Intelligent scheduling method and system based on prediction data

Abstract

The invention discloses an intelligent scheduling method and system based on prediction data, and relates to the technical field of network communication, and the method comprises the steps: dividing historical call data into first historical call data, second historical call data and third historical call data through the comprehensive weight of each historical call record in the historical call data by using quantiles, according to the method, the distribution condition of data can be better understood, the data are divided into different levels, pertinent analysis is facilitated, meanwhile, different models are trained by using different algorithms by using the first historical call data, the second historical call data and the third historical call data, and the data analysis efficiency is improved. And the comprehensive weights of the three kinds of historical call data are divided according to three kinds of different information of timestamps, call durations and problem types, so that the robustness of the model can be increased, the influence of abnormal values or extreme conditions on a prediction result is avoided, and the problem that the intelligent typesetting system cannot perform real-time verification due to the fact that the accuracy and integrity of the data are difficult to verify in time is avoided. And the prediction is not accurate.

Description

Technical Field

[0001] The present invention relates to the field of network communication technology, and in particular to an intelligent scheduling method and system based on prediction data. Background Art

[0002] As the demand for call center services continues to increase, reasonable scheduling becomes particularly important. The intelligent scheduling method based on predictive data can optimize the scheduling, improve resource utilization and customer satisfaction by analyzing historical call volume and employee performance. The system can adjust the schedule in real time to cope with business peaks, reduce customer waiting time, and improve employee work efficiency and satisfaction. Reasonable scheduling not only helps to reduce employee turnover, but also improves the overall service quality, thereby improving the economic benefits of the enterprise. Therefore, studying this method is of great significance to the operation and management of call centers.

[0003] However, the current intelligent scheduling methods based on predictive data are difficult to verify the accuracy and completeness of the data in a timely manner, and missing or erroneous data will lead to inaccurate predictions. In addition, the system's ability to respond to emergencies is limited, and it is impossible to quickly adjust the schedule to cope with sudden peaks. Therefore, it is necessary to provide an intelligent scheduling method and system based on predictive data to solve the above-mentioned problems. Summary of the invention

[0004] In order to solve the above technical problems, an intelligent scheduling method and system based on predictive data are provided. The technical solution solves the problem that the accuracy and completeness of data in the current intelligent typesetting method based on predictive data proposed in the above background technology are difficult to verify in time, and missing or erroneous data will lead to inaccurate predictions, and the system has limited response capabilities to emergencies and cannot quickly adjust the schedule to cope with sudden peaks.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] An intelligent scheduling method based on forecast data, comprising:

[0007] Obtain on-duty employee information and historical call data during the scheduling cycle;

[0008] Obtain predicted call data based on historical call data;

[0009] Collect real-time call data, evaluate predicted call data and real-time call data, and obtain call data impact information;

[0010] The predicted call data is optimized with the real-time call data and the call data impact information to obtain the future call data;

[0011] Determine the work schedule information based on the on-duty employee information and future call data within the scheduling cycle;

[0012] Obtain historical employee performance data through on-the-job employee information during the scheduling cycle, and obtain predicted employee performance data of on-the-job employees during the scheduling cycle in combination with the scheduling work information;

[0013] Evaluate the predicted employee performance data of employees on duty during the scheduling cycle to obtain initial scheduling information;

[0014] Based on the on-the-job employee information within the scheduling cycle, the scheduling impact information is obtained, and the initial scheduling information is optimized based on the scheduling impact information to obtain the preferred scheduling information.

[0015] In an optional embodiment, the acquiring predicted call data according to the historical call data specifically includes:

[0016] Based on historical call data, through W s =ωTf(T)+ω D D+ω C f(C), obtain the comprehensive weight W of each historical call record s , where f(T) is the timestamp function, D is the historical call duration, f(C) is the question type function, ω T ,ω D and ω C They are the weight coefficients of the timestamp function, historical call duration, and question type function;

[0017] Determine the total number N of historical call records in the historical call data, and obtain the quantile position L by L=P(N+1), where P is the quantile coefficient;

[0018] The total number of historical call records N in the historical call data and the comprehensive weight W of each historical call record s , obtain the first quantile Q1, the second quantile Q2 and the third quantile Q3;

[0019] According to the first quantile Q1, the second quantile Q2 and the third quantile Q3, the historical call data is divided to obtain first historical call data, second historical call data and third historical call data;

[0020] The comprehensive weight W s The historical call data less than or equal to the first quantile Q1 is classified as the third historical call data, and the comprehensive weight W s The historical call data with a value greater than the first quantile Q1 and less than or equal to the second quantile Q2 are classified as the second historical call data. s The historical call data greater than the second quantile Q2 is classified as the third historical call data;

[0021] Using a linear regression algorithm, an independent model training is performed on the first historical call data to obtain a first prediction model, and a first prediction value output by the first prediction model is obtained, and the first prediction value is used as a first predicted call volume;

[0022] The expression for obtaining the first prediction value output by the first prediction model is:

[0023]

[0024] In the formula, Y1 is the first predicted value, β m is the model parameter, x i is the total number of calls in the i-th time period.

[0025] In an optional embodiment, the acquiring predicted call data according to the historical call data specifically includes:

[0026] Obtain historical social event information corresponding to historical call data;

[0027] Acquiring historical timestamp information corresponding to the historical social event information from the second historical call data;

[0028] Acquire historical call volume information corresponding to the historical timestamp information from the first historical call data;

[0029] Performing data preprocessing on historical social event information, historical timestamp information, and historical call volume information to obtain first historical integrated data;

[0030] Extracting first key feature data from the first historical integrated data, wherein the first key feature data includes event type, impact scope, and duration;

[0031] Performing correlation analysis on the first key feature data to determine the relationship between the event type and the historical call volume information;

[0032] Using a random forest algorithm, a model is trained based on the first key feature data and the first historical integrated data to obtain a second prediction model, and a second prediction value output by the second prediction model is obtained;

[0033] taking the sum of the second prediction values ​​output by the second prediction model as the second predicted call volume;

[0034] The expression formula for obtaining the second prediction value output by the second prediction model is:

[0035]

[0036] Where, Y2 is the second predicted call volume, F a (z) is the sum of the second prediction values ​​of a historical social events output by the second prediction model, Fa-1 (z) is the sum of the second prediction values ​​of a-1 historical social events output by the second prediction model, γ a is the weight coefficient of the a-th historical social event, L is the model function, y a is the event type label of the ath historical social event, f a (T) is the timestamp corresponding to the ath historical social event, and b is the total number of historical social events.

[0037] In an optional embodiment, the acquiring predicted call data according to the historical call data specifically includes:

[0038] Obtaining historical question type information in the third historical call data;

[0039] Natural language processing is performed on the historical question type information in the third historical call data to obtain key text information;

[0040] Conduct correlation analysis between key text information and historical social event information in historical call data;

[0041] Performing data preprocessing on key text information corresponding to historical social event information in historical call data to obtain related text information;

[0042] Extracting the historical social event information corresponding to the associated text information to obtain the associated historical social event information;

[0043] Acquire historical timestamp information corresponding to the associated historical social event information from the second historical call data to obtain the associated historical timestamp information;

[0044] Acquire historical call volume information corresponding to the associated historical timestamp information from the first historical call data to obtain the associated historical call volume information;

[0045] Performing data preprocessing on the associated historical social event information, the associated historical timestamp information, and the associated historical call volume information to obtain first associated historical integrated data;

[0046] extracting first associated key feature data from the first associated historical integrated data;

[0047] Using a random forest algorithm, a model is trained based on the first associated key feature data and the first associated historical integrated data to obtain a third prediction model, and a third prediction value output by the third prediction model is obtained;

[0048] The sum of the third prediction values ​​output by the third prediction model is used as the third predicted call volume;

[0049] The sum of the first predicted call volume, the second predicted call volume and the third predicted call volume is taken as the predicted call volume to obtain the predicted call data.

[0050] In an optional embodiment, the collecting of real-time call data, and evaluating the predicted call data and the real-time call data to obtain the call data impact information specifically includes:

[0051] Compare the predicted call data with the real-time call data to obtain the call volume difference value;

[0052] Based on the predicted call data, obtain the call volume difference threshold;

[0053] If the call volume difference value exceeds the call volume difference threshold, obtaining real-time social event information related to the real-time call data based on the real-time call data;

[0054] Respectively obtaining real-time event type information in the real-time social event information and historical event type information in the historical social event information;

[0055] The real-time event type information and the historical event type information are evaluated, and the real-time social event information that is different from the historical event type information is extracted to obtain the call data impact information.

[0056] In an optional embodiment, the optimizing the predicted call data with the real-time call data and the call data impact information to obtain the future call data specifically includes:

[0057] Acquire real-time timestamp information corresponding to the call data impact information from the real-time call data;

[0058] The call data impact information and the real-time timestamp information are pre-processed and then input into the second prediction model to obtain the call data impact amount output by the second prediction model;

[0059] Add the predicted call volume in the predicted call data to the call data impact amount to obtain the future call volume;

[0060] The data of future call volume is integrated to obtain future call data.

[0061] In an optional embodiment, determining the scheduling work information based on the on-duty employee information and future call data within the scheduling period specifically includes:

[0062] According to the future call data, the future call volume is obtained;

[0063] Obtain the number of employees on duty during the scheduling period through the on-duty employee information during the scheduling period;

[0064] Based on the future call volume and the number of employees on duty during the scheduling period, obtain the average number of calls per person;

[0065] Determine the work schedule information based on the average number of calls per person.

[0066] In an optional embodiment, evaluating the predicted employee performance data of employees on duty within the scheduling period to obtain initial scheduling information specifically includes:

[0067] Extract the maximum and minimum call volumes of each employee from the predicted employee performance data of employees on duty during the scheduling period, and obtain the customer rating of each employee;

[0068] According to each employee's maximum call volume, minimum call volume and customer score, the on-duty employees within the scheduling cycle are scheduled to obtain initial scheduling information.

[0069] In an optional embodiment, the obtaining of scheduling impact information based on the on-duty employee information within the scheduling period, and optimizing the initial scheduling information based on the scheduling impact information to obtain the preferred scheduling information specifically includes:

[0070] Based on the information of employees on duty within the scheduling cycle, send initial scheduling information to all employees and obtain scheduling feedback information from all employees;

[0071] Evaluate the scheduling feedback information and obtain scheduling impact information;

[0072] The initial shift scheduling information is adjusted using the shift scheduling impact information to obtain the optimal shift scheduling information.

[0073] Furthermore, an intelligent scheduling system based on prediction data is proposed, which is used to implement any of the scheduling methods described above, including:

[0074] A main control module, which is used to obtain on-duty employee information within the scheduling cycle, collect historical call data and real-time call data for training the prediction model;

[0075] An evaluation module, the evaluation module is used to evaluate the predicted call data and the real-time call data, obtain call data impact information, and evaluate the predicted employee performance data of on-the-job employees within the scheduling cycle;

[0076] A scheduling module, wherein the scheduling module is used to obtain initial scheduling information and preferred scheduling information;

[0077] A transmission module, which is used for data transmission between the main control module, the evaluation module, the scheduling module and the display module;

[0078] The display module is used to display the data and information processing process and results of the main control module, to display the evaluation process and results of the evaluation module on the data and information, and to display the scheduling process and results of the scheduling module.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] This scheme proposes an intelligent scheduling method and system based on prediction data. Through the comprehensive weight of each historical call record in the historical call data, and using quantiles to divide it into the first historical call data, the second historical call data and the third historical call data, the distribution of the data can be better understood, and the data can be divided into different levels for targeted analysis. At the same time, different models are trained using different algorithms using the first historical call data, the second historical call data and the third historical call data. The comprehensive weight of the three types of historical call data is divided according to three different information: timestamp, call duration and question type. This can increase the robustness of the model, avoid the prediction results being affected by outliers or extreme situations, and avoid inaccurate predictions when the accuracy and completeness of the data are difficult to verify in time for the intelligent typesetting system.

[0081] This solution proposes an intelligent scheduling method and system based on predictive data. By analyzing historical social event information, training a model based on historical social event information, historical timestamp information, and historical call volume information, and using the model to predict call volume, the system's ability to respond to emergencies is improved, and the layout can be quickly adjusted to cope with sudden peaks. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 A flowchart of an intelligent scheduling method based on forecast data proposed by the present invention;

[0083] Figure 2 This is a flowchart for obtaining the first predicted call volume in the present invention;

[0084] Figure 3 This is a flowchart for obtaining the second predicted call volume in the present invention;

[0085] Figure 4 This is a system framework diagram of an intelligent scheduling system based on predictive data proposed in the present invention. DETAILED DESCRIPTION

[0086] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0087] Reference Figure 1 - Figure 4As shown, an intelligent scheduling method based on forecast data includes:

[0088] Obtain on-duty employee information and historical call data during the scheduling cycle;

[0089] Obtain predicted call data based on historical call data;

[0090] Collect real-time call data, evaluate predicted call data and real-time call data, and obtain call data impact information;

[0091] The predicted call data is optimized with the real-time call data and the call data impact information to obtain the future call data;

[0092] Determine the work schedule information based on the on-duty employee information and future call data within the scheduling cycle;

[0093] Obtain historical employee performance data through on-the-job employee information during the scheduling cycle, and obtain predicted employee performance data of on-the-job employees during the scheduling cycle in combination with the scheduling work information;

[0094] Evaluate the predicted employee performance data of employees on duty during the scheduling cycle to obtain initial scheduling information;

[0095] Based on the on-the-job employee information within the scheduling cycle, the scheduling impact information is obtained, and the initial scheduling information is optimized based on the scheduling impact information to obtain the preferred scheduling information.

[0096] Furthermore, based on the historical call data, the predicted call data is obtained, including:

[0097] Based on historical call data, through W s =ω T f(T)+ω D D+ω C f(C), obtain the comprehensive weight W of each historical call record s , where f(T) is the timestamp function, D is the historical call duration, f(C) is the question type function, ω T ,ω D and ω C They are the weight coefficients of the timestamp function, historical call duration, and question type function;

[0098] Determine the total number N of historical call records in the historical call data, and obtain the quantile position L by L=P(N+1), where P is the quantile coefficient;

[0099] The total number of historical call records N in the historical call data and the comprehensive weight W of each historical call record s , obtain the first quantile Q1, the second quantile Q2 and the third quantile Q3;

[0100] According to the first quantile Q1, the second quantile Q2 and the third quantile Q3, the historical call data is divided to obtain first historical call data, second historical call data and third historical call data;

[0101] The comprehensive weight W s The historical call data less than or equal to the first quantile Q1 is classified as the third historical call data, and the comprehensive weight W s The historical call data with a value greater than the first quantile Q1 and less than or equal to the second quantile Q2 are classified as the second historical call data. s The historical call data greater than the second quantile Q2 is classified as the third historical call data;

[0102] Using a linear regression algorithm, an independent model training is performed on the first historical call data to obtain a first prediction model, and a first prediction value output by the first prediction model is obtained, and the first prediction value is used as a first predicted call volume;

[0103] The expression for obtaining the first prediction value output by the first prediction model is:

[0104]

[0105] In the formula, Y1 is the first predicted value, β m is the model parameter, x i is the total number of calls in the i-th time period.

[0106] Specifically, because the data containing timestamp information in the historical call data may not contain the historical call duration or question type information, the data containing the historical call duration information may not contain the timestamp information or question type information, and the data containing the question type information may not contain the historical call duration or timestamp information, and the historical call data may have some outliers and missing values, so it can be based on the historical call data through W s =ω T f(T)+ω D D+ω C f(C), obtain the comprehensive weight W of each historical call record sAmong them, the timestamp function adopts exponential decay or linear decay, and assigns higher weights to newer call records. The sine or cosine function is used to reflect the time period (such as hours, days of the week, etc.) to ensure that the model can capture the impact of time on call volume. The call duration is directly used as a weight or processed through normalization (such as min-max standardization), and is adapted to the weights of other features in the calculation of the comprehensive weight. At the same time, in order to reduce the impact of extreme values, the call duration can also be square root or logarithmic transformed to adjust its weight. For the weight of the question type function, unique hot encoding is used to convert different question types into numerical features, and then manually assign weights according to their importance, or use machine learning models (such as random forests) to automatically learn the importance of each type. If some question types are more important, they can be directly assigned higher weights, which are usually determined by business experience or historical data analysis.

[0107] It can be understood that the total number N of historical call records in the historical call data and the comprehensive weight W of each historical call record s , obtain the first quantile Q1, the second quantile Q2 and the third quantile Q3, determine the total number N of historical call records in the historical call data, and obtain the quantile position L through L=P(N+1), where P is the quantile coefficient, and the value of P is 0.25, 0.5 and 0.75. If the quantile position L is an integer, then the first quantile Q1, the second quantile Q2 and the third quantile Q3 are the values ​​represented by the position of the quantile position L. If the quantile position L is not an integer, the two adjacent comprehensive weights W are rounded up. s The value of W k and W k+1 , and then use the interpolation formula Q = W k +(Lk)(W k+1 -W k ) calculates the quantiles (first quantile Q1, second quantile Q2, and third quantile Q3).

[0108] Furthermore, based on the historical call data, the predicted call data is obtained, including:

[0109] Obtain historical social event information corresponding to historical call data;

[0110] Acquiring historical timestamp information corresponding to the historical social event information from the second historical call data;

[0111] Acquire historical call volume information corresponding to the historical timestamp information from the first historical call data;

[0112] Performing data preprocessing on historical social event information, historical timestamp information, and historical call volume information to obtain first historical integrated data;

[0113] Extracting first key feature data from the first historical integrated data, the first key feature data including event type, impact scope, and duration;

[0114] Performing correlation analysis on the first key feature data to determine the relationship between the event type and the historical call volume information;

[0115] Using a random forest algorithm, a model is trained based on the first key feature data and the first historical integrated data to obtain a second prediction model, and a second prediction value output by the second prediction model is obtained;

[0116] taking the sum of the second prediction values ​​output by the second prediction model as the second predicted call volume;

[0117] The expression formula for obtaining the second prediction value output by the second prediction model is:

[0118]

[0119] Where, Y2 is the second predicted call volume, F a (z) is the sum of the second prediction values ​​of a historical social events output by the second prediction model, F a-1 (z) is the sum of the second prediction values ​​of a-1 historical social events output by the second prediction model, γ a is the weight coefficient of the a-th historical social event, L is the model function, y a is the event type label of the ath historical social event, f a (Y) is the timestamp corresponding to the ath historical social event, and b is the total number of historical social events.

[0120] Furthermore, based on the historical call data, the predicted call data is obtained, including:

[0121] Obtaining historical question type information in the third historical call data;

[0122] Natural language processing is performed on the historical question type information in the third historical call data to obtain key text information;

[0123] Conduct correlation analysis between key text information and historical social event information in historical call data;

[0124] Performing data preprocessing on key text information corresponding to historical social event information in historical call data to obtain related text information;

[0125] Extracting the historical social event information corresponding to the associated text information to obtain the associated historical social event information;

[0126] Acquire historical timestamp information corresponding to the associated historical social event information from the second historical call data to obtain the associated historical timestamp information;

[0127] Acquire historical call volume information corresponding to the associated historical timestamp information from the first historical call data to obtain the associated historical call volume information;

[0128] Performing data preprocessing on the associated historical social event information, the associated historical timestamp information, and the associated historical call volume information to obtain first associated historical integrated data;

[0129] extracting first associated key feature data from the first associated historical integrated data;

[0130] Using a random forest algorithm, a model is trained based on the first associated key feature data and the first associated historical integrated data to obtain a third prediction model, and a third prediction value output by the third prediction model is obtained;

[0131] The sum of the third prediction values ​​output by the third prediction model is used as the third predicted call volume;

[0132] The sum of the first predicted call volume, the second predicted call volume and the third predicted call volume is taken as the predicted call volume to obtain the predicted call data.

[0133] Furthermore, real-time call data is collected, and the predicted call data and real-time call data are evaluated to obtain call data impact information, including:

[0134] Compare the predicted call data with the real-time call data to obtain the call volume difference value;

[0135] Based on the predicted call data, obtain the call volume difference threshold;

[0136] If the call volume difference value exceeds the call volume difference threshold, obtaining real-time social event information related to the real-time call data based on the real-time call data;

[0137] Respectively obtaining real-time event type information in the real-time social event information and historical event type information in the historical social event information;

[0138] The real-time event type information and the historical event type information are evaluated, and the real-time social event information that is different from the historical event type information is extracted to obtain the call data impact information.

[0139] Furthermore, the predicted call data is optimized with the real-time call data and the call data impact information to obtain future call data, including:

[0140] Acquire real-time timestamp information corresponding to the call data impact information from the real-time call data;

[0141] The call data impact information and the real-time timestamp information are pre-processed and then input into the second prediction model to obtain the call data impact amount output by the second prediction model;

[0142] Add the predicted call volume in the predicted call data to the call data impact amount to obtain the future call volume;

[0143] The data of future call volume is integrated to obtain future call data.

[0144] Furthermore, the on-duty employee information and future call data within the scheduling period are used to determine the scheduling work information, including:

[0145] According to the future call data, the future call volume is obtained;

[0146] Obtain the number of employees on duty during the scheduling period through the on-duty employee information during the scheduling period;

[0147] Based on the future call volume and the number of employees on duty during the scheduling period, obtain the average number of calls per person;

[0148] Determine the work schedule information based on the average number of calls per person.

[0149] Furthermore, the predicted employee performance data of employees on duty during the scheduling period is evaluated to obtain initial scheduling information, including:

[0150] Extract the maximum and minimum call volumes of each employee from the predicted employee performance data of employees on duty during the scheduling period, and obtain the customer rating of each employee;

[0151] According to each employee's maximum call volume, minimum call volume and customer score, the on-duty employees within the scheduling cycle are scheduled to obtain initial scheduling information.

[0152] Furthermore, based on the on-the-job employee information within the scheduling cycle, the scheduling impact information is obtained, and the initial scheduling information is optimized based on the scheduling impact information to obtain the preferred scheduling information, specifically including:

[0153] Based on the information of employees on duty within the scheduling cycle, send initial scheduling information to all employees and obtain scheduling feedback information from all employees;

[0154] Evaluate the scheduling feedback information and obtain scheduling impact information;

[0155] The initial shift scheduling information is adjusted using the shift scheduling impact information to obtain the optimal shift scheduling information.

[0156] Furthermore, an intelligent scheduling system based on prediction data is proposed, which is used to implement any of the scheduling methods described above, including:

[0157] The main control module is used to obtain the on-duty employee information within the scheduling cycle, collect historical call data and real-time call data for training the prediction model;

[0158] Evaluation module, which is used to evaluate the predicted call data and real-time call data, obtain the call data impact information, and evaluate the predicted employee performance data of employees on duty during the scheduling cycle;

[0159] Scheduling module: The typesetting module is used to obtain initial scheduling information and preferred scheduling information;

[0160] Transmission module, which is used for data transmission between the main control module, the evaluation module, the scheduling module and the display module;

[0161] Display module: The display module is used to display the data and information processing process and results of the main control module, to display the evaluation process and results of the evaluation module on the data and information, and to display the scheduling process and results of the scheduling module.

[0162] In summary, the advantages of the present invention are: by using the comprehensive weight of each historical call record in the historical call data and using quantiles to divide it into the first historical call data, the second historical call data and the third historical call data, the distribution of the data can be better understood, and the data can be divided into different levels for targeted analysis. At the same time, different models are trained using different algorithms using the first historical call data, the second historical call data and the third historical call data, and the comprehensive weights of the three types of historical call data are divided according to three different information, namely, timestamp, call duration and question type, which can increase the robustness of the model, avoid the prediction results being affected by outliers or extreme situations, and avoid the intelligent typesetting system from being unable to verify the accuracy and completeness of the data in a timely manner, resulting in inaccurate predictions;

[0163] By analyzing historical social event information, training models based on historical social event information, historical timestamp information, and historical call volume information, and using the model to predict call volume, the system's ability to respond to emergencies is improved, and the layout can be quickly adjusted to cope with sudden peaks.

[0164] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. An intelligent scheduling method based on forecast data, characterized in that: include: Obtain employee information and historical call data during the shift schedule; Obtain predicted call data based on historical call data; Collect real-time call data, evaluate predicted call data and real-time call data, and obtain call data impact information; The predicted call data is optimized with the real-time call data and the call data impact information to obtain the future call data; Determine the work schedule information based on the on-duty employee information and future call data within the scheduling cycle; Obtain historical employee performance data through on-the-job employee information during the scheduling cycle, and obtain predicted employee performance data of on-the-job employees during the scheduling cycle in combination with the scheduling work information; Evaluate the predicted employee performance data of employees on duty during the scheduling cycle to obtain initial scheduling information; Based on the on-the-job employee information within the scheduling cycle, the scheduling impact information is obtained, and the initial scheduling information is optimized based on the scheduling impact information to obtain the preferred scheduling information.

2. The intelligent scheduling method based on prediction data according to claim 1 is characterized in that: The obtaining of predicted call data according to the historical call data specifically includes: Based on historical call data, through W s =ω T f(T)+ω D D+ω C f(C), obtain the comprehensive weight W of each historical call record s , where f(T) is the timestamp function, D is the historical call duration, f(C) is the question type function, ω T ,ω D and ω C They are the weight coefficients of the timestamp function, historical call duration, and question type function; Determine the total number N of historical call records in the historical call data, and obtain the quantile position L by L=P(N+1), where P is the quantile coefficient; The total number of historical call records N in the historical call data and the comprehensive weight W of each historical call record s , obtain the first quantile Q1, the second quantile Q2 and the third quantile Q3; According to the first quantile Q1, the second quantile Q2 and the third quantile Q3, the historical call data is divided to obtain first historical call data, second historical call data and third historical call data; The comprehensive weight W s The historical call data less than or equal to the first quantile Q1 is classified as the third historical call data, and the comprehensive weight W s The historical call data with a value greater than the first quantile Q1 and less than or equal to the second quantile Q2 are classified as the second historical call data. s The historical call data greater than the second quantile Q2 is classified as the third historical call data; Using a linear regression algorithm, an independent model training is performed on the first historical call data to obtain a first prediction model, and a first prediction value output by the first prediction model is obtained, and the first prediction value is used as a first predicted call volume; The expression for obtaining the first prediction value output by the first prediction model is: In the formula, Y1 is the first predicted value, β m is the model parameter, x i is the total number of calls in the i-th time period.

3. The intelligent scheduling method based on forecast data according to claim 2 is characterized in that: The obtaining of predicted call data according to the historical call data specifically includes: Obtain historical social event information corresponding to historical call data; Acquiring historical timestamp information corresponding to the historical social event information from the second historical call data; Acquire historical call volume information corresponding to the historical timestamp information from the first historical call data; Performing data preprocessing on historical social event information, historical timestamp information, and historical call volume information to obtain first historical integrated data; Extracting first key feature data from the first historical integrated data, wherein the first key feature data includes event type, impact scope, and duration; Performing correlation analysis on the first key feature data to determine the relationship between the event type and the historical call volume information; Using a random forest algorithm, a model is trained based on the first key feature data and the first historical integrated data to obtain a second prediction model, and a second prediction value output by the second prediction model is obtained; taking the sum of the second prediction values ​​output by the second prediction model as the second predicted call volume; The expression formula for obtaining the second prediction value output by the second prediction model is: Where, Y2 is the second predicted call volume, F a (z) is the sum of the second prediction values ​​of a historical social events output by the second prediction model, F a-1 (z) is the sum of the second prediction values ​​of a-1 historical social events output by the second prediction model, γ a is the weight coefficient of the a-th historical social event, L is the model function, y a is the event type label of the ath historical social event, f a (T) is the timestamp corresponding to the ath historical social event, and b is the total number of historical social events.

4. The intelligent scheduling method based on forecast data according to claim 3 is characterized in that: The obtaining of predicted call data according to the historical call data specifically includes: Obtaining historical question type information in the third historical call data; Natural language processing is performed on the historical question type information in the third historical call data to obtain key text information; Conduct correlation analysis between key text information and historical social event information in historical call data; Performing data preprocessing on key text information corresponding to historical social event information in historical call data to obtain related text information; Extracting the historical social event information corresponding to the associated text information to obtain the associated historical social event information; Acquire historical timestamp information corresponding to the associated historical social event information from the second historical call data to obtain the associated historical timestamp information; Acquire historical call volume information corresponding to the associated historical timestamp information from the first historical call data to obtain the associated historical call volume information; Performing data preprocessing on the associated historical social event information, the associated historical timestamp information, and the associated historical call volume information to obtain first associated historical integrated data; extracting first associated key feature data from the first associated historical integrated data; Using a random forest algorithm, a model is trained based on the first associated key feature data and the first associated historical integrated data to obtain a third prediction model, and a third prediction value output by the third prediction model is obtained; The sum of the third prediction values ​​output by the third prediction model is used as the third predicted call volume; The sum of the first predicted call volume, the second predicted call volume and the third predicted call volume is taken as the predicted call volume to obtain the predicted call data.

5. The intelligent scheduling method based on forecast data according to claim 4 is characterized in that: The collecting of real-time call data, and evaluating the predicted call data and the real-time call data to obtain the call data impact information specifically includes: Compare the predicted call data with the real-time call data to obtain the call volume difference value; Based on the predicted call data, obtain the call volume difference threshold; If the call volume difference value exceeds the call volume difference threshold, obtaining real-time social event information related to the real-time call data based on the real-time call data; Respectively obtaining real-time event type information in the real-time social event information and historical event type information in the historical social event information; The real-time event type information and the historical event type information are evaluated, and the real-time social event information that is different from the historical event type information is extracted to obtain the call data impact information.

6. The intelligent scheduling method based on forecast data according to claim 5, characterized in that: The optimizing the predicted call data by using the real-time call data and the call data impact information to obtain the future call data specifically includes: Acquire real-time timestamp information corresponding to the call data impact information from the real-time call data; The call data impact information and the real-time timestamp information are pre-processed and then input into the second prediction model to obtain the call data impact amount output by the second prediction model; Add the predicted call volume in the predicted call data to the call data impact amount to obtain the future call volume; The data of future call volume is integrated to obtain future call data.

7. The intelligent scheduling method based on forecast data according to claim 6, characterized in that: The above-mentioned method of determining the shift work information based on the on-duty employee information and future call data within the shift cycle specifically includes: According to the future call data, the future call volume is obtained; Obtain the number of employees on duty during the scheduling period through the on-duty employee information during the scheduling period; Based on the future call volume and the number of employees on duty during the scheduling period, obtain the average number of calls per person; Determine the work schedule information based on the average number of calls per person.

8. The intelligent scheduling method based on forecast data according to claim 7 is characterized in that: The predicted employee performance data of employees on duty within the scheduling period is evaluated to obtain initial scheduling information, specifically including: Extract the maximum and minimum call volumes of each employee from the predicted employee performance data of employees on duty during the scheduling period, and obtain the customer rating of each employee; According to each employee's maximum call volume, minimum call volume and customer score, the on-duty employees within the scheduling cycle are scheduled to obtain initial scheduling information.

9. The intelligent scheduling method based on prediction data according to claim 8, characterized in that: The method of obtaining the scheduling impact information based on the on-the-job employee information within the scheduling cycle, and optimizing the initial scheduling information based on the scheduling impact information to obtain the preferred scheduling information specifically includes: Based on the information of employees on duty within the scheduling cycle, send initial scheduling information to all employees and obtain scheduling feedback information from all employees; Evaluate the scheduling feedback information and obtain scheduling impact information; The initial shift scheduling information is adjusted using the shift scheduling impact information to obtain the optimal shift scheduling information.

10. An intelligent scheduling system based on prediction data, used to implement the scheduling method according to any one of claims 1 to 9, characterized in that: include: A main control module, which is used to obtain on-duty employee information within the scheduling cycle, collect historical call data and real-time call data for training the prediction model; An evaluation module, the evaluation module is used to evaluate the predicted call data and the real-time call data, obtain call data impact information, and evaluate the predicted employee performance data of on-the-job employees within the scheduling cycle; A scheduling module, wherein the scheduling module is used to obtain initial scheduling information and preferred scheduling information; A transmission module, which is used for data transmission between the main control module, the evaluation module, the scheduling module and the display module; The display module is used to display the data and information processing process and results of the main control module, to display the evaluation process and results of the evaluation module on the data and information, and to display the scheduling process and results of the scheduling module.

Images