A distributed scheduling method and system for elevator groups based on machine learning

Abstract

The present invention discloses a distributed scheduling method and system for elevator groups based on machine learning, which is used in the field of elevator control. The distributed scheduling method for the elevator group includes the following steps: using the pre-configured sensors on each floor of the building to monitor the flow of people and the load of elevators in real time, and collecting the historical data of the flow of people and the load of elevators at the same time; using the prophet prediction algorithm to predict the flow of people within a preset time in the future, based on the prediction results, combined with the real-time monitored flow of people data, using a multi-objective optimization algorithm to balance the waiting time of passengers, the operating efficiency of elevators and energy consumption; using the real-time monitored flow of people data and elevator load data, combined with the floor height and the predicted flow of people data, dynamically adjusting the layered service of the elevator. The present invention reduces the waiting time of passengers and improves the passenger experience by accurately predicting peak hours and adjusting elevator scheduling.

Description

Technical Field

[0001] The present invention relates to the field of elevator control, and more specifically, to a distributed scheduling method and system for elevator groups based on machine learning. Background Art

[0002] As a means of vertical transportation, elevators are indispensable for modern high-rise buildings. They not only improve the utilization efficiency of buildings, but are also important facilities connecting different floors, greatly facilitating people's daily life and work. As the height of buildings increases, the design and technology of elevators are also constantly improving. From early mechanical elevators to current intelligent elevator systems, the safety, comfort and efficiency of elevators have been significantly improved.

[0003] Elevator groups refer to multiple elevators in the same building that are connected by a dispatching system. This system allows multiple elevators to work together to optimize passenger service processes and effectively manage passenger flow and carrying capacity. The elevator group dispatching system can intelligently allocate passengers, reduce waiting and travel time, and improve the efficiency and energy utilization of the elevator system.

[0004] Traditional elevator systems may not be able to respond quickly to increased passenger demand during peak hours, causing passengers to wait for a long time; previous elevator dispatching systems may have failed to optimize energy consumption, resulting in energy waste during elevator operation during non-peak hours; traditional systems may not fully utilize historical and real-time data for analysis, resulting in the inability to flexibly adjust elevator dispatching strategies according to actual needs. Previous elevator dispatching decisions may have relied more on experience rather than data-driven, lacking precise trend analysis and forecasting support. When encountering emergencies or unconventional demands, traditional elevator dispatching systems may find it difficult to quickly adjust strategies to respond.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In order to overcome the above problems, the present invention aims to propose a distributed scheduling method and system for elevator groups based on machine learning, with the aim of solving the problem that traditional systems may not make full use of historical and real-time data for analysis, resulting in the inability to flexibly adjust elevator scheduling strategies according to actual needs. Previous elevator scheduling decisions may rely more on experience rather than data-driven, lacking accurate trend analysis and prediction support. When encountering emergencies or unconventional demands, traditional elevator scheduling systems may find it difficult to quickly adjust strategies to cope with the problem.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a distributed scheduling method for an elevator group based on machine learning is provided, and the distributed scheduling method for an elevator group comprises the following steps:

[0009] S1. Use pre-configured sensors on each floor of the building to monitor the flow of people and elevator load data in real time, and collect historical data on flow of people and elevator load;

[0010] S2. Based on the historical data of passenger flow and elevator load, the prophet prediction algorithm is used to predict the passenger flow within a preset time in the future. Based on the prediction results and combined with the real-time monitored passenger flow data, a multi-objective optimization algorithm is used to balance the passenger waiting time, elevator operation efficiency and energy consumption, and formulate an elevator operation scheduling strategy;

[0011] S3. Using the real-time monitored passenger flow data and elevator load data, combined with the floor height and predicted passenger flow data, dynamically adjust the elevator's tiered services and optimize the elevator operation scheduling strategy, while monitoring the performance indicators of the optimized elevator operation scheduling strategy in real time;

[0012] Among them, based on the historical data of passenger flow and elevator load, the prophet prediction algorithm is used to predict the passenger flow within the preset time in the future. Based on the prediction results, combined with the real-time monitored passenger flow data, a multi-objective optimization algorithm is used to balance passenger waiting time, elevator operation efficiency and energy consumption, and the elevator operation scheduling strategy is formulated, including the following steps:

[0013] S21, obtaining historical data of passenger flow and elevator load;

[0014] S22. Use the prophet prediction algorithm to perform time series analysis on historical data and extract trend items and cycle items respectively;

[0015] S23, based on the extracted trend items and cycle items, using the prophet prediction model to predict the trend and cycle changes of passenger flow within a preset time in the future;

[0016] S24, collecting real-time passenger flow and elevator load data, and adjusting the prediction results of the prophet prediction model according to the real-time data;

[0017] S25. extract random fluctuation items from the verification sequence, use the random fluctuation items to train the extreme gradient boosting model, and predict random fluctuations in the future time period;

[0018] S26. Combining the trend term and the period term predicted by the prophet prediction model with the random fluctuation term predicted by the extreme gradient boosting model to form a complete prediction result of passenger flow and elevator load;

[0019] S27. Define optimization objectives, apply multi-objective optimization algorithms, and formulate elevator operation scheduling strategies based on the complete passenger flow and elevator load prediction results.

[0020] Optionally, using the prophet prediction algorithm to perform time series analysis on historical data, extracting trend items and period items respectively includes the following steps:

[0021] S221, constructing a prophet prediction model and configuring parameters of the prophet prediction model;

[0022] S222, inputting historical data into the prophet prediction model, and extracting trend items and seasonal cycle items;

[0023] S223, analyzing trend items and periodic items to identify long-term change trends and periodic fluctuation patterns of historical data;

[0024] S224. Evaluate the prediction accuracy of the prophet prediction model and the degree of fit between the trend and cyclical terms through statistical error indicators.

[0025] Optionally, based on the extracted trend items and cycle items, using the prophet prediction model to predict the trend and cycle changes of the flow of people within a preset time in the future includes the following steps:

[0026] S231, setting a future preset time range;

[0027] S232, creating a data framework of future time points based on a future preset time range;

[0028] S233, input the future date data frame into the prophet prediction model, and use the predict method to make predictions;

[0029] S234. Extract the overall trend and periodic changes of passenger flow from the prediction results.

[0030] Optionally, extracting random fluctuation terms from the verification sequence, training the extreme gradient boosting model using the random fluctuation terms, and predicting random fluctuations in a future time period include the following steps:

[0031] S251, selecting a period of passenger flow and elevator load data from actual applications as a verification sequence;

[0032] S252, using the prophet prediction model to predict the verification sequence, obtain the predicted values ​​of the trend term and the cycle term, and subtract the predicted values ​​of the trend term and the cycle term from the actual observed value to obtain the estimated value of the random fluctuation term;

[0033] S253, taking the random fluctuation term as the target variable, and preparing the feature variable for training the extreme gradient boosting model;

[0034] S254, using the random fluctuation term and the characteristic variable as inputs of the extreme gradient boosting model, training the extreme gradient boosting model at the same time, and verifying the trained extreme gradient boosting model;

[0035] S255. Use the verified extreme gradient boosting model to predict random fluctuations in future time periods.

[0036] Optionally, the expression for the estimated value of the random fluctuation term is:

[0037] ;

[0038] In the formula, represents the random fluctuation term at time point t;

[0039] represents the actual observation value at time point t;

[0040] represents the trend term at time point t;

[0041] represents the periodic term at time point t.

[0042] Optionally, combining the trend term and the period term predicted by the prophet prediction model with the random fluctuation term predicted by the extreme gradient boosting model to form a complete prediction result of passenger flow and elevator load includes the following steps:

[0043] S261, extracting the forecast values ​​of the trend item and the cycle item from the forecast results of the prophet forecast model;

[0044] S262. Use the trained extreme gradient boosting model to predict random fluctuations in the same future time period;

[0045] S263, adding the prediction results of the trend item, the period item and the random fluctuation item to calculate the complete predicted values ​​of the passenger flow and the elevator load;

[0046] S264. Compare the calculated predicted value with the actual value, and use the root mean square error evaluation index to quantify the accuracy of the prediction.

[0047] Optionally, the root mean square error evaluation indicator is used to quantify the accuracy of the prediction:

[0048] ;

[0049] In the formula, represents the root mean square error;

[0050] n represents the total number of data points utilized when evaluating the prediction accuracy;

[0051] Fi represents the value of passenger flow or elevator load predicted by the model at the i-th time point;

[0052] Ti represents the actual passenger flow or elevator load value at the i-th time point.

[0053] Optionally, defining optimization objectives, applying a multi-objective optimization algorithm, and formulating an elevator operation scheduling strategy based on the complete prediction results of passenger flow and elevator load include the following steps:

[0054] S271. Define optimization goals based on the key performance indicators of elevator system operation;

[0055] S272. According to the multi-objective characteristics of the elevator dispatching problem, a Pareto optimal genetic algorithm suitable for the multi-objective problem is selected;

[0056] S273. Based on the Pareto optimal genetic algorithm, a mathematical model of elevator dispatching is established, and decision variables, constraints and objective functions are defined, and parameters of the Pareto optimal genetic algorithm are set;

[0057] S274, inputting the prediction results of passenger flow and elevator load into a mathematical model to predict the expected working condition of the elevator system in a future period;

[0058] S275. Run the Pareto optimal genetic algorithm using the set parameters, search for the Pareto optimal solution set through an iterative process, and select the final elevator scheduling strategy from the Pareto optimal solution set.

[0059] Optionally, using the real-time monitored passenger flow data and elevator load data, combined with the floor height and predicted passenger flow data, dynamically adjusting the elevator's stratified service and optimizing the elevator operation scheduling strategy, while real-time monitoring the performance indicators of the optimized elevator operation scheduling strategy includes the following steps:

[0060] S31. Collect the load data in the elevator and the flow of people on each floor in real time, combine historical patterns and prediction models, estimate future utilization trends, and build a comprehensive operational data framework;

[0061] S32. In the operational data framework, analyze the collected data to identify peak hours and hot spots with high traffic volume;

[0062] S33. Based on the recognition results, dynamically adjust the elevator's hierarchical service strategy and use genetic algorithms to perform intelligent optimization of elevator scheduling;

[0063] S34. Implement real-time monitoring mechanism, evaluate the operation performance of elevators and collect passenger feedback;

[0064] S35. Dynamically adjust the elevator dispatch strategy based on real-time monitoring data and passenger feedback, and regularly summarize operation reports.

[0065] According to another aspect of the present invention, there is also provided a distributed dispatching system for elevator groups based on machine learning, the distributed dispatching system comprising: a data collection and monitoring module, a prediction and strategy planning module, and a dynamic strategy adjustment and performance monitoring module;

[0066] Data collection and monitoring module, which is used to monitor the flow of people and elevator load data in real time using pre-configured sensors on each floor of the building, and collect historical data on flow of people and elevator load;

[0067] The prediction and strategy planning module is used to predict the passenger flow within a preset time in the future based on the historical data of passenger flow and elevator load using the prophet prediction algorithm. Based on the prediction results and combined with the real-time monitored passenger flow data, a multi-objective optimization algorithm is used to balance passenger waiting time, elevator operation efficiency and energy consumption, and formulate elevator operation scheduling strategies;

[0068] The dynamic strategy adjustment and performance monitoring module is used to utilize real-time monitored passenger flow data and elevator load data, combined with floor height and predicted passenger flow data, to dynamically adjust the elevator's tiered services and optimize the elevator operation scheduling strategy, while also monitoring the performance indicators of the optimized elevator operation scheduling strategy in real time.

[0069] Compared with the prior art, this application has the following beneficial effects:

[0070] 1. The Prophet model in the present invention is used to process time series data with strong seasonal characteristics, such as elevator utilization, accurately predict trends and cyclical changes, and adjust and plan elevator operations. Through the Pareto optimal genetic algorithm, the optimal balance is found between multiple objectives, such as minimizing passenger waiting time and maximizing energy efficiency, thereby optimizing the overall performance of the elevator. By accurately predicting peak hours and adjusting elevator scheduling, the waiting time of passengers is reduced and the passenger utilization experience is improved. According to the prediction results and real-time data, the elevator scheduling strategy can be flexibly adjusted to better respond to emergencies and unconventional demands; through in-depth analysis of trends and cycles, managers can gain deep insights into elevator utilization patterns and support data-based decision-making. By combining the prediction of trends, cycles and random fluctuations, a comprehensive forecast view is provided to help managers understand and predict elevator load conditions more comprehensively. Statistical indicators such as the root mean square error are used to regularly evaluate the performance of the forecast model to ensure continuous optimization and adjustment of the forecast results.

[0071] 2. Through real-time monitoring and historical data analysis, the present invention can accurately identify peak hours and hot floors with dense traffic, thereby achieving rapid response and scheduling optimization of elevator services, significantly reducing passengers' waiting time, and improving overall service efficiency. Based on the identified data, resources can be allocated more reasonably. The intelligent optimization of elevator scheduling using genetic algorithms provides a high degree of adaptability and flexibility. The integrated data analysis and prediction model provides real-time data support for elevator scheduling, which makes decision-making more scientific and accurate. The prophet prediction algorithm can not only predict future passenger flow trends, but also identify periodic changes and assist in formulating more reasonable elevator scheduling strategies. By reducing waiting time and optimizing elevator operation strategies, the overall passenger experience is improved. Quick response and efficient service can improve passengers' overall satisfaction with building management. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The above characteristics, features and advantages of the present invention and their implementation methods and methods will become more clearly understood in conjunction with the following description of the embodiments, which are described in detail in conjunction with the accompanying drawings. Herein, a schematic diagram is shown:

[0073] Figure 1 is a flow chart of a distributed scheduling method for elevator groups based on machine learning according to an embodiment of the present invention;

[0074] Figure 2 It is a principle block diagram of a distributed scheduling system for elevator groups based on machine learning according to an embodiment of the present invention.

[0075] In the figure:

[0076] 1. Data collection and monitoring module; 2. Forecasting and strategy planning module; 3. Dynamic strategy adjustment and performance monitoring module. DETAILED DESCRIPTION

[0077] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0078] According to an embodiment of the present invention, a distributed scheduling method and system for elevator groups based on machine learning are provided.

[0079] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1As shown, according to one embodiment of the present invention, a distributed scheduling method for an elevator group based on machine learning is provided, and the distributed scheduling method for an elevator group includes the following steps:

[0080] S1. Use the pre-configured sensors on each floor of the building to monitor the passenger flow data and elevator load data in real time, and collect the historical data of passenger flow and elevator load.

[0081] It needs to be explained that sensors are pre-configured on each floor of the building to monitor the flow of people and the load of the elevator in real time. Flow sensors (such as infrared sensors, video cameras, etc.) are installed at the entrance and exit of the floor to count the number of people entering and leaving; load sensors (such as weight sensors) are installed inside the elevator to monitor the real-time load of the elevator. Infrared sensors count by detecting infrared radiation emitted by the human body, while weight sensors use pressure changes to measure weight. The collected data provides accurate real-time monitoring capabilities; real-time data provided by the sensors is collected, combined with historical data, and real-time data and historical data are integrated and analyzed to provide decision support for elevator scheduling.

[0082] S2. Based on the historical data of passenger flow and elevator load, the prophet prediction algorithm is used to predict the passenger flow within a preset time in the future. Based on the prediction results and combined with the real-time monitored passenger flow data, a multi-objective optimization algorithm is used to balance passenger waiting time, elevator operation efficiency and energy consumption, and an elevator operation scheduling strategy is formulated.

[0083] Among them, based on the historical data of passenger flow and elevator load, the prophet prediction algorithm is used to predict the passenger flow within the preset time in the future. Based on the prediction results, combined with the real-time monitored passenger flow data, a multi-objective optimization algorithm is used to balance passenger waiting time, elevator operation efficiency and energy consumption, and the elevator operation scheduling strategy is formulated, including the following steps:

[0084] S21. Obtain historical data of passenger flow and elevator load.

[0085] S22. Use the prophet prediction algorithm to perform time series analysis on historical data and extract trend items and cycle items respectively.

[0086] Preferably, using the prophet prediction algorithm to perform time series analysis on historical data, respectively extracting trend items and period items includes the following steps:

[0087] S221, constructing a prophet prediction model and configuring parameters of the prophet prediction model;

[0088] S222, inputting historical data into the prophet prediction model, and extracting trend items and seasonal cycle items;

[0089] S223, analyzing trend items and periodic items to identify long-term change trends and periodic fluctuation patterns of historical data;

[0090] S224. Evaluate the prediction accuracy of the prophet prediction model and the degree of fit between the trend and cyclical terms through statistical error indicators.

[0091] It should be explained that a Prophet model instance needs to be initialized first. Prophet is an open source time series forecasting library developed by Facebook. It can handle seasonal changes and holiday effects and is suitable for daily (or finer) time series data. In this step, set the parameters of the model, which may include seasonal configuration, holiday effect processing, and trend change flexibility, load the historical data set into the model and fit it. The dataset must contain two columns: ds (date stamp) and y (predictor variable); in this step, Prophet uses historical data to learn and extract trends and seasonal changes, and uses the fitted model to analyze and extract trends and cyclical changes. In this step, the trend and seasonal components can be visualized to identify key patterns and cyclical fluctuations in the data. The Prophet model decomposes the time series into three main components: trend, annual seasonality, and weekly seasonality, making the analysis more intuitive and the understanding deeper; to evaluate the prediction accuracy of the model, the root mean square error (RMSE) or other statistical indicators are usually used to measure the model's fit quality to historical data and its ability to predict future data. The root mean square error (RMSE) is a common method to measure prediction accuracy. It provides the square root mean of the difference between the predicted value and the actual value.

[0092] S23. Based on the extracted trend items and cycle items, the prophet prediction model is used to predict the trend and cycle changes of passenger flow within a preset time in the future.

[0093] Preferably, based on the extracted trend items and cycle items, using the prophet prediction model to predict the trend and cycle changes of the flow of people within a preset time in the future includes the following steps:

[0094] S231, setting a future preset time range;

[0095] S232, creating a data framework of future time points based on a future preset time range;

[0096] S233, input the future date data frame into the prophet prediction model, and use the predict method to make predictions;

[0097] S234. Extract the overall trend and periodic changes of passenger flow from the prediction results.

[0098] It should be explained that the future time period to be predicted should be determined, and the duration of the prediction should be determined according to the needs, such as the next week, month or year. The setting of the time range should take into account the periodic characteristics of the data and the analysis needs to ensure the practicality and accuracy of the prediction, generate a data frame for future dates, and prepare input data for the prediction. At the same time, the make_future_dataframe method of the Prophet model is used to generate a future date data frame according to the set time range, pass the future date data frame as input to the Prophet model, and use the model's predict method for prediction. The predict method will calculate the predicted value of the future date based on the previous training results of the model (learning of trend items and cycle items), analyze the prediction results, extract information about the trend and periodic fluctuations of future passenger flow, check the components of the Prophet prediction output, such as trends, annual cycles, weekly cycles, etc., and interpret how these components affect the predicted passenger flow. The prediction results usually include several parts, such as trend lines, seasonal changes, etc., to help managers understand future traffic changes and adjust their operating strategies accordingly.

[0099] Inputting the future date data frame into the prophet prediction model, using the predict method to make predictions includes the following steps:

[0100] When making predictions, the make_future_dataframe method of the Prophet model is first used to generate a data frame containing future dates according to the set future time range; then, this data frame is passed as input to the trained Prophet model, and the model's predict method is used to calculate and output the prediction results for each future time point, including trends and cyclical changes; this process integrates the model's learning results of historical data to generate predictions for future states.

[0101] S24. Collect real-time passenger flow and elevator load data, and adjust the prediction results of the prophet prediction model according to the real-time data.

[0102] It needs to be explained that the sensor system is deployed to continuously monitor the flow of people on each floor and the load of the elevator, and the data is transmitted to the central monitoring system in real time. The collected real-time data is formatted and preprocessed, such as denoising and normalization, to make it suitable for model input. The real-time data is input into the prophet model, and the model parameters are adjusted or incremental learning is performed using the new input to refine the model's prediction of future conditions. The adjusted model is used for a new round of predictions to generate updated prediction results, which will more accurately reflect the real-time and future operating status.

[0103] S25. Extract random fluctuation terms from the verification sequence, use the random fluctuation terms to train the extreme gradient boosting model, and predict random fluctuations in future time periods.

[0104] Preferably, extracting random fluctuation terms from the verification sequence, using the random fluctuation terms to train the extreme gradient boosting model, and predicting random fluctuations in a future time period comprises the following steps:

[0105] S251, selecting a period of passenger flow and elevator load data from actual applications as a verification sequence;

[0106] S252, using the prophet prediction model to predict the verification sequence, obtain the predicted values ​​of the trend term and the cycle term, and subtract the predicted values ​​of the trend term and the cycle term from the actual observed value to obtain the estimated value of the random fluctuation term;

[0107] S253, taking the random fluctuation term as the target variable, and preparing the feature variable for training the extreme gradient boosting model;

[0108] S254, using the random fluctuation term and the characteristic variable as inputs of the extreme gradient boosting model, training the extreme gradient boosting model at the same time, and verifying the trained extreme gradient boosting model;

[0109] S255. Use the verified extreme gradient boosting model to predict random fluctuations in future time periods.

[0110] Preferably, the expression of the estimated value of the random fluctuation term is:

[0111] ;

[0112] In the formula, represents the random fluctuation term at time point t;

[0113] represents the actual observation value at time point t;

[0114] represents the trend term at time point t;

[0115] represents the periodic term at time point t.

[0116] It should be explained that the actual passenger flow and elevator load data within a specific time period are extracted from the central database. The selection of the verification sequence should represent typical or special operating conditions so that the performance of the model can be tested in different scenarios. The Prophet model is used to predict the trend and periodic changes in the verification sequence, and then these predicted values ​​are subtracted from the actual data. The remaining part is the random fluctuation term. Unpredictable random fluctuations are separated by known trend and periodic adjustments. These fluctuations may be caused by irregular events or abnormal situations. Features related to random fluctuations (such as time points, holidays, special events, etc.) are selected and paired with random fluctuation terms as model inputs. Feature selection is crucial to the performance of the model. Appropriate features can help the model better learn and predict random fluctuations. The extreme gradient boosting (XGBoost) algorithm is used for model training, and cross-validation and other technologies are used to ensure the generalization ability and accuracy of the model. Extreme gradient boosting is an efficient integrated learning technology that improves the accuracy of predictions by optimizing a collection of decision trees. It is particularly suitable for complex nonlinear relationships. The trained and verified XGBoost model is used to predict future random fluctuations based on the latest feature inputs.

[0117] S26. Combine the trend term and period term predicted by the prophet prediction model with the random fluctuation term predicted by the extreme gradient boosting model to form a complete prediction result of passenger flow and elevator load.

[0118] Preferably, combining the trend term and the period term predicted by the prophet prediction model with the random fluctuation term predicted by the extreme gradient boosting model to form a complete prediction result of passenger flow and elevator load includes the following steps:

[0119] S261, extracting the forecast values ​​of the trend item and the cycle item from the forecast results of the prophet forecast model;

[0120] S262. Use the trained extreme gradient boosting model to predict random fluctuations in the same future time period;

[0121] S263, adding the prediction results of the trend item, the period item and the random fluctuation item to calculate the complete predicted values ​​of the passenger flow and the elevator load;

[0122] S264. Compare the calculated predicted value with the actual value, and use the root mean square error evaluation index to quantify the accuracy of the prediction.

[0123] Preferably, the expression for quantifying the accuracy of the prediction using the root mean square error evaluation index is:

[0124] ;

[0125] In the formula, represents the root mean square error;

[0126] n represents the total number of data points utilized when evaluating the prediction accuracy;

[0127] Fi represents the value of passenger flow or elevator load predicted by the model at the i-th time point;

[0128] Ti represents the actual passenger flow or elevator load value at the i-th time point.

[0129] It should be explained that the output of the Prophet model is analyzed to extract the forecast data of the trend and periodic components at each future time point. The trend term represents the long-term change direction, while the period term describes the fluctuation within a fixed period such as a week or a year. The trained extreme gradient boosting model is used to predict the random fluctuations at future time points based on the current and historical feature data. The random fluctuation term reflects the fluctuations caused by irregular factors that the forecasting model fails to capture. The trend and periodic term forecast values ​​of the Prophet model are added to the random fluctuation forecast values ​​of the extreme gradient boosting model to obtain the final forecast result. In this way, the long-term forecasting ability and responsiveness of the model to random events can be integrated to improve the accuracy and practicality of the forecast. The actual passenger flow and elevator load data at the forecast time point are collected and compared with the forecast results of the model, and statistical indicators such as the root mean square error (RMSE) are calculated. The root mean square error (RMSE) is a commonly used indicator to measure the deviation between the forecast value and the actual value. The smaller the value, the higher the accuracy of the forecast model.

[0130] S27. Define optimization objectives, apply multi-objective optimization algorithms, and formulate elevator operation scheduling strategies based on the complete passenger flow and elevator load prediction results.

[0131] Preferably, defining the optimization goal, applying the multi-objective optimization algorithm, and formulating the elevator operation scheduling strategy based on the complete prediction results of passenger flow and elevator load include the following steps:

[0132] S271. Define optimization goals based on the key performance indicators of elevator system operation;

[0133] S272. According to the multi-objective characteristics of the elevator dispatching problem, a Pareto optimal genetic algorithm suitable for the multi-objective problem is selected;

[0134] S273. Based on the Pareto optimal genetic algorithm, a mathematical model of elevator dispatching is established, and decision variables, constraints and objective functions are defined, and parameters of the Pareto optimal genetic algorithm are set;

[0135] S274, inputting the prediction results of passenger flow and elevator load into a mathematical model to predict the expected working condition of the elevator system in a future period;

[0136] S275. Run the Pareto optimal genetic algorithm using the set parameters, search for the Pareto optimal solution set through an iterative process, and select the final elevator scheduling strategy from the Pareto optimal solution set.

[0137] It needs to be explained that key performance indicators such as minimizing passenger waiting time and maximizing elevator utilization efficiency are selected as optimization goals; the definition of optimization goals should be based on actual operational needs and the characteristics of the elevator system to ensure the practicality and effectiveness of the goals. The Pareto genetic algorithm is selected because it can effectively handle multiple conflicting goals and find a set of optimal solutions rather than a single solution. The Pareto genetic algorithm optimizes problems by simulating the mechanisms of natural selection and inheritance. It is suitable for scenarios that need to meet multiple optimization goals at the same time. It defines decision variables (such as elevator selection, start time, etc.), constraints (such as elevator capacity, running time, etc.), and objective functions (such as minimizing total waiting time). At the same time, the algorithm parameters are set, such as population size, crossover rate and mutation rate. The elevator scheduling model needs to accurately describe the dynamics and complexity of elevator operation. At the same time, the setting of algorithm parameters has a key impact on the performance of the algorithm and the quality of the solution. The prediction results of trends, cycles and random fluctuations are integrated as model input to simulate future changes in elevator demand. The prediction results are used for decision support in practical applications, which can significantly improve the response speed and accuracy of the scheduling strategy. The algorithm is run, and through multiple generations of selection, crossover and mutation, a solution set that can balance multiple objectives is gradually found. The most suitable strategy for practical application is selected from these solutions. The Pareto optimal solution set represents the best trade-off between various objectives. The feasibility and efficiency of actual operation should be considered when selecting the final strategy.

[0138] When running the Pareto optimal genetic algorithm, first initialize a population in which individuals represent different strategies for elevator scheduling; through an iterative process, the algorithm continuously updates the population using selection, crossover, and mutation operations to find a solution set that simultaneously meets multiple optimization objectives; in each iteration, evaluate the fitness of the individuals, and select solutions based on the Pareto optimal principle, gradually building an optimal solution set that includes multiple optimization objective trade-offs; finally, select the final elevator scheduling strategy from this solution set to ensure that it achieves the best balance between efficiency and satisfaction in actual application.

[0139] S3. Utilize the real-time monitored passenger flow data and elevator load data, combined with the floor height and predicted passenger flow data, to dynamically adjust the elevator's tiered services and optimize the elevator operation scheduling strategy, while also monitoring the performance indicators of the optimized elevator operation scheduling strategy in real time.

[0140] Preferably, using the real-time monitored passenger flow data and elevator load data, combined with the floor height and predicted passenger flow data, dynamically adjusting the elevator's stratified service, and optimizing the elevator operation scheduling strategy, while real-time monitoring the performance indicators of the optimized elevator operation scheduling strategy includes the following steps:

[0141] S31. Collect the load data in the elevator and the flow of people on each floor in real time, combine historical patterns and prediction models, estimate future utilization trends, and build a comprehensive operational data framework;

[0142] S32. In the operational data framework, analyze the collected data to identify peak hours and hot spots with high traffic volume;

[0143] S33. Based on the recognition results, dynamically adjust the elevator's hierarchical service strategy and use genetic algorithms to perform intelligent optimization of elevator scheduling;

[0144] S34. Implement real-time monitoring mechanism, evaluate the operation performance of elevators and collect passenger feedback;

[0145] S35. Dynamically adjust the elevator dispatch strategy based on real-time monitoring data and passenger feedback, and regularly summarize operation reports.

[0146] It should be explained that sensors and monitoring systems are used to continuously collect elevator load and floor passenger flow data; historical data and prediction models are combined to analyze and estimate future utilization trends; data analysis techniques, such as cluster analysis or time series analysis, are used to identify passenger flow peaks and hot floors. This information will directly affect elevator scheduling decisions. By identifying hot areas and peak hours, resources can be allocated more effectively and elevator service efficiency can be optimized; according to the results of data analysis, the tiered service strategy of elevators in specific time periods is adjusted; genetic algorithms are used to optimize elevator scheduling, and the optimal scheduling strategy is found by simulating the evolutionary process. Genetic algorithms can effectively solve complex optimization problems by simulating the natural selection and genetic mechanisms of biological evolution, and are particularly suitable for scheduling problems in dynamic environments. A real-time monitoring system is set up to track the operation of the elevator, and passengers' opinions and suggestions are collected through questionnaires, application feedback, etc. Real-time monitoring and passenger feedback are the key to continuously improving service quality, which can help managers understand problems in elevator operation and make timely adjustments. Based on monitoring data and passenger feedback, analysis is conducted, and scheduling strategies are regularly adjusted to improve efficiency and passenger satisfaction. Operation reports are regularly generated to summarize experience and adjustment measures.

[0147] In the operational data framework, the specific steps to analyze the collected data and identify peak hours and hot spots with high traffic volume are as follows:

[0148] First, ensure the completeness and accuracy of the collection of real-time and historical traffic and elevator load data; this usually involves sensors and monitoring systems on each floor to capture the number of people entering and exiting and elevator utilization;

[0149] Clean and format the collected data, handle any errors or missing values, and standardize the data format for analysis; this may include steps such as noise removal, data interpolation, and outlier handling;

[0150] Use statistical analysis and time series analysis methods, such as moving average or exponential smoothing, to identify key trends in the data; this helps understand long-term patterns in passenger flow and elevator loads;

[0151] By analyzing the periodic changes in time series data, tools such as Fourier transform can be used to identify peak hours of traffic. Cluster analysis can be used to group time periods and identify time periods with abnormally high utilization frequencies.

[0152] Use cluster analysis, heat maps or geospatial analysis to identify floors with high traffic volume; these analyses will be based on floor entry and exit traffic data;

[0153] Visualize analysis results using tools such as charts and heat maps to more intuitively understand data patterns and identify hot spots; these visualization tools can help decision makers quickly grasp key information;

[0154] Based on the analysis results, specific adjustment suggestions are put forward, such as increasing the elevator service frequency during specific time periods, optimizing the elevator allocation strategy, or adjusting the elevator stop floor strategy.

[0155] Based on the recognition results, the specific steps of dynamically adjusting the elevator's hierarchical service strategy and using genetic algorithms to perform intelligent optimization of elevator scheduling are as follows:

[0156] First, based on the peak hours and crowded floors obtained through real-time and historical data analysis, these data points indicate the hot spots and times of elevator service demand;

[0157] Dynamically adjust the elevator's tiered service strategy based on the identified hot spots and floors; for example, if a floor often experiences high traffic during a specific period, the elevator's stop frequency for that floor can be increased during that period, or more elevator resources can be allocated to that floor first;

[0158] Build a genetic algorithm model to optimize elevator scheduling; define decision variables in the model (e.g., the route and floors of each elevator), as well as the objective function (e.g., minimizing total waiting time and maximizing elevator utilization);

[0159] Set the relevant parameters of the genetic algorithm, including population size, crossover rate, mutation rate, etc. These parameters will affect the efficiency and effect of the algorithm in finding the optimal solution;

[0160] By initializing the population, the elevator scheduling strategy is optimized using genetic operations such as selection, crossover, and mutation. Each individual in the population of each generation represents a possible elevator scheduling solution, and the optimal elevator operation strategy is found through iterative improvements.

[0161] The optimal elevator dispatching strategy obtained by the genetic algorithm is implemented in actual operation, and the effectiveness of the strategy is evaluated through real-time monitoring and user feedback to ensure maximum efficiency of elevator services and passenger satisfaction.

[0162] Specific examples are as follows:

[0163] Consider a large shopping mall located in a busy commercial district with multiple elevators and multiple entrances on different floors. This center has a large flow of people on weekends and holidays, so an efficient elevator dispatching system is needed to ensure smooth passenger flow.

[0164] 1) Initial setup:

[0165] Infrared sensors and weight sensors are installed at each floor entrance and inside the elevator to monitor the flow of people and elevator load in real time;

[0166] Real-time data together with historical data are stored in a central database to provide data support for analysis and prediction.

[0167] 2) Forecasting and scheduling strategy formulation:

[0168] Use the Prophet prediction algorithm to analyze historical passenger flow and elevator load data to predict passenger flow trends and cycle changes on specific days (such as weekends or holidays);

[0169] Combined with real-time monitoring data, the prediction model is adjusted to improve accuracy.

[0170] 3) Elevator dispatching strategy:

[0171] Adjust elevator service levels in advance during predicted peak hours, such as increasing the frequency of elevators to retail floors during peak shopping hours;

[0172] Apply multi-objective optimization algorithm (Pareto optimal genetic algorithm) to formulate scheduling strategies to balance passenger waiting time and elevator operation efficiency.

[0173] 4) Performance Monitoring:

[0174] Implement a real-time monitoring system to track elevator operation status and passenger satisfaction;

[0175] Collect passenger feedback, such as satisfaction surveys on elevator waiting times, to dynamically adjust scheduling strategies;

[0176] Regularly analyze elevator operation reports and adjust prediction models and scheduling strategies based on the latest data.

[0177] According to another embodiment of the present invention, Figure 2 As shown, a distributed dispatching system for elevator groups based on machine learning is also provided, the distributed dispatching system comprising: a data collection and monitoring module 1, a prediction and strategy planning module 2, and a dynamic strategy adjustment and performance monitoring module 3;

[0178] Data collection and monitoring module 1, used to monitor the pedestrian flow data and elevator load data in real time using pre-configured sensors on each floor of the building, and to collect historical data on pedestrian flow and elevator load;

[0179] Prediction and strategy planning module 2 is used to predict the passenger flow within a preset time in the future based on the historical data of passenger flow and elevator load using the prophet prediction algorithm. Based on the prediction results and combined with the real-time monitored passenger flow data, a multi-objective optimization algorithm is used to balance the passenger waiting time, elevator operation efficiency and energy consumption, and formulate an elevator operation scheduling strategy;

[0180] The dynamic strategy adjustment and performance monitoring module 3 is used to utilize the real-time monitored passenger flow data and elevator load data, combined with the floor height and predicted passenger flow data, to dynamically adjust the elevator's stratified services and optimize the elevator operation scheduling strategy, while also monitoring the performance indicators of the optimized elevator operation scheduling strategy in real time.

[0181] In summary, with the help of the above technical solutions of the present invention, the Prophet model in the present invention is used to process time series data with strong seasonal characteristics, such as elevator utilization, accurately predict trends and cyclical changes, and adjust and plan elevator operations. Through the Pareto optimal genetic algorithm, the optimal balance is found between multiple objectives, such as minimizing passenger waiting time and maximizing energy efficiency, thereby optimizing the overall performance of the elevator. By accurately predicting peak hours and adjusting elevator scheduling, the waiting time of passengers is reduced and the passenger utilization experience is improved. According to the prediction results and real-time data, the elevator scheduling strategy can be flexibly adjusted to better respond to emergencies and unconventional demands; through in-depth analysis of trends and cycles, managers can gain deep insights into elevator utilization patterns and support data-based decision-making. By combining the prediction of trends, cycles and random fluctuations, a comprehensive forecast view is provided to help managers understand and predict elevator load conditions more comprehensively. The performance of the prediction model is regularly evaluated using statistical indicators such as the root mean square error to ensure continuous optimization and adjustment of the prediction results. Through real-time monitoring and historical data analysis, the present invention can accurately identify peak hours and hot floors with dense traffic, thereby achieving rapid response and scheduling optimization of elevator services, significantly reducing passengers' waiting time, and improving overall service efficiency. Based on the identified data, resources can be allocated more reasonably. The intelligent optimization of elevator scheduling using genetic algorithms provides a high degree of adaptability and flexibility. The integrated data analysis and prediction model provides real-time data support for elevator scheduling, which makes decision-making more scientific and accurate. The prophet prediction algorithm can not only predict future passenger flow trends, but also identify periodic changes, and assist in formulating more reasonable elevator scheduling strategies. By reducing waiting time and optimizing elevator operation strategies, the overall passenger experience is improved. Quick response and efficient service can improve passengers' overall satisfaction with building management.

[0182] Although the present invention has been disclosed as above with preferred embodiments, the embodiments are merely examples for the convenience of description and are not intended to limit the present invention. Those skilled in the art may make several changes and modifications without departing from the spirit and scope of the present invention. The scope of protection claimed by the present invention shall be based on the claims.

Claims

1. A distributed scheduling method for elevator groups based on machine learning, characterized in that: The distributed scheduling method of the elevator group includes the following steps: S1. Use pre-configured sensors on each floor of the building to monitor the flow of people and elevator load data in real time, and collect historical data on flow of people and elevator load; S2. Based on the historical data of passenger flow and elevator load, the prophet prediction algorithm is used to predict the passenger flow within a preset time in the future. Based on the prediction results and combined with the real-time monitored passenger flow data, a multi-objective optimization algorithm is used to balance the passenger waiting time, elevator operation efficiency and energy consumption, and formulate an elevator operation scheduling strategy; S3. Using the real-time monitored passenger flow data and elevator load data, combined with the floor height and predicted passenger flow data, dynamically adjust the elevator's tiered services and optimize the elevator operation scheduling strategy, while monitoring the performance indicators of the optimized elevator operation scheduling strategy in real time; The method of using the prophet prediction algorithm to predict the passenger flow within a preset time in the future based on the historical data of passenger flow and elevator load, and using the multi-objective optimization algorithm to balance passenger waiting time, elevator operation efficiency and energy consumption based on the prediction results and combined with the real-time monitored passenger flow data, and formulating the elevator operation scheduling strategy includes the following steps: S21, obtaining historical data of passenger flow and elevator load; S22. Use the prophet prediction algorithm to perform time series analysis on historical data and extract trend items and cycle items respectively; S23, based on the extracted trend items and cycle items, using the prophet prediction model to predict the trend and cycle changes of passenger flow within a preset time in the future; S24, collecting real-time passenger flow and elevator load data, and adjusting the prediction results of the prophet prediction model according to the real-time data; S25. extract random fluctuation items from the verification sequence, use the random fluctuation items to train the extreme gradient boosting model, and predict random fluctuations in the future time period; S26. Combining the trend term and the period term predicted by the prophet prediction model with the random fluctuation term predicted by the extreme gradient boosting model to form a complete prediction result of passenger flow and elevator load; S27. Define optimization objectives, apply multi-objective optimization algorithms, and formulate elevator operation scheduling strategies based on the complete prediction results of passenger flow and elevator load; The method of extracting random fluctuation items from the verification sequence, using the random fluctuation items to train the extreme gradient boosting model, and predicting random fluctuations in a future time period includes the following steps: S251, selecting a period of passenger flow and elevator load data from actual applications as a verification sequence; S252, using the prophet prediction model to predict the verification sequence, obtain the predicted values ​​of the trend term and the cycle term, and subtract the predicted values ​​of the trend term and the cycle term from the actual observed value to obtain the estimated value of the random fluctuation term; S253, taking the random fluctuation term as the target variable, and preparing the feature variable for training the extreme gradient boosting model; S254, using the random fluctuation term and the characteristic variable as inputs of the extreme gradient boosting model, training the extreme gradient boosting model at the same time, and verifying the trained extreme gradient boosting model; S255. Use the verified extreme gradient boosting model to predict random fluctuations in the future time period; The expression of the estimated value of the random fluctuation term is: ； In the formula, Indicates at a point in time t The random fluctuation term of Indicates at a point in time t The actual observed value of Indicates at a point in time t Trend items; Indicates at a point in time t The periodic term of The method of combining the trend term and the period term predicted by the prophet prediction model with the random fluctuation term predicted by the extreme gradient boosting model to form a complete prediction result of passenger flow and elevator load includes the following steps: S261. Extracting the forecast values ​​of the trend item and the cycle item from the forecast results of the prophet forecast model; S262. Use the trained extreme gradient boosting model to predict random fluctuations in the same future time period; S263, adding the prediction results of the trend item, the period item and the random fluctuation item to calculate the complete predicted values ​​of the passenger flow and the elevator load; S264, comparing the calculated predicted value with the actual value, and quantifying the accuracy of the prediction using a root mean square error evaluation index; The expression for quantifying the accuracy of the prediction using the root mean square error evaluation index is: ； In the formula, represents the root mean square error; n Represents the total number of data points utilized when evaluating forecast accuracy; F i Indicates the passenger flow or elevator load predicted by the model in the i The value at a time point; T i Indicates the actual passenger flow or elevator load in the i The value at a time point; The method of utilizing the real-time monitored passenger flow data and elevator load data, combining the floor height and predicted passenger flow data, dynamically adjusting the elevator's stratified service, optimizing the elevator operation scheduling strategy, and simultaneously monitoring the performance indicators of the optimized elevator operation scheduling strategy in real time includes the following steps: S31. Collect the load data in the elevator and the flow of people on each floor in real time, combine historical patterns and prediction models, estimate future utilization trends, and build a comprehensive operational data framework; S32. In the operational data framework, analyze the collected data to identify peak hours and hot spots with high traffic volume; S33. Based on the recognition results, dynamically adjust the elevator's hierarchical service strategy and use genetic algorithms to perform intelligent optimization of elevator scheduling; S34. Implement real-time monitoring mechanism, evaluate the operation performance of elevators and collect passenger feedback; S35. Dynamically adjust the elevator dispatch strategy based on real-time monitoring data and passenger feedback, and regularly summarize operation reports.

2. The distributed scheduling method for elevator groups based on machine learning according to claim 1, characterized in that: The method of using the prophet prediction algorithm to perform time series analysis on historical data and extract trend items and period items respectively includes the following steps: S221, constructing a prophet prediction model and configuring parameters of the prophet prediction model; S222, inputting historical data into the prophet prediction model, and extracting trend items and seasonal cycle items; S223, analyzing trend items and periodic items to identify long-term change trends and periodic fluctuation patterns of historical data; S224. Evaluate the prediction accuracy of the prophet prediction model and the degree of fit between the trend and cyclical terms through statistical error indicators.

3. The distributed scheduling method for elevator groups based on machine learning according to claim 2, characterized in that: The method of using the prophet prediction model to predict the trend and periodic changes of the flow of people within a preset time in the future based on the extracted trend items and periodic items includes the following steps: S231, setting a future preset time range; S232, creating a data framework of future time points based on a future preset time range; S233, input the future date data frame into the prophet prediction model, and use the predict method to make predictions; S234. Extract the overall trend and periodic changes of passenger flow from the prediction results.

4. The distributed scheduling method for elevator groups based on machine learning according to claim 3, characterized in that: Defining the optimization goal, applying the multi-objective optimization algorithm, and formulating the elevator operation scheduling strategy based on the complete prediction results of passenger flow and elevator load include the following steps: S271. Define optimization goals based on the key performance indicators of elevator system operation; S272. According to the multi-objective characteristics of the elevator dispatching problem, a Pareto optimal genetic algorithm suitable for the multi-objective problem is selected; S273. Based on the Pareto optimal genetic algorithm, a mathematical model of elevator dispatching is established, and decision variables, constraints and objective functions are defined, and parameters of the Pareto optimal genetic algorithm are set; S274, inputting the prediction results of passenger flow and elevator load into a mathematical model to predict the expected working condition of the elevator system in a future period; S275. Run the Pareto optimal genetic algorithm using the set parameters, search for the Pareto optimal solution set through an iterative process, and select the final elevator scheduling strategy from the Pareto optimal solution set.

5. A distributed dispatching system for elevator groups based on machine learning, used to implement the distributed dispatching method for elevator groups based on machine learning according to any one of claims 1 to 4, characterized in that: The distributed scheduling system includes: a data collection and monitoring module, a prediction and strategy planning module, and a dynamic strategy adjustment and performance monitoring module; Data collection and monitoring module, which is used to monitor the flow of people and elevator load data in real time using pre-configured sensors on each floor of the building, and collect historical data on flow of people and elevator load; The prediction and strategy planning module is used to predict the passenger flow within a preset time in the future based on the historical data of passenger flow and elevator load using the prophet prediction algorithm. Based on the prediction results and combined with the real-time monitored passenger flow data, a multi-objective optimization algorithm is used to balance passenger waiting time, elevator operation efficiency and energy consumption, and formulate elevator operation scheduling strategies; The dynamic strategy adjustment and performance monitoring module is used to utilize real-time monitored passenger flow data and elevator load data, combined with floor height and predicted passenger flow data, to dynamically adjust the elevator's tiered services and optimize the elevator operation scheduling strategy, while also monitoring the performance indicators of the optimized elevator operation scheduling strategy in real time.

Images