layer and a single output neuron. Both the hidden and output layers used the sigmoid function as the activation function.

[0016] Furthermore, the preprocessing includes:

[0017] Let the gender of the caregiver and the gender of the person being cared for be G∈{M,F}. Calculate the percentage of service time spent by male and female employees:

[0018]

[0019] N = 1 - M

[0020] Where M represents the percentage of service time for male employees, N represents the percentage of service time for female employees, n represents the number of female employees, m represents the number of male employees, and S... i This represents the service duration of the i-th employee;

[0021] By calculating the ratio of service time between male and female employees, preprocessing is performed to assign gender-appropriate caregivers to those being cared for.

[0022] Enter the characteristics of the caregiver type according to the requirements of the caregiver:

[0023] P(r)

[0024] Where r represents the requirements of the person being cared for, and pi represents the type of caregiver;

[0025] Caregiver job types include full-time, part-time, internship, and retired. Full-time caregiver positions are given priority in allocation. The formula for the average number of caregiver positions per type is:

[0026]

[0027] Where μ iRepresents the average number of orders for the i-th type of caregiver, count(S) i ) represents the total number of employees of the i-th type; order(S) ij Let represent the order volume of the j-th caregiver of the i-th type; after calculating the average order volume of caregivers for each job type, substitute it into the softmax function to standardize it while amplifying the weights of advantageous job types:

[0028]

[0029] Among them, S i The feature input representing the i-th type of caregiver;

[0030] The preprocessing also includes:

[0031] Calculate the time difference between the previous order and the next order based on the end time of the previous order and the start time of the next order:

[0032]

[0033] Where T e Indicates the end time of the previous order, T. s Indicates the start time of the next order, T i This represents the threshold time difference. If the time difference between an order and a new order exceeds the threshold time difference, it is considered that a new order can be inserted between these two orders. The threshold time difference is defined according to the business situation.

[0034] W t P represents the estimated travel time for employees. LN Feature inputs representing the time elapsed since the end of the previous order or the start time of the next order;

[0035] Calculate the caregiver's estimated travel time based on the estimated travel time from the previous order to the current order and the estimated travel time from the current order to the next order:

[0036]

[0037] Where (x) i y i ) and (x j y j ) represents the latitude and longitude coordinates of the two locations, v represents the estimated speed of the vehicle used by the caregiver, and ρ represents the time redundancy, which is preset based on the actual situation of the caregiver; the standardized mean and variance normalization of the travel time is applied using the modified method.

[0038]

[0039] Where X is the travel time required, μ is the average travel time of the caregiver, and σ is the variance of the travel time of the caregiver.

[0040] Furthermore, it also includes:

[0041] Extract the existing and future work characteristics of caregivers;

[0042] Based on existing and future work characteristics, the most suitable scheduling scheme function is derived through data analysis and machine learning.

[0043] Schedule future work for idle personnel based on the scheduling scheme function; and

[0044] Qualified caregivers will be selected based on the future work schedule.

[0045] Furthermore, the characteristics of the caregiver include one or more of the following: caregiver gender, caregiver age, caregiver type, caregiver work experience, work experience with this company, caregiver employee star rating, caregiver employee score, number of complaints against the caregiver, total number of caregiver orders, caregiver highest education level, caregiver job type, caregiver marital status, caregiver height, caregiver weight, number of orders per day, end time of the previous order, start time of the next order, distance from the previous order, and distance from the next order.

[0046] Furthermore, the characteristics of the caregiver include one or more of the following: the caregiver's gender, the caregiver's age, the length of the caregiver's contract, the caregiver's health level, the caregiver's average rating, the number of complaints the caregiver has received, the number of times the caregiver has been served by the current caregiver, and the number of times the caregiver has been served by the current caregiver in the past 30 days.

[0047] Furthermore, it is characterized by:

[0048] After constructing the BP neural network, weights are assigned to the first layer of the neural network by initializing the weight module;

[0049] The initial weight module uses an expert system to calculate the weights of the first layer of the neural network. The expert system employs a large-scale multi-expert collaborative system, multi-knowledge representation, comprehensive knowledge base, self-organizing problem-solving mechanism, multi-disciplinary collaborative problem-solving and parallel reasoning, expert system tools and environment, and artificial neural network knowledge acquisition and learning mechanism to calculate the weights of the first layer of the neural network, which has multiple knowledge bases and multiple subjects.

[0050] Furthermore, the weights of the first layer of the neural network include:

[0051] The weights for the caregiver's highest education level, marital status, height, weight, gender, and gender of the person being cared for are 0.1.

[0052] The weight of the health level of the person being cared for is 0.2;

[0053] The weights for caregiver type, caregiver job type, caregiver work experience, and length of contract with the caregiver are 0.3;

[0054] The weight of work experience in this company is 0.4;

[0055] The weights for caregiver employee star rating, caregiver employee score, number of complaints against caregivers, total number of caregiver cases, caregiver age, age of caregivers, average score of caregivers, and number of complaints against caregivers are all 0.5.

[0056] The weight of each caregiver's daily order volume is 0.6;

[0057] The weighting of the caregiver's previous job end time, the caregiver's next job start time, and the number of times the caregiver has served the patient in the past 30 days is 0.8; and

[0058] The weights for the distance of the caregiver's previous job, the distance of the caregiver's next job, and the number of times the caregiver has served the patient are all 0.7.

[0059] Furthermore, the model training process includes:

[0060] The feature input vector is:

[0061] x = (x0, x1, x2, ..., x...) p x p+1 ,·,x n ) T

[0062] Where x1 to x p As a characteristic of caregivers, x p+1 To x n The features of the person being cared for are represented by x0, which is the bias node; the weights of the first layer of the neural network are:

[0063]

[0064] Where w ij Let represent the weights from the j-th node in the input layer to the i-th node in the hidden layer. Then the output of the hidden layer is...

[0065] Z = Wx

[0066] Substituting it into the Sigmoid function, we get:

[0067]

[0068] After obtaining the hidden layer input, forward propagation continues. The weights of the second layer of the neural network are randomly initialized and represented as follows:

[0069] V = (v0, v1, v2, ..., v n)

[0070] Where v i This represents the i-th weight from the hidden layer node to the output layer node; then the final output is:

[0071]

[0072] Once the predicted value ^y is obtained, the loss function is defined using MSE. The loss function is as follows:

[0073]

[0074] Where m represents the total number of samples; by performing gradient descent on the neural network through backpropagation, the weights of the first layer of the neural network are adjusted as follows:

[0075]

[0076] Where η is the learning rate, and the partial derivative is... for:

[0077]

[0078] Adjusting the weights of the first layer of the neural network, according to the chain rule, yields:

[0079]

[0080] in, for:

[0081]

[0082] After the update, the first round of the loop is completed; repeat the above loop until the changes between two iterations approach 0, then stop the iteration and the model training is completed; for subsequent scheduling, after the staff schedule is completed, continue to feed new data into the original model for training and continuously optimize the model;

[0083] Once the model is trained, when scheduling shifts, the characteristics of caregivers and patients are preprocessed and then imported into the trained model to generate caregiver ratings.

[0084] Score i =f(X) i X j )

[0085] Where X i X represents the feature vector of the caregiver. j Let f represent the feature vector of the person being cared for, and f be the trained model function, with score as the denominator. i The score is assigned to caregiver i; finally, the caregiver with the highest score is selected from the available caregivers.

[0086] The inventors of this invention discovered through research that the large distance between two service locations causes caregivers to spend most of their time traveling, and the large distance between two service times causes caregivers to spend a lot of time waiting at the elderly person's door for their service time to begin. This results in an unreasonable allocation of work for existing caregivers, exacerbating the impact of the shortage of caregivers.

[0087] Based on the above insights, this invention provides an intelligent recommendation and scheduling system for caregivers. By extracting caregiver characteristics, elderly person characteristics (caregiver characteristics), existing job characteristics, and future job characteristics, a mathematical model is established. Through data analysis and machine learning, the most suitable scheduling scheme function is learned. Finally, the system schedules future jobs for idle personnel according to the model. In addition, a neural network is constructed based on caregiver and elderly person characteristics, and historical data is used for training. The final result is a score between 0 and 1, which represents the degree to which a caregiver should be assigned to an elderly person. Caregivers who meet the criteria are ranked by score and recommended to the scheduling staff. The scheduling staff prioritizes the highest-scoring caregiver, which can achieve the optimal solution for caregiver job allocation and greatly alleviate the drawbacks caused by short-term caregiver assignment. Attached Figure Description

[0088] Figure 1 This is a neural network diagram of a nursing staff intelligent recommendation scheduling system according to an embodiment of the present invention. Detailed Implementation

[0089] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0090] It should be noted that the components in the various figures may be shown exaggeratedly for illustrative purposes and are not necessarily to scale. In each figure, the same reference numerals are used for components that are identical or have the same function.

[0091] In this invention, unless otherwise specified, "arranged on," "arranged above," and "arranged on" do not exclude the possibility of an intermediate element between them. Furthermore, "arranged on or above" merely indicates the relative positional relationship between two components, and in certain cases, such as when the product orientation is reversed, it can also be converted to "arranged below or under," and vice versa.

[0092] In this invention, the various embodiments are merely intended to illustrate the solutions of the invention and should not be construed as limiting.

[0093] In this invention, unless otherwise specified, the quantifiers “a” and “one” do not exclude scenarios involving multiple elements.

[0094] It should also be noted that, in the embodiments of the present invention, only a portion of the components or parts may be shown for clarity and simplicity. However, those skilled in the art will understand that, under the teachings of the present invention, necessary components or parts can be added as needed for specific scenarios. Furthermore, unless otherwise stated, features in different embodiments of the present invention can be combined with each other. For example, a feature in the second embodiment can replace a corresponding or functionally identical or similar feature in the first embodiment, and the resulting embodiment will also fall within the scope of disclosure or description of this application.

[0095] It should also be noted that, within the scope of this invention, the terms "same," "equal," and "equal to" do not imply that the two values ​​are absolutely equal, but rather allow for a certain reasonable margin of error. In other words, the terms also encompass "substantially the same," "substantially equal," and "substantially equal to." Similarly, in this invention, the directional terms "perpendicular to," "parallel to," etc., also encompass the meanings of "substantially perpendicular to" and "substantially parallel to."

[0096] Furthermore, the numbering of the steps in the methods of the present invention does not limit the execution order of the method steps. Unless otherwise specified, the method steps may be executed in different orders.

[0097] The intelligent recommendation and scheduling system for caregivers proposed in this invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The advantages and features of this invention will become clearer from the following description. It should be noted that the drawings are all in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of this invention.

[0098] The purpose of this invention is to provide an intelligent recommendation and scheduling system for caregivers, in order to solve the problem of unreasonable work allocation for caregivers, which exacerbates the impact of caregiver shortages.

[0099] Existing AI algorithms include automated order dispatch algorithms for the ride-hailing and food delivery industries. The ride-hailing industry currently has a very mature order dispatch algorithm. The primary principle of this algorithm is "proximity principle," assigning orders to the driver closest to the current user. In addition, it considers driver characteristics such as ratings, order volume, and vehicle type, and user characteristics such as complaint volume and destination, making a comprehensive comparison to ultimately select a suitable driver for dispatch. Food delivery order dispatch algorithms automatically dispatch orders by building machine learning models. These models automatically calculate "food preparation time estimates," "delivery time estimates," "future order estimates," and "route time estimates." Simultaneously, based on domain knowledge, they perform "route planning" and "automatic order dispatch."

[0100] However, the inventors discovered through research that existing AI algorithms are not suitable for dispatching caregiver home visits. For example, compared to automatic dispatching for ride-hailing services, ride-hailing dispatch is real-time, meaning a driver is assigned based on the user's needs only when the user places an order. In contrast, caregiver home visit services are dispatched in advance. Clients and companies sign long-term contracts, and caregivers are scheduled to visit at fixed times each week. The characteristics that ride-hailing services need to consider are not applicable to the home-based elderly care industry; they cannot be directly applied and cannot be converted. Furthermore, ride-hailing dispatch does not require consideration of whether the driver's skills match the user's needs. However, in the home-based elderly care industry, it is necessary to consider whether the caregiver possesses the skills required by the client, such as massage or acupressure. Finally, ride-hailing departs from the current location for the next order, and the next destination is uncertain. But caregiver services depart from the home of the previous client, and the destination for the next order is fixed.

[0101] Furthermore, the inventors compared home delivery order dispatching by caregivers with automated food delivery order dispatching and found that food delivery order dispatching requires consideration of the number of orders currently held by the delivery person, allowing a single delivery person to handle multiple orders simultaneously. However, caregivers can only serve one order at a time; only after completing that order can they start the next. Food delivery has high requirements for route planning, while caregivers do not require extensive analysis of their service routes. Food delivery services end upon arrival at the customer's home. However, caregivers incur a service time limit upon arrival at the customer's home, and only after this time can they proceed to the next order. In summary, existing order dispatching algorithms cannot effectively solve the scheduling problem for caregivers, necessitating a customized scheduling method tailored to the specific characteristics of their work. This invention provides an intelligent recommendation and scheduling system for caregivers, belonging to the field of home-based elderly care technology, primarily aimed at rationally arranging caregiver work.

[0102] This invention discloses a method for intelligently scheduling caregivers' future work based on their existing work assignments. This invention is applicable only to the home-based elderly care service industry. Schedulers can use the rules provided by this invention to optimally schedule their staff, rationally allocate work time, reduce time fragmentation, and maximize company profits. This solution extracts caregiver characteristics, existing work characteristics, and future work characteristics to establish a mathematical model. Through data analysis and machine learning, it learns the most suitable scheduling function. Finally, it schedules future work for available staff based on the model. This invention will be described in detail, including but not limited to: background, technical solution, beneficial effects, applicable scenarios, instructions for use, and model explanation. By reading this article, readers can understand how to use this invention and apply it in practical scenarios.

[0103] To achieve the above objectives, the present invention provides an intelligent recommendation and scheduling system for caregivers, comprising: an information extraction module configured to extract caregiver characteristics and caregiver characteristics of each caregiver; a neural network construction module configured to construct a neural network based on the caregiver characteristics and caregiver characteristics, inputting historical data of the caregiver characteristics and caregiver characteristics into training, and obtaining a recommendation score for each pair of caregiver characteristics and each caregiver characteristic, wherein the recommendation score represents the degree to which the corresponding caregiver should be assigned to the corresponding caregiver; and a scheduling module configured to rank eligible caregivers by score and recommend them to the scheduling staff, with the scheduling staff prioritizing the caregiver with the highest score.

[0104] Embodiments of the present invention provide an intelligent recommendation and scheduling system for caregivers, comprising: an information extraction module configured to extract caregiver characteristics and caregiver characteristics of each caregiver; a neural network construction module configured to construct a neural network based on the caregiver characteristics and caregiver characteristics, inputting historical data of the caregiver characteristics and caregiver characteristics into training, and obtaining a recommendation score for each pair of caregiver characteristics and each caregiver characteristic, wherein the recommendation score represents the degree to which the corresponding caregiver should be assigned to the corresponding caregiver; and a scheduling module configured to rank eligible caregivers by score and recommend them to scheduling personnel, with scheduling personnel prioritizing the selection of the highest-scoring caregivers.

[0105] In one embodiment of the present invention, in the aforementioned intelligent recommendation and scheduling system for caregivers, the information extraction module is further configured to extract the existing work characteristics and future work characteristics of each caregiver; the neural network construction module includes a mathematical model building module, which is configured to derive the most suitable scheduling scheme function based on the existing work characteristics and future work characteristics through data analysis and machine learning; the scheduling module is further configured to perform the following actions: schedule future work for idle personnel according to the scheduling scheme function; and obtain caregivers who meet the conditions based on the scheduling of future work.

[0106] This invention extracts features from caregivers and elderly individuals, constructs a neural network, and trains it using historical data. The resulting score, between 0 and 1, represents the degree to which a caregiver should be assigned to an elderly individual. Eligible caregivers are ranked by score and recommended to scheduling staff, who, in principle, prioritize the caregiver with the highest score.

[0107] In one embodiment of the present invention, in the aforementioned intelligent caregiver recommendation and scheduling system, caregiver characteristics include caregiver gender, caregiver age, caregiver type, caregiver work experience, company work experience, caregiver employee star rating, caregiver employee score, number of complaints against the caregiver, total number of caregiver orders, caregiver highest education level, caregiver job type, caregiver marital status, caregiver height, caregiver weight, number of orders for the day, end time of the previous order, start time of the next order, distance from the previous order, and distance from the next order.

[0108] In one embodiment of the present invention, in the aforementioned intelligent recommendation and scheduling system for caregivers, the characteristics of the caregiver include the caregiver's gender, age, contract duration, health level, average score, number of complaints, number of times the caregiver has been served by the current caregiver, and number of times the caregiver has been served by the current caregiver in the past 30 days. The characteristics to be considered in the elderly care industry differ from those in traditional industries.

[0109] In one embodiment of the present invention, the intelligent recommendation and scheduling system for caregivers further includes: a preprocessing module configured to preprocess caregiver characteristics and / or careee characteristics before inputting them into the neural network construction module; the preprocessing includes:

[0110] Let the gender of the caregiver and the gender of the person being cared for be G∈{M,F}. Calculate the percentage of service time spent by male and female employees:

[0111]

[0112] N = 1 - M

[0113] Where M represents the percentage of service time for male employees, N represents the percentage of service time for female employees, n represents the number of female employees, m represents the number of male employees, and Si represents the service time of the i-th employee; Preprocessing is performed by calculating the percentage of service time for male and female employees to assign gender-appropriate caregivers to the recipients; Employee position is a relatively special feature, as different positions have different levels of difficulty, and the different needs of the elderly will also affect the choice of position. The input of this feature needs to be combined with the requirements of the elderly, and the caregiver type feature should be input according to the requirements of the recipients:

[0114]

[0115] Where r represents the requirements of the person being cared for, and p i This refers to the types of caregivers; the caregiver job types include full-time, part-time, intern, and retired, with full-time accounting for the largest proportion and being the type that should be prioritized for allocation. The priority allocation type for caregiver positions is full-time, and the formula for the average number of caregiver positions for each type is:

[0116]

[0117] Where μ i Represents the average number of orders for the i-th type of caregiver, count(S) i ) represents the total number of employees of the i-th type; order(S) ij Let represent the order volume of the j-th caregiver of the i-th type; after calculating the average order volume of caregivers for each job type, substitute it into the softmax function to standardize it while amplifying the weights of advantageous job types:

[0118]

[0119] Among them, S i The feature input represents the i-th type of caregiver.

[0120] In one embodiment of the present invention, in the aforementioned intelligent nursing assistant recommendation scheduling system, the preprocessing further includes: the time remaining until the end (start) of the previous (next) order. The order to be scheduled for the nursing assistant needs to consider the time of the current order and the time of the nursing assistant's previous and next orders. Simply directly applying a negative correlation is not reasonable and requires special processing. Based on the end time of the previous order and the start time of the next order, the time difference between the previous and next orders is calculated:

[0121]

[0122] Where T e Indicates the end time of the previous order, T. s Indicates the start time of the next order, T i This represents the threshold time difference. If the time difference between an order and a new order exceeds the threshold time difference, then a new order is considered to be inserted between these two orders. The threshold time difference is defined based on business requirements; T i This indicates that if the time interval exceeds this threshold, a new order can be inserted between these two orders. This value is a constant defined based on business requirements.

[0123] W t P represents the estimated travel time for employees. LN Feature inputs representing the time elapsed since the end of the previous order or the start time of the next order;

[0124] Calculate the caregiver's estimated travel time based on the estimated travel time from the previous order to the current order and the estimated travel time from the current order to the next order:

[0125]

[0126] Where (x) i y i ) and (xj y j The coordinates () represent the latitude and longitude of the two locations, v represents the estimated speed of the vehicle used by the caregiver, and ρ represents the time redundancy, which provides the caregiver with a certain amount of time tolerance. This value needs to be specified by the staff based on the actual situation of the caregiver; therefore, the time redundancy is preset according to the actual situation of the caregiver. The standardization of the travel time is achieved using the modified mean and variance normalization:

[0127]

[0128] Where X represents the travel time, μ represents the average travel time for caregivers, and σ represents the variance of travel time for caregivers. The final neural network input features are summarized in Table 1.

[0129] Table 1: Characteristics of Caregivers

[0130]

[0131] The input features of caregivers are processed according to the method shown in the table. The initial weights are pre-judged by business personnel based on experience (not rigorous, and can be modified arbitrarily according to subjective wishes). This reduces the number of training iterations and speeds up model training. The characteristics of the elderly are shown in Table 2.

[0132] Table 2: Nursing Staff Characteristics Table

[0133]

[0134] In one embodiment of the present invention, in the aforementioned intelligent recommendation and scheduling system for caregivers, the neural network construction module is further configured to: construct a BP neural network based on the preprocessed input features of caregivers and caregivers, wherein the BP neural network has one hidden layer and a single neuron as its output layer, and both the hidden and output layers of the BP neural network use the Sigmoid function as the activation function; after constructing the BP neural network, weights are assigned to the first layer of the neural network through an initialization weight module; the initialization weight module uses an expert system to calculate the weights of the first layer of the neural network, wherein the expert system employs a large-scale multi-expert collaborative system, multiple knowledge representations, a comprehensive knowledge base, a self-organizing problem-solving mechanism, multi-disciplinary collaborative problem-solving and parallel reasoning, expert system tools and environment, and an artificial neural network knowledge acquisition and learning mechanism, to obtain the weights of the first layer of the neural network with multiple knowledge bases and multiple subjects.

[0135] In one embodiment of the present invention, in the aforementioned intelligent recommendation and scheduling system for caregivers, the weights of the first layer of the neural network include: 0.1 for caregiver's highest education level, marital status, height, weight, gender, and gender of the caregiver; 0.2 for the caregiver's health level; 0.3 for caregiver type, job type, work experience, and contract duration of the caregiver; 0.4 for work experience within the company; 0.5 for caregiver employee star rating, employee score, number of complaints against the caregiver, total number of caregiver orders, age of the caregiver, age of the caregiver, average score of the caregiver, and number of complaints against the caregiver; 0.6 for the number of orders the caregiver receives that day; 0.8 for the caregiver's last order end time, next order start time, and number of times the caregiver has served the caregiver in the past 30 days; and 0.7 for the distance between the caregiver's last order, next order, and the number of times the caregiver has served the caregiver.

[0136] In one embodiment of the present invention, in the intelligent caregiver recommendation and scheduling system, historical data is collected and fed into a neural network for learning. A final model is then derived for scheduling. After staff scheduling, new data is continuously fed into the existing model for training, continuously optimizing the model. When a station manager needs to assign a caregiver to a patient, qualified caregivers are queried. A vector is formed by combining the qualified caregiver and patient data as input to the model. The resulting output is sorted and recommended to the station manager, who then selects the first caregiver for scheduling. The input features of the patient are preprocessed, where the weights calculated by the initialization module are pre-judged by an expert system based on experience, reducing the number of training iterations and accelerating model training. After collecting the features of caregivers and patients, a backpropagation (BP) neural network is constructed with one hidden layer and a single output neuron. Both the hidden and output layers use the sigmoid function as the activation function. The neural network is as follows: Figure 1 As shown.

[0137] In one embodiment of the present invention, the training process of the model in the intelligent recommendation and scheduling system for caregivers includes:

[0138] The feature input vector is:

[0139] x = (x0, x1, x2, ..., x...) p x p+1 ,·,x n ) T

[0140] Where x1 to x p As a characteristic of caregivers, x p+1 To x nThe features of the person being cared for are represented by x0, which is the bias node; the weights of the first layer of the neural network are:

[0141]

[0142] Where w ij Let represent the weights from the j-th node in the input layer to the i-th node in the hidden layer. Then the output of the hidden layer is...

[0143] Z = Wx

[0144] Substituting it into the Sigmoid function, we get:

[0145]

[0146] After obtaining the hidden layer input, forward propagation continues. The weights of the second layer of the neural network are randomly initialized and represented as follows:

[0147] V = (v0, v1, v2, ..., v n )

[0148] Where v i This represents the i-th weight from the hidden layer node to the output layer node; then the final output is:

[0149]

[0150] Once the predicted value ^y is obtained, the loss function is defined using MSE. The loss function is as follows:

[0151]

[0152] Where m represents the total number of samples; by performing gradient descent on the neural network through backpropagation, the weights of the first layer of the neural network are adjusted as follows:

[0153]

[0154] Where η is the learning rate, and the partial derivative is... for:

[0155]

[0156] Adjusting the weights of the first layer of the neural network, according to the chain rule, yields:

[0157]

[0158] in, for:

[0159]

[0160] After the update, the first round of the loop is completed; repeat the above loop until the changes between two iterations approach 0, then stop the iteration and the model training is completed; for subsequent scheduling, after the staff schedule is completed, continue to feed new data into the original model for training and continuously optimize the model;

[0161] Once the model is trained, when scheduling shifts, the characteristics of caregivers and patients are preprocessed and then imported into the trained model to generate caregiver ratings.

[0162] Score i =f(X) i X j )

[0163] Where X i X represents the feature vector of the caregiver. j Let f represent the feature vector of the person being cared for, and f be the trained model function, with score as the denominator. i The system assigns a score to each caregiver; ultimately, the caregiver with the highest score is selected from the available caregivers. After implementing this invention, station managers no longer need to manually screen caregivers; they can simply schedule shifts based on the recommended caregivers. The accuracy rate is currently around 80%, and it gradually increases with the amount of scheduling data.

[0164] Embodiments of the present invention also provide a method for intelligent recommendation and scheduling of caregivers, comprising: an information extraction module extracting caregiver characteristics and caregiver characteristics of caregivers; a neural network construction module performing the following actions: constructing a neural network based on caregiver characteristics and caregiver characteristics; and feeding historical data of caregiver characteristics and caregiver characteristics into training to obtain recommendation scores for pairwise pairing of caregiver characteristics and caregiver characteristics, wherein the recommendation scores represent the degree to which the corresponding caregiver should be assigned to the corresponding caregiver; and a scheduling module configured to perform the following actions: ranking caregivers according to the recommendation scores; and scheduling caregivers to provide care to caregivers according to the rankings.

[0165] Because the two service locations are too far apart, caregivers spend most of their time traveling. The large gap between the two service times also means that caregivers spend a lot of time waiting at the elderly person's door for their service time to begin. This has led to an unreasonable allocation of work among caregivers, exacerbating the impact of the caregiver shortage.

[0166] Based on the above insights, this invention provides an intelligent recommendation and scheduling system for caregivers. By extracting caregiver characteristics, elderly person characteristics (caregiver characteristics), existing job characteristics, and future job characteristics, a mathematical model is established. Through data analysis and machine learning, the most suitable scheduling scheme function is learned. Finally, the system schedules future jobs for idle personnel according to the model. In addition, a neural network is constructed based on caregiver and elderly person characteristics, and historical data is used for training. The final result is a score between 0 and 1, which represents the degree to which a caregiver should be assigned to an elderly person. Caregivers who meet the criteria are ranked by score and recommended to the scheduling staff. The scheduling staff prioritizes the highest-scoring caregiver, which can achieve the optimal solution for caregiver job allocation and greatly alleviate the drawbacks caused by short-term caregiver assignment.

[0167] In summary, the above embodiments have provided detailed descriptions of different configurations of the intelligent recommendation and scheduling system for caregivers. Of course, this invention includes, but is not limited to, the configurations listed in the above embodiments. Any modifications made based on the configurations provided in the above embodiments are within the scope of protection of this invention. Those skilled in the art can apply the principles described in the above embodiments to other similar applications.

[0168] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0169] The above description is merely a description of preferred embodiments of the present invention and is not intended to limit the scope of the present invention in any way. Any changes or modifications made by those skilled in the art based on the above disclosure shall fall within the protection scope of the claims.

Claims

1. An intelligent scheduling method, characterized in that, include: Extract caregiver characteristics from caregivers and caregiver characteristics from caregivers; Preprocessing of caregiver characteristics and / or caregiver characteristics; Based on the characteristics of caregivers and caregivers, a backpropagation neural network is constructed with one hidden layer and a single neuron in the output layer. The sigmoid function is used as the activation function for both the hidden and output layers. Historical data on caregiver and careee characteristics are fed into the training to derive recommendation scores for pairwise pairings of careee and caregiver characteristics, wherein the recommendation scores represent the degree to which a corresponding caregiver is assigned to a corresponding careee. Caregivers are ranked and recommended to staff based on their recommended scores. as well as Staff selected the caregiver with the highest score for scheduling; The preprocessing includes: Let the gender of the caregiver and the gender of the person being cared for be G∈{M,F}. Calculate the percentage of service time spent by male and female employees: Where M represents the percentage of service time for male employees, N represents the percentage of service time for female employees, n represents the number of female employees, m represents the number of male employees, and S... i This represents the service duration of the i-th employee; By calculating the ratio of service time between male and female employees, preprocessing is performed to assign gender-appropriate caregivers to those being cared for. Enter the characteristics of the caregiver type according to the requirements of the caregiver: Where r represents the requirements of the person being cared for, and p i Types of caregivers; Caregiver job types include full-time, part-time, internship, and retired. Full-time caregiver positions are given priority in allocation. The formula for the average number of caregiver positions per type is: Where µ i Represents the average number of orders for the i-th type of caregiver, count(S) i ) represents the total number of employees of the i-th type; order(S) ij Let represent the order volume of the j-th caregiver of the i-th type; after calculating the average order volume of caregivers for each job type, substitute it into the softmax function to standardize it while amplifying the weights of advantageous job types: Among them, S i The feature input representing the i-th type of caregiver; The preprocessing also includes: Calculate the time difference between the previous order and the next order based on the end time of the previous order and the start time of the next order: Where T e Indicates the end time of the previous order, T. s Indicates the start time of the next order, T i This represents the threshold time difference. If the time difference between an order and a new order exceeds the threshold time difference, it is considered that a new order can be inserted between these two orders. The threshold time difference is defined according to the business situation. W t P represents the estimated travel time for employees. LN Feature inputs representing the time elapsed since the end of the previous order or the start time of the next order; Calculate the caregiver's estimated travel time based on the estimated travel time from the previous order to the current order and the estimated travel time from the current order to the next order: Where (x) i ,y i ) and (x j ,y j ) represents the latitude and longitude coordinates of the two locations, v represents the estimated speed of the vehicle used by the caregiver, and ρ represents the time redundancy, which is preset based on the actual situation of the caregiver; the standardized mean and variance normalization of the travel time is applied using the modified method. Where X is the travel time required, µ is the average travel time for caregivers, and σ is the variance of travel time for caregivers.

2. The method according to claim 1, characterized in that, Historical data is collected and fed into a neural network for learning to obtain a final model for scheduling. After the staff schedule is completed, new data is fed into the original model for training to continuously optimize the model. When staff need to assign caregivers to caregivers, they can find caregivers who meet the criteria. The data of the caregivers who meet the criteria and the caregivers who are being cared for are combined into a vector and used as the input to the model. The output is sorted and recommended to the staff, who then select the first caregiver to schedule the caregiver. The input features of the care recipient are preprocessed, and the weights calculated by the initialization module are obtained by the expert system based on experience. After collecting features of caregivers and caregivers, a backpropagation (BP) neural network was constructed with one hidden layer and a single output neuron. Both the hidden and output layers used the sigmoid function as the activation function.

3. The method according to claim 2, characterized in that, Also includes: Extract the existing and future work characteristics of caregivers; Based on existing and future work characteristics, the most suitable scheduling scheme function is derived through data analysis and machine learning. Schedule future work for idle personnel based on the scheduling scheme function; as well as Qualified caregivers will be selected based on the future work schedule.

4. The method according to claim 2, characterized in that, The characteristics of the caregiver include one or more of the following: caregiver gender, caregiver age, caregiver type, caregiver work experience, work experience with this company, caregiver employee star rating, caregiver employee score, number of complaints against the caregiver, total number of caregiver orders, caregiver highest education level, caregiver job type, caregiver marital status, caregiver height, caregiver weight, number of orders per day, end time of the previous order, start time of the next order, distance from the previous order, and distance from the next order.

5. The method according to claim 2, characterized in that, The characteristics of the caregiver include one or more of the following: the caregiver's gender, age, duration of the caregiver's contract, health level, average score of the caregiver, number of complaints by the caregiver, number of times the caregiver has been served by the current caregiver, and number of times the caregiver has been served by the current caregiver in the past 30 days.

6. The method according to claim 2, characterized in that: After constructing the BP neural network, weights are assigned to the first layer of the neural network by initializing the weight module; The initial weight module uses an expert system to calculate the weights of the first layer of the neural network. The expert system employs a large-scale multi-expert collaborative system, multi-knowledge representation, comprehensive knowledge base, self-organizing problem-solving mechanism, multi-disciplinary collaborative problem-solving and parallel reasoning, expert system tools and environment, and artificial neural network knowledge acquisition and learning mechanism to calculate the weights of the first layer of the neural network, which has multiple knowledge bases and multiple subjects.

7. The method according to claim 6, characterized in that, The weights of the first layer of the neural network include: The weights for the caregiver's highest education level, marital status, height, weight, gender, and gender of the person being cared for are 0.

1. The weight of the health level of the person being cared for is 0.2; The weights for caregiver type, caregiver job type, caregiver work experience, and length of contract with the caregiver are 0.3; The weight of work experience in this company is 0.4; The weights for caregiver employee star rating, caregiver employee score, number of complaints against caregivers, total number of caregiver cases, caregiver age, age of caregivers, average score of caregivers, and number of complaints against caregivers are all 0.

5. The weight of each caregiver's daily order volume is 0.6; The weighting of the caregiver's previous job end time, the caregiver's next job start time, and the number of times the caregiver has served the patient in the past 30 days is 0.8; and The weights for the distance of the caregiver's previous job, the distance of the caregiver's next job, and the number of times the caregiver has served the patient are all 0.

7.

8. The method according to claim 7, characterized in that, The training process of the model includes: The feature input vector is: Where x1 to x p As a characteristic of caregivers, x p+1 To x n The features of the person being cared for are represented by x0, which is the bias node; the weights of the first layer of the neural network are: Where w ij Let represent the weights from the j-th node in the input layer to the i-th node in the hidden layer. Then the output of the hidden layer is... Substituting it into the Sigmoid function, we get: After obtaining the hidden layer input, forward propagation continues. The weights of the second layer of the neural network are randomly initialized and represented as follows: Where v i This represents the i-th weight from the hidden layer node to the output layer node; then the final output is: Once the predicted value ˆy is obtained, the loss function is defined using MSE. The loss function is as follows: Where m represents the total number of samples; by performing gradient descent on the neural network through backpropagation, the weights of the first layer of the neural network are adjusted as follows: Where η is the learning rate, and the partial derivative ∂L / ∂v is: Adjusting the weights of the first layer of the neural network, according to the chain rule, yields: Where ∂v / ∂w is: After the update, the first round of the loop is completed; repeat the above loop until the changes between two iterations approach 0, then stop the iteration and the model training is completed; for subsequent scheduling, after the staff schedule is completed, continue to feed new data into the original model for training and continuously optimize the model; Once the model is trained, when scheduling shifts, the characteristics of caregivers and patients are preprocessed and then imported into the trained model to generate caregiver ratings. Where X i X represents the feature vector of the caregiver. j Let f represent the feature vector of the person being cared for, and f be the trained model function, with score as the denominator. i The score is assigned to caregiver i; finally, the caregiver with the highest score is selected from the available caregivers.

Images