CN116588776B - Elevator passenger flow prediction and optimal scheduling method and system - Google Patents
Elevator passenger flow prediction and optimal scheduling method and system Download PDFInfo
- Publication number
- CN116588776B CN116588776B CN202310882406.6A CN202310882406A CN116588776B CN 116588776 B CN116588776 B CN 116588776B CN 202310882406 A CN202310882406 A CN 202310882406A CN 116588776 B CN116588776 B CN 116588776B
- Authority
- CN
- China
- Prior art keywords
- elevator
- passenger flow
- passengers
- floor
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 71
- 238000012545 processing Methods 0.000 claims abstract description 39
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 230000001364 causal effect Effects 0.000 claims abstract description 20
- 238000004422 calculation algorithm Methods 0.000 claims description 26
- 230000007246 mechanism Effects 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000005457 optimization Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 5
- 238000013135 deep learning Methods 0.000 claims description 5
- 230000003068 static effect Effects 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 9
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 238000003062 neural network model Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000007621 cluster analysis Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000001174 ascending effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000005206 flow analysis Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B66—HOISTING; LIFTING; HAULING
- B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
- B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
- B66B5/0006—Monitoring devices or performance analysers
- B66B5/0012—Devices monitoring the users of the elevator system
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B66—HOISTING; LIFTING; HAULING
- B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
- B66B1/00—Control systems of elevators in general
- B66B1/02—Control systems without regulation, i.e. without retroactive action
- B66B1/06—Control systems without regulation, i.e. without retroactive action electric
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B66—HOISTING; LIFTING; HAULING
- B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
- B66B1/00—Control systems of elevators in general
- B66B1/34—Details, e.g. call counting devices, data transmission from car to control system, devices giving information to the control system
- B66B1/3415—Control system configuration and the data transmission or communication within the control system
- B66B1/3423—Control system configuration, i.e. lay-out
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B66—HOISTING; LIFTING; HAULING
- B66B—ELEVATORS; ESCALATORS OR MOVING WALKWAYS
- B66B5/00—Applications of checking, fault-correcting, or safety devices in elevators
- B66B5/0006—Monitoring devices or performance analysers
- B66B5/0018—Devices monitoring the operating condition of the elevator system
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
- Y02B50/00—Energy efficient technologies in elevators, escalators and moving walkways, e.g. energy saving or recuperation technologies
Landscapes
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Computer Networks & Wireless Communication (AREA)
- Indicating And Signalling Devices For Elevators (AREA)
- Elevator Control (AREA)
Abstract
The invention discloses a method and a system for predicting and optimally scheduling elevator passenger flow, comprising the following steps: acquiring the number of passengers in the elevator in real time; forming an elevator traffic network by a plurality of elevators which are similar in space in a building, wherein each building is a node, and constructing an elevator traffic flow; and analyzing the elevator traffic flow data on a space domain by using a transducer model, and processing the time domain information through a causal neural network to obtain the passenger flow distribution of the whole elevator network at a certain moment in the future. The elevators are dispatched by minimizing the sum of the times all passengers wait for the elevator and ride on the elevator. Information in elevator passenger flow data is extracted from two dimensions of space and time, elevator passenger flow distribution is effectively predicted, and finally optimal scheduling of the elevator is completed by minimizing macroscopic shortest elevator taking time of passengers.
Description
Technical Field
The invention belongs to the technical field of intelligent elevator dispatching, and relates to a method and a system for predicting and optimizing elevator passenger flow.
Background
How to scientifically and efficiently schedule elevators becomes an important problem in the development of elevator technology. Currently, although some scheduling methods have been tried and applied, these methods still have limitations such as low intelligence, and inability to automatically adjust scheduling policies according to environmental changes.
Traditional elevator dispatching technology is limited by factors such as insufficient hardware facilities, low software algorithm efficiency and the like, and real-time passenger flow prediction and dispatching optimization are difficult to perform. In the field of intelligent elevator dispatching, methods such as expert systems, fuzzy control, genetic algorithms and the like are applied to a certain degree and have certain achievements, but most of the methods are difficult to solve by using an accurate mathematical model, have poor flexibility and adaptability, and cannot ensure that dispatching can be rapidly adapted to the change of passenger flow conditions.
Application number 2023100005150 discloses a cooperative dispatching method of high-rise elevators based on the Internet of things, which comprises the steps of obtaining the number of people loaded at the current moment of each elevator in an elevator group, the corresponding target distance at the current moment of each elevator and the number of waiting people at the current moment of each floor; predicting the number of people waiting for ascending and the number of people waiting for descending at the current moment of each floor based on the monitoring video data of the elevator doorway every day in the preset historical days, constructing a feature vector corresponding to the current moment of each elevator and a feature vector corresponding to the current moment of the skyscraper, and further obtaining a state vector corresponding to the current moment; each elevator is controlled based on the state vector and the ES-reinforcement learning network. According to the method, the existing OpenPose model is utilized to detect key points of the head, the shoulders and the feet of a human body in images of elevator gates of all floors at the current moment respectively, the number of people waiting for an elevator at the elevator gates of all floors at the current moment is obtained, the statistical method is adopted to analyze and predict the traffic flow of the elevator, the calculated amount is large, and the prediction accuracy has an optimization space.
Disclosure of Invention
The invention aims to provide a method and a system for predicting and optimally dispatching elevator passenger flows, which are characterized in that an MCNN deep learning algorithm is used for identifying the number of elevator passengers, then an elevator passenger flow prediction neural network model based on a Transformer and a causal neural network is constructed, information in elevator passenger flow data is extracted from two dimensions of space and time, elevator passenger flow distribution is effectively predicted, and finally the optimal dispatching of the elevator is completed by minimizing the macroscopic shortest elevator taking time of passengers.
The technical solution for realizing the purpose of the invention is as follows:
an elevator passenger flow prediction method comprises the following steps:
s01: acquiring the number of passengers in the elevator in real time;
s02: forming an elevator traffic network by a plurality of elevators which are similar in space in a building, wherein each building is a node, and constructing an elevator traffic flow;
s03: and analyzing the elevator traffic flow data on a space domain by using a transducer model, and processing the time domain information through a causal neural network to obtain the passenger flow distribution of the whole elevator network at a certain moment in the future.
In a preferred technical solution, in the step S01, the MCNN deep learning algorithm is used to identify the number of passengers in the elevator, and the method includes:
MCNN algorithm is trained using a population density map, using an impact functionRepresenting the generated crowd density map：
wherein ,for the number of pixels +.>Index for the pixel;
using gaussian convolution versus population density mapProcessing is carried out, and the expansion parameters of the Gaussian kernel are determined through the average distance between a certain head and the heads around the head, so that a crowd density map is finally generated:
wherein ,is Gaussian kernel->Is Gaussian kernel parameter->For pixels +.>Average distance between the head and the surrounding heads, < ->Is a constant proportionality coefficient.
In a preferred embodiment, the method for constructing an elevator traffic flow in step S02 includes:
the method comprises the steps of abstracting elevators into traffic links with a plurality of nodes, wherein each floor is one node, and a plurality of elevators which are similar in space in one floor form an elevator traffic network:
wherein ,representing the set of links between adjacent nodes in an elevator, < >>For node set, ++>Is an adjacency matrix between nodes;
the floor number of one building is set asThe passenger flow distribution of one elevator in space is as follows:
wherein ,refers to a certain elevator, is->Indicating the floor at which the current is located, +.>Indicating that the current is in->First stop in a building>Total number of passengers waiting and riding in the elevator;
at a certain moment in a buildingElevator passenger flow distribution->The method comprises the following steps:
wherein ,indicating the number of elevators;
static distribution of elevator passenger flow distribution in time domain, expanding elevator traffic flow data in a period of time in time domain to obtain tensor of dynamic elevator passenger flow。
In a preferred embodiment, the method for analyzing the elevator traffic flow data in the spatial domain using the transducer model in the step S03 includes:
s31: setting the passenger flow of a node at a certain moment asThen->The relation with other nodes is:
wherein , and />Respectively representing the current floor and time, +.>For a certain period of time->For elevator traffic distribution->Predicting a relation for data to be established;
s32: analysis of elevator traffic flow data over a spatial domain using a modified transducer model, including byConvolution vs. data tensor->Performing dimension increasing and decreasing operations to complete cross-channel information fusion of data, and adding one +_ respectively in an Encoder and a Decode>Meanwhile, a Feed Forward layer in the Encoder is canceled, and the nonlinear characteristic is increased on the premise of keeping the scale of the feature map unchanged;
raw dataFirst go through->After convolution, the result is added with position code and sent into a transducer model, and the position code is +.>Is calculated as follows:
wherein , and />Respectively indicates a certain floor and total floor number, +.> and />For distinguishing even and odd dimensions and satisfying +.>，/>And->The result after addition is set to +.>Processing the data by using a multi-head self-attention mechanism;
after calculation using the attention mechanism, the network uses a residual structure to combine the result with the input dataAdding and normalizing, then using +.>Convolution is carried out to obtain an output result of the Encoder module;
the input data of the Decoder module comprises two parts, wherein one part is a sequence obtained by sequentially right shifting network input dataAnother part is the output of the Encoder module, < >>Processing is first performed using a mask matrix, the resulting result is processed using the same multi-headed attention mechanism as in the Encoder, the resulting result is added to the output of the Encoder and normalized, processing is again performed using the attention mechanism, and then processing is performed by +.>And after convolution, full-connection neural network and Softmax layer processing, obtaining a final output result.
In a preferred embodiment, the method for processing the time domain information in step S03 through the causal neural network includes:
partially removing the layer-by-layer links in the fully connected neural network in a mask processing mode, and reserving the front-to-back links;
processing data by using one-dimensional cavity convolution, and adding step spacing parameters in the moving process of convolution kernelStep size interval parameter->, wherein />Is an index of the network layer where it is located.
The invention also discloses an elevator optimal scheduling method, comprising the elevator passenger flow prediction method, wherein the step S03 further comprises the following steps:
the elevators are dispatched by minimizing the sum of the times all passengers wait for the elevator and ride on the elevator.
In a preferred embodiment, the method for minimizing the sum of the times of waiting and taking the elevator for all passengers comprises:
constructing a macroscopic shortest boarding time objective function for passengers：
wherein ,for the number of elevators>The floor number is 1 st floor, at +.>In a floor, waiting for an elevatorThe number of passengers is->The destination floor of each person is +.>, wherein />To wait for the index of elevator passengers, the number of passengers on the same destination floor is +>An elevator is provided with->The number of current passengers is +.>The total number of people that the elevator can accommodate is；
By minimizing the objective function。
The invention also discloses an elevator passenger flow prediction system, which comprises:
the passenger vision counting module is used for acquiring the real-time number of passengers in the elevator;
the elevator traffic flow construction module is used for constructing an elevator traffic network by forming a plurality of elevators which are similar in space in one building, wherein each building is a node, and the elevator traffic flow is constructed;
and the passenger flow prediction module is used for analyzing the elevator traffic flow data on a space domain by using a transducer model, and processing the time domain information through a causal neural network to obtain passenger flow distribution of the whole elevator network at a certain moment in the future.
The invention also discloses an elevator optimal scheduling system, which comprises the elevator passenger flow prediction system and further comprises:
the dispatch optimization module dispatches the elevator by minimizing the sum of the time for all passengers to wait for the elevator and ride on the elevator.
The invention also discloses a computer storage medium, on which a computer program is stored, which when executed implements the above-mentioned elevator passenger flow prediction method.
Compared with the prior art, the invention has the remarkable advantages that:
firstly, accurately identifying the number of elevator passengers by using an MCNN neural network algorithm, then constructing an elevator passenger flow prediction neural network model based on a transducer and a causal neural network, extracting information in elevator passenger flow data from two dimensions of space and time, and realizing accurate prediction of elevator passenger flow distribution. And the elevator passenger flow distribution is effectively predicted by fusing a multi-module network architecture. The minimum macroscopic shortest elevator taking time of passengers is used as an optimization target, the aim of elevator dispatching optimization is finally achieved, and the overall operation efficiency of the elevator is effectively improved.
Drawings
Fig. 1 is a flow chart of a preferred embodiment elevator passenger flow prediction method;
fig. 2 is a block diagram of an elevator optimization scheduling system;
fig. 3 is a diagram of the overall architecture of an intelligent elevator dispatching system;
fig. 4 is a schematic diagram of an elevator dynamic passenger flow tensor;
fig. 5 is a schematic diagram of an elevator traffic prediction network architecture;
FIG. 6 is a diagram of dataPerforming convolution treatment;
FIG. 7 is a schematic diagram of the construction of an Encoder and a Decoder;
FIG. 8 is a block diagram of a causal convolutional neural network;
FIG. 9 is a block diagram of a cavitation causal convolutional neural network;
FIG. 10 is a diagram of a support vector machine prediction error;
FIG. 11 is a graph of Kalman filtering algorithm prediction error;
FIG. 12 prediction error for an artificial immune algorithm;
FIG. 13 shows the prediction error of the algorithm of the present invention.
Detailed Description
The principle of the invention is as follows: firstly, an MCNN deep learning algorithm is used for identifying the number of elevator passengers, then an elevator passenger flow prediction neural network model based on a transducer and a causal neural network is constructed, information in elevator passenger flow data is extracted from two dimensions of space and time, elevator passenger flow distribution is effectively predicted, and finally, the optimal dispatching of the elevator is completed by minimizing the macroscopic shortest elevator taking time of passengers.
Example 1:
as shown in fig. 1, an elevator passenger flow prediction method comprises the following steps:
s01: acquiring the number of passengers in the elevator in real time;
s02: forming an elevator traffic network by a plurality of elevators which are similar in space in a building, wherein each building is a node, and constructing an elevator traffic flow;
s03: and analyzing the elevator traffic flow data on a space domain by using a transducer model, and processing the time domain information through a causal neural network to obtain the passenger flow distribution of the whole elevator network at a certain moment in the future.
In a preferred embodiment, the MCNN deep learning algorithm is used in step S01 to identify the number of passengers in the elevator, and the method includes:
MCNN algorithm is trained using a population density map, using an impact functionRepresenting the generated crowd density map：
wherein ,for the number of pixels +.>Index for the pixel;
using gaussian convolution versus population density mapProcessing is carried out, and the expansion parameters of the Gaussian kernel are determined through the average distance between a certain head and the heads around the head, so that a crowd density map is finally generated:
wherein ,is Gaussian kernel->Is Gaussian kernel parameter->For pixels +.>Average distance between the head and the surrounding heads, < ->Is a constant proportionality coefficient.
In a preferred embodiment, the method for constructing an elevator traffic flow in step S02 includes:
the method comprises the steps of abstracting elevators into traffic links with a plurality of nodes, wherein each floor is one node, and a plurality of elevators which are similar in space in one floor form an elevator traffic network:
wherein ,representing the set of links between adjacent nodes in an elevator, < >>For node set, ++>Is an adjacency matrix between nodes;
the floor number of one building is set asThe passenger flow distribution of one elevator in space is as follows:
wherein ,refers to a certain elevator, is->Indicating the floor at which the current is located, +.>Indicating that the current is in->First stop in a building>Total number of passengers waiting and riding in the elevator;
at a certain moment in a buildingElevator passenger flow distribution->The method comprises the following steps:
wherein ,indicating the number of elevators;
static distribution of elevator passenger flow distribution in time domain, expanding elevator traffic flow data in a period of time in time domain to obtain tensor of dynamic elevator passenger flow。
In a preferred embodiment, the method for analyzing the elevator traffic flow data in the spatial domain using the transducer model in step S03 includes:
s31: setting the passenger flow of a node at a certain moment asThen->The relation with other nodes is:
wherein , and />Respectively representing the current floor and time, +.>For a certain period of time->For elevator traffic distribution->Predicting a relation for data to be established;
s32: analysis of elevator traffic flow data over a spatial domain using a modified transducer model, including byConvolution vs. data tensor->Performing dimension increasing and decreasing operations to complete cross-channel information fusion of data, and adding one +_ respectively in an Encoder and a Decode>Meanwhile, a Feed Forward layer in the Encoder is canceled, and the nonlinear characteristic is increased on the premise of keeping the scale of the feature map unchanged;
original, originalDataFirst go through->After convolution, the result is added with position code and sent into a transducer model, and the position code is +.>Is calculated as follows:
wherein , and />Respectively indicates a certain floor and total floor number, +.> and />For distinguishing even and odd dimensions and satisfying +.>，/>And->The result after addition is set to +.>Processing the data by using a multi-head self-attention mechanism;
after calculation using the attention mechanism, the network uses a residual structure to combine the result with the input dataAdding and normalizing, then using +.>Convolution is carried out to obtain an output result of the Encoder module;
the input data of the Decoder module comprises two parts, wherein one part is a sequence obtained by sequentially right shifting network input dataAnother part is the output of the Encoder module, < >>Processing is first performed using a mask matrix, the resulting result is processed using the same multi-headed attention mechanism as in the Encoder, the resulting result is added to the output of the Encoder and normalized, processing is again performed using the attention mechanism, and then processing is performed by +.>And after convolution, full-connection neural network and Softmax layer processing, obtaining a final output result.
In a preferred embodiment, the method for processing the time domain information in step S03 through the causal neural network includes:
partially removing the layer-by-layer links in the fully connected neural network in a mask processing mode, and reserving the front-to-back links;
processing data by using one-dimensional cavity convolution, and adding step spacing parameters in the moving process of convolution kernelStep size interval parameter->, wherein />Is an index of the network layer where it is located.
In another embodiment, an elevator optimization scheduling method includes the elevator passenger flow prediction method, and after step S03, the method further includes:
the elevators are dispatched by minimizing the sum of the times all passengers wait for the elevator and ride on the elevator.
In a preferred embodiment, the method of minimizing the sum of all passengers waiting for and riding in the elevator comprises:
constructing a macroscopic shortest boarding time objective function for passengers：
wherein ,for the number of elevators>The floor number is 1 st floor, at +.>In a floor, waiting for an elevatorThe number of passengers is->The destination floor of each person is +.>, wherein />To wait for the index of elevator passengers, the number of passengers on the same destination floor is +>An elevator is provided with->The number of current passengers is +.>The total number of people that the elevator can accommodate is；
By minimizing the objective function。
In another embodiment, an elevator passenger flow prediction system comprises:
the passenger vision counting module is used for acquiring the real-time number of passengers in the elevator;
the elevator traffic flow construction module is used for constructing an elevator traffic network by forming a plurality of elevators which are similar in space in one building, wherein each building is a node, and the elevator traffic flow is constructed;
and the passenger flow prediction module is used for analyzing the elevator traffic flow data on a space domain by using a transducer model, and processing the time domain information through a causal neural network to obtain passenger flow distribution of the whole elevator network at a certain moment in the future.
In another embodiment, as shown in fig. 2, an elevator optimization scheduling system includes the elevator passenger flow prediction system, and further includes:
the dispatch optimization module dispatches the elevator by minimizing the sum of the time for all passengers to wait for the elevator and ride on the elevator.
In another embodiment, a computer storage medium has a computer program stored thereon, which when executed implements the elevator traffic prediction method described above.
Specifically, the following describes the working procedure of the elevator optimization scheduling system by taking a preferred embodiment as an example:
the overall architecture of the invention is shown in fig. 3, the current passenger number is firstly obtained through a visual sensor and a passenger real-time technology algorithm based on MCNN, then the elevator passenger distribution condition at a certain moment in the future is obtained through a prediction model based on a transducer and a causal convolutional neural network, and finally the intelligent optimal scheduling of the elevator is realized through the priority scheduling index of the passenger macroscopic shortest waiting time.
Step one: MCNN-based elevator passenger real-time counting algorithm.
The elevator passenger number statistics is divided into two parts of elevator inner passenger and elevator outer waiting passenger, passengers taking and waiting for the elevator are usually more dense in places such as a large mall or in periods such as going to and going from work, serious shielding situations can occur among people, and more accurate passenger numbers are difficult to obtain by adopting algorithms such as traditional wavelet transformation, edge detection, HOG feature extraction and the like. Therefore, on the basis of a Multi-column convolutional neural network (Multi-column Convolutional Neutral Network, MCNN) neural network algorithm, the method improves the elevator scene and realizes accurate statistics on the number of elevator passengers. The MCNN is a novel crowd counting convolutional neural network based on a density map, a high-quality density map is generated by adopting a self-adaptive Gaussian kernel, feature extraction is performed by using receptive field filters with different sizes, and accurate quantity statistical information can be obtained for dense crowds.
The MCNN algorithm is trained using a population density map if a pixel of the imageIf there is a human head, the label value of the corresponding place in the crowd density map is 1, so that the impact function can be used>To represent a crowd density map as shown in the following formula, wherein +.>Representing the generated crowd density map,/->For the number of pixels +.>Is like a figureAnd (5) indexing the element.
（11）
In order to makeThe method can continuously carry out Gaussian kernel convolution on the crowd density map by the MCNN, and meanwhile, the expansion parameters of the Gaussian kernel can be determined through the average distance between a certain head and the heads around the head, so that the finally generated crowd density map is shown as a formula (12), wherein +_>Is Gaussian kernel->Is Gaussian kernel parameter->For pixels +.>Average distance between the head and the surrounding heads, < ->Is a constant proportionality coefficient +.>The value is 0.3.
（12）
Although the MCNN network has better effect on crowd statistics, the MCNN network aims at dense crowds with larger range, the imaging sizes of the crowds can be greatly different when the crowds are located at a place close to and far from a camera, the crowds are usually limited to a smaller range inside and outside an elevator in an elevator scene, the imaging of passenger's head area is relatively larger, the size difference between the imaging is smaller, and a k-means clustering algorithm is usedAfter the super-cluster analysis, the inventionThe value is 0.42.
Step two: and constructing an elevator traffic flow analysis and prediction neural network model.
Elevator traffic flow data is time and space serialized data, but the time of an elevator run for one cycle is typically short, so its time series data is primarily short-term dependent in the time domain. The number of the nodes in the elevators is usually large, most of the elevators currently have tens of nodes (each layer is a node), and the spatial sequence data of the elevators have long-term dependency on a spatial domain. Most elevator dispatching algorithms at present consider elevator traffic flow data as a time sequence, and usually, SVM, markov chain, cluster analysis and other methods are used for predicting the elevator traffic flow, but the methods have the defects of low robustness, low accuracy and the like. With the development of technologies such as neural networks, a Recursive Neural Network (RNN) -based elevator traffic flow prediction method has achieved a good effect because it can effectively capture the time dependence between data. But most of these methods only consider the characteristics of the time dimension, while ignoring the characteristics of the space dimension of the elevator traffic flow. The method is characterized in that the method is used for modeling the sequence data based on the attention mechanism, and particularly has more excellent processing results on the data with long dependency relationship.
Step 21: representation of elevator traffic flow.
The elevators can be abstracted into traffic links with a plurality of nodes, each floor is one node, and a plurality of elevators which are similar in space in one floor form an elevator traffic network, and the elevator traffic networkCan be expressed as follows:
（13）
wherein Representing the set of links between adjacent nodes in an elevator, < >>For node set, ++>Is an adjacency matrix between nodes.
The floor number of one building is set asThe spatial passenger flow distribution of an elevator can then be described by the formula (14), wherein +.>Refers to a certain elevator, is->Indicating the floor at which the current is located, +.>Indicating that the current is in->First stop in a building>Total number of passengers waiting and riding in the elevator.
（14）
According to the formula (14), a certain moment in a building can be further obtainedElevator passenger flow distribution->Represented by formula (15), wherein>Indicating the number of elevators>Is an elevator traffic network represented by formula (13).
（15）
Equation (15) is a static distribution of elevator traffic distribution in time domain, and the tensor capable of dynamically reflecting elevator traffic is obtained by expanding elevator traffic data in time domainAs shown in FIG. 4, wherein +.>For a certain period of time.
Step 22: and constructing an elevator traffic flow prediction model.
The intelligent dispatching of the elevator is realized, and the passenger flow of a certain node in the elevator traffic network at a certain future moment needs to be predicted. Meanwhile, the passenger flow volume change of one node has a spatial and time dependency relationship with other nodes, so that the passenger flow volume of the current node can be predicted according to the passenger flow volume data of other nodes. Setting the passenger flow of a node at a certain moment asThen->The relation with the traffic of other nodes can be expressed by the formula (16) wherein +.> and />Respectively representing the current floor and time, +.>For a certain period of time->For the elevator passenger flow distribution as shown in formula (15)>Namely, the data prediction relation to be established.
（16）
The elevator traffic flow data has short dependency relationship in time domain and long dependency relationship in space domain, so the invention uses a transducer model to analyze the elevator traffic flow data in space domain, uses a causal convolutional neural network to analyze in time domain, and an elevator passenger flow prediction network overall architecture model is shown in figure 5, wherein the elevator passenger flow prediction network overall architecture model comprisesAn Encoder and Decoder Module,>different values may be set according to the specific circumstances.
The following improvements are made in the present invention over the conventional transducer model. First byConvolution vs. data tensor->Performing dimension increasing and reducing operations to complete cross-channel information fusion of data, wherein the dimension of the original data is defined byUnified as->As shown in fig. 6. Next, add a +/in each of the Encoder and Decoder>Meanwhile, a Feed Forward layer in the Encoder is canceled, and nonlinear characteristics are added on the premise of keeping the scale of the feature map unchanged, so that the depth of the network can be deepened, and a better feature extraction effect is realized. The construction of the Encoder and Decode is shown in FIG. 7.
Raw dataFirst go through->After convolution, the result is added with the position code and sent into a transducer model. Position coding->The calculation of (1) is shown in the formula, wherein +.> and />Respectively indicates a certain floor and total floor number, +.> and />For distinguishing even and odd dimensions and satisfying +.> 。
（17）/>
And->The result after addition is set to +.>It is processed using a multi-headed self-attention mechanism as shown in FIG. 7, wherein +.>、/>、/>All are linear transformation matrixes, and the values of the linear transformation matrixes are obtained by model training. The calculation of the multi-headed self-attentive mechanism is shown in formula (18), wherein +.>For the dimension of data, +.>Is->Transpose of->I.e. the calculation of the scaling dot product.
（18）
After calculation using the attention mechanism, the network uses a residual structure to combine the result with the input dataAdding and normalizing, then using +.>The convolution is processed to obtain the output result of the Encoder module.
For the Decoder module shown in fig. 7, the input data includes two parts, one part is a sequence obtained by sequentially right shifting the network input dataThe other part is the output of the Encoder module. />First, a mask matrix is used to process the sequence data to avoid the influence of the data at the back on the front data. The mask matrix is a lower triangle 0-1 matrix, e.g. a +.>The mask matrix of (2) is shown in equation (19). />After mask matrix processing, the resulting result is processed using the same multi-headed attention mechanism as in the Encoder, the resulting result is added to the output of the Encoder and normalized, again processed using the attention mechanism, and then passed->And after convolution, full-connection neural network and Softmax layer processing, obtaining a final output result.
（19）
The method processes the spatial domain information in the data through a transducer technology, and then processes the time domain information through a causal neural network. The causal neural network is formed by partially removing the layer-by-layer links in the fully connected neural network in a mask processing mode, and only the front-to-back links are reserved, so that the front-to-back dependency relationship of data in time is met, as shown in fig. 8. But causal convolutionThe receptive field of the nodes in the network is small, so that the network is too deep, and the calculated amount is increased. To increase receptive field, the present invention uses one-dimensional hole convolution to process data. The cavity convolution is to add step spacing parameters in the moving process of the convolution kernelAs shown in fig. 9, comparing with fig. 8, it can be seen that the receptive field of the node is greatly improved after the cavity convolution is used for the network with the same depth. Step size interval parameter->, wherein />Is an index of the network layer where it is located.
Through the data processing, the passenger flow data of a certain node of the elevator can be predicted, and further the passenger flow distribution of the whole elevator network at a certain moment in the future can be obtained, so that the purposes of analyzing and predicting the passenger flow of the elevator are achieved.
Step three: the passenger macroscopic shortest boarding time is scheduled preferentially.
The macroscopic minimum elevator taking time of the passengers refers to the sum of the time for all the passengers to wait for the elevator and the time for the passengers to take the elevator, and the aim of optimizing elevator dispatching is achieved by minimizing the macroscopic minimum waiting time of the passengers, so that the elevator dispatching is realized in a wider range and high efficiency and unification are realized. The number of the elevators isThe number of floors is->The bottom layer of the elevator is the 1 st layer, in the +.>In a floor, waiting for an elevator->The number of passengers is->The destination floor of each person is +.>, wherein />To wait for the index of elevator passengers, the number of passengers on the same destination floor is +>. Elevator with->The number of current passengers is +.>The total number of people that the elevator can accommodate is +.>Then constructing the macroscopic shortest boarding time objective function of the passenger as shown in formula (20)>By minimizing +.>The elevator is scheduled and regulated, and the aim of improving the elevator scheduling efficiency is fulfilled.
（20）
In order to verify the effectiveness of the invention, firstly, elevator simulation software Elevate is used for generating simulation data, wherein the number of elevators is 4, the number of floors is 8, a time period between 8 am and 8 pm is selected as test data generation, the distribution data of elevator passengers are recorded every 5 minutes, the average value of the distribution errors of each layer of elevator passengers is used as a measurement standard, the average value of the distribution errors of each layer of elevator passengers is compared with a support vector machine, kalman filtering and artificial immunity algorithm which are commonly used for predicting the current elevator passenger flow distribution, the prediction errors of various algorithms are shown as fig. 10-13, the prediction of the invention on the elevator passenger flow is more accurate, the prediction result is more stable, and the data comparison of the various algorithms on the passenger flow prediction can be seen.
Meanwhile, for different time periods of each day, corresponding elevator passenger flow simulation data are generated, and compared with elevator scheduling algorithms such as a scanning algorithm, a SCAN-EDF algorithm and the like which are used more currently, the results are shown in Table 2:
from this it can be seen that the invention enables a more efficient scheduling of elevators.
The foregoing examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the foregoing examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principles of the present invention should be made therein and are intended to be equivalent substitutes within the scope of the present invention.
Claims (9)
1. The elevator passenger flow prediction method is characterized by comprising the following steps of:
s01: acquiring the number of passengers in the elevator in real time;
s02: forming an elevator traffic network by a plurality of elevators which are similar in space in a building, wherein each building is a node, and constructing an elevator traffic flow; the method for constructing the elevator traffic flow comprises the following steps:
the method comprises the steps of abstracting elevators into traffic links with a plurality of nodes, wherein each floor is one node, and a plurality of elevators which are similar in space in one floor form an elevator traffic network:
，
wherein ,representing the set of links between adjacent nodes in an elevator, < >>For node set, ++>Is an adjacency matrix between nodes;
the floor number of one building is set asThe passenger flow distribution of one elevator in space is as follows:
，
wherein ,refers to a certain elevator, is->Indicating the floor at which the current is located, +.>Indicating that the current is in->First stop in a building>Total number of passengers waiting and riding in the elevator;
at a certain moment in a buildingElevator passenger flow distribution->The method comprises the following steps:
，
wherein ,indicating the number of elevators;
static distribution of elevator passenger flow distribution in time domain, expanding elevator traffic flow data in a period of time in time domain to obtain tensor of dynamic elevator passenger flow；
S03: and analyzing the elevator traffic flow data on a space domain by using a transducer model, and processing the time domain information through a causal neural network to obtain the passenger flow distribution of the whole elevator network at a certain moment in the future.
2. The elevator passenger flow prediction method according to claim 1, wherein the step S01 uses MCNN deep learning algorithm to identify the number of elevator passengers, and the method comprises:
MCNN algorithm is trained using a population density map, using an impact functionRepresenting the generated crowd density map->：
，
wherein ,for the number of pixels +.>Index for the pixel;
using gaussian convolution versus population density mapProcessing is carried out, and the expansion parameters of the Gaussian kernel are determined through the average distance between a certain head and the heads around the head, so that a crowd density map is finally generated:
，
wherein ,is Gaussian kernel->Is Gaussian kernel parameter->For pixels +.>Average distance between the head and the surrounding heads, < ->Is a constant proportionality coefficient.
3. The method for predicting elevator traffic as defined in claim 1, wherein the method for analyzing the elevator traffic data in the spatial domain using the transducer model in step S03 comprises:
s31: setting the passenger flow of a node at a certain moment asThen->The relation with other nodes is:
，
wherein , and />Respectively representing the current floor and time, +.>For a certain period of time->For elevator traffic distribution->Predicting a relation for data to be established;
s32: analysis of elevator traffic flow data over a spatial domain using a modified transducer model, including byConvolution vs. data tensor->Performing dimension increasing and decreasing operations to complete cross-channel information fusion of data, and adding one +_ respectively in an Encoder and a Decode>Meanwhile, a Feed Forward layer in the Encoder is canceled, and the nonlinear characteristic is increased on the premise of keeping the scale of the feature map unchanged;
raw dataFirst go through->After convolution, the result is added with position code and sent into a transducer model, and the position codeIs calculated as follows:
，
wherein , and />Respectively indicates a certain floor and total floor number, +.> and />For distinguishing even and odd dimensions and satisfying +.>，/>And->The result after addition is set to +.>Processing the data by using a multi-head self-attention mechanism;
after calculation using the attention mechanism, network miningThe residual structure is used for combining the result with the input dataAdding and normalizing, then using +.>Convolution is carried out to obtain an output result of the Encoder module;
the input data of the Decoder module comprises two parts, wherein one part is a sequence obtained by sequentially right shifting network input dataAnother part is the output of the Encoder module, < >>Processing is first performed using a mask matrix, the resulting result is processed using the same multi-headed attention mechanism as in the Encoder, the resulting result is added to the output of the Encoder and normalized, processing is again performed using the attention mechanism, and then processing is performed by +.>And after convolution, full-connection neural network and Softmax layer processing, obtaining a final output result.
4. The elevator passenger flow prediction method according to claim 1, wherein the method for processing the time domain information through the causal neural network in step S03 comprises:
partially removing the layer-by-layer links in the fully connected neural network in a mask processing mode, and reserving the front-to-back links;
processing data by using one-dimensional cavity convolution, and adding step spacing parameters in the moving process of convolution kernelStep size interval parameter->, wherein />Is an index of the network layer where it is located.
5. An optimized elevator dispatching method, characterized by comprising the elevator passenger flow prediction method according to any one of claims 1-4, and further comprising after step S03:
the elevators are dispatched by minimizing the sum of the times all passengers wait for the elevator and ride on the elevator.
6. The optimal scheduling method for an elevator according to claim 5, wherein the method of minimizing the sum of times for all passengers to wait for an elevator and ride an elevator comprises:
constructing a macroscopic shortest boarding time objective function for passengers：
，
wherein ,for the number of elevators>The floor number is 1 st floor, at +.>In a floor, waiting for an elevator->The number of passengers is->The destination floor of each person is +.>, wherein />To wait for the index of elevator passengers, the number of passengers on the same destination floor is +>An elevator is provided with->The number of current passengers is +.>The total number of people that the elevator can accommodate is +.>；
By minimizing the objective function。
7. An elevator traffic prediction system, comprising:
the passenger vision counting module is used for acquiring the real-time number of passengers in the elevator;
the elevator traffic flow construction module is used for constructing an elevator traffic network by forming a plurality of elevators which are similar in space in one building, wherein each building is a node, and the elevator traffic flow is constructed; the method for constructing the elevator traffic flow comprises the following steps:
the method comprises the steps of abstracting elevators into traffic links with a plurality of nodes, wherein each floor is one node, and a plurality of elevators which are similar in space in one floor form an elevator traffic network:
，
wherein ,representing the set of links between adjacent nodes in an elevator, < >>For node set, ++>Is an adjacency matrix between nodes;
the floor number of one building is set asThe passenger flow distribution of one elevator in space is as follows:
，
wherein ,refers to a certain elevator, is->Indicating the floor at which the current is located, +.>Indicating that the current is in->First stop in a building>Total number of passengers waiting and riding in the elevator;
at a certain moment in a buildingElevator passenger flow distribution->The method comprises the following steps:
，
wherein ,indicating the number of elevators;
static distribution of elevator passenger flow distribution in time domain, expanding elevator traffic flow data in a period of time in time domain to obtain tensor of dynamic elevator passenger flow；
And the passenger flow prediction module is used for analyzing the elevator traffic flow data on a space domain by using a transducer model, and processing the time domain information through a causal neural network to obtain passenger flow distribution of the whole elevator network at a certain moment in the future.
8. An elevator optimization scheduling system comprising the elevator traffic prediction system of claim 7, further comprising:
the dispatch optimization module dispatches the elevator by minimizing the sum of the time for all passengers to wait for the elevator and ride on the elevator.
9. A computer storage medium on which a computer program is stored, characterized in that the computer program, when executed, implements the elevator passenger flow prediction method of any one of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310882406.6A CN116588776B (en) | 2023-07-19 | 2023-07-19 | Elevator passenger flow prediction and optimal scheduling method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310882406.6A CN116588776B (en) | 2023-07-19 | 2023-07-19 | Elevator passenger flow prediction and optimal scheduling method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116588776A CN116588776A (en) | 2023-08-15 |
| CN116588776B true CN116588776B (en) | 2023-11-03 |
Family
ID=87606624
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310882406.6A Active CN116588776B (en) | 2023-07-19 | 2023-07-19 | Elevator passenger flow prediction and optimal scheduling method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116588776B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102556783A (en) * | 2011-07-12 | 2012-07-11 | 江苏镇安电力设备有限公司 | Subarea-based elevator traffic prediction group control method and elevator monitoring implementation |
| CN102765643A (en) * | 2012-05-31 | 2012-11-07 | 天津大学 | Elevator fault diagnosis and early-warning method based on data drive |
| CN111994748A (en) * | 2020-08-04 | 2020-11-27 | 广州广日电梯工业有限公司 | Method and system for simulating elevator passenger flow in peak period |
| CN112061908A (en) * | 2020-09-21 | 2020-12-11 | 深圳炳麟科技有限公司 | Elevator control method and system |
| CN114707587A (en) * | 2022-03-24 | 2022-07-05 | 电子科技大学中山学院 | Traffic pattern recognition method based on genetic algorithm and fuzzy neural network |
| CN115303901A (en) * | 2022-08-05 | 2022-11-08 | 北京航空航天大学 | Elevator traffic flow identification method based on computer vision |
-
2023
- 2023-07-19 CN CN202310882406.6A patent/CN116588776B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102556783A (en) * | 2011-07-12 | 2012-07-11 | 江苏镇安电力设备有限公司 | Subarea-based elevator traffic prediction group control method and elevator monitoring implementation |
| CN102765643A (en) * | 2012-05-31 | 2012-11-07 | 天津大学 | Elevator fault diagnosis and early-warning method based on data drive |
| CN111994748A (en) * | 2020-08-04 | 2020-11-27 | 广州广日电梯工业有限公司 | Method and system for simulating elevator passenger flow in peak period |
| CN112061908A (en) * | 2020-09-21 | 2020-12-11 | 深圳炳麟科技有限公司 | Elevator control method and system |
| CN114707587A (en) * | 2022-03-24 | 2022-07-05 | 电子科技大学中山学院 | Traffic pattern recognition method based on genetic algorithm and fuzzy neural network |
| CN115303901A (en) * | 2022-08-05 | 2022-11-08 | 北京航空航天大学 | Elevator traffic flow identification method based on computer vision |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116588776A (en) | 2023-08-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108133188B (en) | Behavior identification method based on motion history image and convolutional neural network | |
| CN112532439B (en) | Network flow prediction method based on attention multi-component space-time cross-domain neural network model | |
| Nguyen et al. | EO-CNN: an enhanced CNN model trained by equilibrium optimization for traffic transportation prediction | |
| CN113469356A (en) | Improved VGG16 network pig identity recognition method based on transfer learning | |
| CN107636691A (en) | Method and apparatus for identifying the text in image | |
| CN111476285B (en) | Training method of image classification model, image classification method and storage medium | |
| CN113852432B (en) | Spectrum Prediction Sensing Method Based on RCS-GRU Model | |
| CN116832588B (en) | Acid regeneration flue gas purifying device and method thereof | |
| CN116187555A (en) | Traffic flow prediction model construction method and prediction method based on self-adaptive dynamic diagram | |
| CN110675632B (en) | Vehicle short-time trajectory prediction control method aiming at multi-feature space and data sparseness | |
| CN115508073B (en) | Prototype adaptive mechanical equipment fault diagnosis method based on multi-scale attention | |
| CN111127360A (en) | Gray level image transfer learning method based on automatic encoder | |
| CN110210656A (en) | Shared bicycle method for predicting and system based on website behavioural analysis | |
| CN114898121A (en) | Concrete dam defect image description automatic generation method based on graph attention network | |
| CN115131618A (en) | Semi-supervised image classification method based on causal reasoning | |
| CN116894977A (en) | Remote sensing target detection method based on lightweight transducer | |
| Kaburlasos et al. | Intervals' numbers (INs) interpolation/extrapolation | |
| Terziyan et al. | Causality-aware convolutional neural networks for advanced image classification and generation | |
| CN114584406A (en) | Industrial big data privacy protection system and method for federated learning | |
| CN116588776B (en) | Elevator passenger flow prediction and optimal scheduling method and system | |
| CN117636183A (en) | Small sample remote sensing image classification method based on self-supervision pre-training | |
| CN112669186A (en) | Optimal intervention method and system for positive emotion and negative emotion in crowd evacuation | |
| CN116805162A (en) | Transformer model training method based on self-supervision learning | |
| CN109583584A (en) | The CNN with full articulamentum can be made to receive the method and system of indefinite shape input | |
| CN115331052A (en) | Garbage data labeling system and method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |