A dynamic partition energy-saving elevator dispatching method for group control peak period

Abstract

The invention relates to the technical field of elevator dispatching, and discloses a group control peak period dynamic partition energy-saving elevator dispatching method, comprising the following steps: step S101, collecting the operation data of M elevators and the flow of people on N floors at the current moment, and constructing graph structure data; step S102, constructing an elevator dispatching scheme generation model; step S103, generating a second adjacency matrix as a sample label of a training sample for training the elevator dispatching scheme generation model through a spotted kingfisher optimization algorithm; step S104, inputting the graph structure data into the trained elevator dispatching scheme generation model, and outputting a first adjacency matrix; step S105, decoding the first adjacency matrix to obtain an elevator dispatching scheme; the invention comprehensively considers the load and energy consumption of the elevator through the elevator dispatching scheme generation model combined with the spotted kingfisher optimization algorithm, and completes the dynamic partition of the elevator according to the flow of people on different floors in real time, thereby reducing the energy consumption of the elevator.

Description

Technical Field

[0001] The present invention relates to the technical field of elevator dispatching, and more specifically, to a group control peak period dynamic partition energy-saving elevator dispatching method. Background Art

[0002] In modern high-rise buildings, elevators are one of the key facilities connecting the floors, and their operating efficiency directly affects the flow efficiency of people in the building. Existing elevator scheduling methods usually adopt a static zoning method, that is, the service areas of the elevators are pre-set, and these service areas will not be automatically adjusted over time or with changes in the flow of people. For example, in an office building, elevators may be divided into high and low zones, with low-zone elevators only responsible for lower floors (such as 1-10 floors), and high-zone elevators responsible for higher floors (such as 11-20 floors).

[0003] Although the existing elevator dispatching method can reduce the up and down distance of the elevator to a certain extent and improve the flow efficiency of personnel, it still has the following defects:

[0004] 1. Static zoning cannot flexibly respond to changes in actual demand during different time periods. For example, during rush hour on weekdays, a large number of people flock from the lower floors to the higher floors. At this time, if the elevators still operate according to the fixed high and low zoning, it may cause the elevators in the lower floors to be empty, while the elevators in the higher floors will stop frequently due to excessive crowding, resulting in increased waiting time.

[0005] 2. The existing elevator dispatching method does not take into account the energy consumption of the elevator. The elevator will consume a lot of electricity when running in an unloaded state or frequently starting and stopping. In the static partitioning mode, the elevator service area cannot be automatically adjusted according to the flow of people in real time, resulting in unnecessary energy consumption. Summary of the invention

[0006] The present invention provides a group control peak period dynamic partition energy-saving elevator dispatching method to solve the technical problems in the above-mentioned background technology.

[0007] The present invention provides a group control peak period dynamic partition energy-saving elevator dispatching method, comprising the following steps:

[0008] Step S101, collecting the operation data of M elevators and the passenger flow of N floors at the current moment, and constructing graph structure data;

[0009] Graph structure data includes: nodes, initial vectors of nodes, and edges between nodes;

[0010] The nodes include: M first nodes and N second nodes;

[0011] The M first nodes respectively establish mapping relationships with the M elevators;

[0012] The N second nodes respectively establish mapping relationships with the N floors;

[0013] The initial vectors of the M first nodes are represented by the operation data of the elevator;

[0014] The initial vectors of the N second nodes are represented by the flow of people on the floor;

[0015] Step S102, constructing an elevator dispatching plan generation model;

[0016] The input of the elevator scheduling plan generation model is graph structure data;

[0017] The output of the elevator scheduling plan generation model is the first adjacency matrix;

[0018] The element value of the first adjacency matrix is ​​represented by 0 or 1. The element value of the mth column of the nth row of the first adjacency matrix is ​​1, which means that the mth elevator stops at the nth floor, otherwise it means that the mth elevator does not stop at the nth floor, where 1≤m≤M, 1≤n≤N;

[0019] Step S103, generating a second adjacency matrix by using the Pied Kingfisher optimization algorithm as sample labels of training samples for training an elevator scheduling solution generation model;

[0020] The representation of the second adjacency matrix is ​​the same as the representation of the first adjacency matrix;

[0021] Step S104, inputting the graph structure data into the trained elevator scheduling solution generation model, and outputting a first adjacency matrix;

[0022] Step S105: Decode the first adjacency matrix to obtain an elevator scheduling solution.

[0023] Furthermore, the operating data of the elevator includes: the floor where the elevator is located, the direction of elevator travel, the current load, the maximum load, the current number of passengers, the maximum number of passengers, the number of historical maintenance times, the number of historical accidents and unit energy consumption, among which the directions of elevator travel include: up, down and stay, which are represented by real number codes, the maximum load and the maximum number of passengers are set by the elevator manufacturer, and the unit energy consumption is represented by the power consumption of the elevator when it travels one floor in an empty state.

[0024] Furthermore, the edge between the nodes includes: the edge constructed between the first node and the second node indicates that the floor that establishes a mapping relationship with the second node generates a dispatch instruction for the elevator that establishes a mapping relationship with the first node.

[0025] Furthermore, the elevator dispatching scheme generation model includes:

[0026] First hidden layer, second hidden layer, third hidden layer;

[0027] The first hidden layer inputs the initial vector of the first node and outputs the first update vector;

[0028] The second hidden layer inputs the initial vector of the second node and outputs the second update vector;

[0029] The first update vector and the second update vector have the same number of dimensions;

[0030] That is, the initial vector of the node of the graph structure data is represented by the update vector;

[0031] The third hidden layer inputs the graph structure data and outputs the first adjacency matrix.

[0032] Furthermore, the calculation formula of the first hidden layer includes: ; ;

[0033] Where 1≤i≤H, H represents the number of dimensions of the initial vector of the first node, that is, H is equal to 9, represents the first update vector of the first hidden layer output, Represents the value of the i-th dimension of the initial vector of the first node, and Respectively represent the mean and standard deviation of the initial vector of the first node, It means to concatenate the values ​​of H dimensions, NormInit means the normalized vector, represents the first weight parameter, represents the first bias parameter, ReLU represents the ReLU activation function;

[0034] The calculation formula for the second hidden layer is as follows: ;

[0035] in represents the second update vector for the output of the second hidden layer, The initial vector of the second node representing the input of the second hidden layer, represents the second weight parameter, Represents the second bias parameter.

[0036] Furthermore, the calculation formula of the third hidden layer includes: ; ; ;

[0037] Where S represents the first adjacency matrix of the third hidden layer output. If the element value in S is greater than or equal to 0.5, it is assigned to 1, otherwise it is assigned to 0. represents the intermediate vector of the u-th node, and Represent the update vectors of the u-th node and the v-th node respectively, represents the set of nodes that build edges with the u-th node, , and represent the third weight parameter, the fourth weight parameter and the fifth weight parameter respectively, represents the attention coefficient between the uth node and the vth node, MLP represents a multi-layer perceptron, and pile represents a stacking operation. Indicates stacking the intermediate vectors of N second nodes. Indicates that the intermediate vectors of the M first nodes are stacked, T indicates a transpose operation, and sigmoid indicates a sigmoid activation function.

[0038] Further, generating a second adjacency matrix by using the Pied Kingfisher optimization algorithm comprises the following steps:

[0039] Step S201, randomly generate an initialization population consisting of K individuals that meet the constraints, and initialize the current number of iterations to 0;

[0040] Where K is a custom parameter;

[0041] The constraints include: the load capacity assigned to the elevator cannot be greater than the elevator's maximum load capacity; the number of passengers assigned to the elevator cannot be greater than the elevator's maximum load capacity;

[0042] The coding of individuals is represented by the second adjacency matrix;

[0043] Step S202, calculating the fitness values ​​of K individuals through the objective function;

[0044] Step S203, generating a first random number in the range of 0 to 1, when the first random number is less than the first threshold, proceeding to step S204, otherwise proceeding to step S205;

[0045] The first threshold is a custom parameter;

[0046] Step S204, when the first random number is greater than or equal to the second threshold, the code of the individual is updated through the perching strategy, otherwise the code of the individual is updated through the hovering strategy;

[0047] The second threshold is a custom parameter;

[0048] Step S205, updating the individual code through the diving strategy;

[0049] Step S206, updating the individual code through the local escape strategy;

[0050] Step S207, the current iteration number is increased by 1. When the iteration termination condition is met, the code of the individual with the largest fitness value in the initialized population is output, otherwise, the process returns to step S202 to continue execution.

[0051] Furthermore, the objective function is calculated as follows: ;

[0052] Where Fit represents the fitness value of the individual, Q represents the number of elements with a value of 1 in the first adjacency matrix, that is, the floor to which the elevator is assigned. Indicates the floor where the mth elevator is located at the current moment, represents the qth floor assigned to the mth elevator, represents the passenger flow of the qth floor assigned to the mth elevator, Indicates the maximum number of people that can be carried by the mth elevator. represents the unit energy consumption of the mth elevator, and denote the first weight coefficient and the second weight coefficient respectively, and The sum of is 1.

[0053] Furthermore, the iteration termination conditions include: the current iteration number is greater than or equal to the maximum iteration number, where the maximum iteration number is a custom parameter; within three consecutive iterations, the rate of change of the maximum value of the fitness value is within a third threshold, where the third threshold is a custom parameter.

[0054] Furthermore, when the element value of the first adjacency matrix is ​​1, the elevator corresponding to the column stops at the floor corresponding to the row, and when the element value of the first adjacency matrix is ​​0, the elevator corresponding to the column does not stop at the floor corresponding to the row.

[0055] The beneficial effect of the present invention is that the present invention comprehensively considers the load and energy consumption of the elevator through the elevator scheduling scheme generation model combined with the kingfisher optimization algorithm, and completes the dynamic partitioning of the elevator according to the passenger flow on different floors in real time, thereby reducing the energy consumption of the elevator. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of a group control peak period dynamic partition energy-saving elevator dispatching method of the present invention;

[0057] Figure 2 is a flow chart of generating a second adjacency matrix by using the Pied Kingfisher optimization algorithm of the present invention;

[0058] Figure 3 It is a comparison chart of the experimental results of the present invention. DETAILED DESCRIPTION

[0059] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0060] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0061] like Figure 1 to Figure 3 As shown, a group control peak period dynamic partition energy-saving elevator scheduling method includes the following steps:

[0062] Step S101, collecting the operation data of M elevators and the passenger flow of N floors at the current moment, and constructing graph structure data;

[0063] Graph structure data includes: nodes, initial vectors of nodes, and edges between nodes;

[0064] The nodes include: M first nodes and N second nodes;

[0065] The M first nodes respectively establish mapping relationships with the M elevators;

[0066] The N second nodes respectively establish mapping relationships with the N floors;

[0067] The initial vectors of the M first nodes are represented by the operation data of the elevator;

[0068] The initial vectors of the N second nodes are represented by the flow of people on the floor;

[0069] Step S102, constructing an elevator dispatching plan generation model;

[0070] The input of the elevator scheduling plan generation model is graph structure data;

[0071] The output of the elevator scheduling plan generation model is the first adjacency matrix;

[0072] The element value of the first adjacency matrix is ​​represented by 0 or 1. The element value of the mth column of the nth row of the first adjacency matrix is ​​1, which means that the mth elevator stops at the nth floor, otherwise it means that the mth elevator does not stop at the nth floor, where 1≤m≤M, 1≤n≤N;

[0073] Step S103, generating a second adjacency matrix by using the Pied Kingfisher optimization algorithm as sample labels of training samples for training an elevator scheduling solution generation model;

[0074] The representation of the second adjacency matrix is ​​the same as the representation of the first adjacency matrix;

[0075] Step S104, inputting the graph structure data into the trained elevator scheduling solution generation model, and outputting a first adjacency matrix;

[0076] Step S105: Decode the first adjacency matrix to obtain an elevator scheduling solution.

[0077] In one embodiment of the present invention, the operation data of the elevator include: the floor where the elevator is located, the direction of the elevator, the current load, the maximum load, the current number of passengers, the maximum number of passengers, the number of historical maintenance times, the number of historical accidents and the unit energy consumption, wherein the direction of the elevator includes: upward, downward and stop, which are represented by real number codes. For example, the directions of the elevator upward, downward and stop are represented by real numbers 1.1, 1.2 and 1.3 respectively. The maximum load and the maximum number of passengers are set by the elevator manufacturer. The unit energy consumption is represented by the power consumption of the elevator when it travels one floor in an empty state.

[0078] It should be noted that the current number of passengers and the flow of people on each floor can be obtained through a convolutional neural network model or a target detection model. The target detection model can be a YOLO model, etc., which will not be elaborated here.

[0079] In one embodiment of the present invention, the edge between nodes includes: the edge constructed between the first node and the second node indicates that the floor mapping with the second node generates a dispatch instruction for the elevator mapping with the first node.

[0080] It should be noted that the graph structure data represents the static partitioning of elevators. For example, the 10th floor generates a dispatching instruction for the 1st elevator and the 2nd elevator, then an edge is constructed between the 10th second node and the 1st and 2nd first nodes, and the dispatching instructions include: an uplink dispatching instruction and a downlink dispatching instruction. The graph structure data is updated by the elevator dispatching scheme generation model to generate a first adjacency matrix, and the second adjacency matrix generated by the Pied Kingfisher optimization algorithm is approximated to achieve dynamic partitioning of elevators.

[0081] In one embodiment of the present invention, the elevator dispatching scheme generation model includes:

[0082] First hidden layer, second hidden layer, third hidden layer;

[0083] The first hidden layer inputs the initial vector of the first node and outputs the first update vector;

[0084] The second hidden layer inputs the initial vector of the second node and outputs the second update vector;

[0085] The first update vector and the second update vector have the same number of dimensions;

[0086] That is, the initial vector of the node of the graph structure data is represented by the update vector;

[0087] The third hidden layer inputs the graph structure data and outputs the first adjacency matrix.

[0088] In one embodiment of the present invention, the calculation formula of the first hidden layer includes: ; ;

[0089] Where 1≤i≤H, H represents the number of dimensions of the initial vector of the first node, that is, H is equal to 9, represents the first update vector of the first hidden layer output, Represents the value of the i-th dimension of the initial vector of the first node, and Respectively represent the mean and standard deviation of the initial vector of the first node, It means to concatenate the values ​​of H dimensions, NormInit means the normalized vector, represents the first weight parameter, represents the first bias parameter, ReLU represents the ReLU activation function; the calculation formula of the second hidden layer is as follows: in represents the second update vector for the output of the second hidden layer, The initial vector of the second node representing the input of the second hidden layer, represents the second weight parameter, Represents the second bias parameter.

[0090] It should be noted that the number of dimensions of the normalized vector is the same as the number of dimensions of the initial vector of the first node, both of which are 1×9. Therefore, the size of the first weight parameter can be designed as a matrix of size 9×16. The size of the first update vector obtained by multiplying the two is 1×16. Then the size of the second weight parameter can be designed to be 1×16, that is, the size of the second update vector is also 1×16.

[0091] It should be noted that H is equal to 9, which corresponds to the floor where the elevator is located, the direction of the elevator, the current load, the maximum load, the current number of people carrying the load, the maximum number of people carrying the load, the number of historical maintenance times, the number of historical accidents, and the unit energy consumption.

[0092] In one embodiment of the present invention, the calculation formula of the third hidden layer includes: ; ; ; Where S represents the first adjacency matrix of the third hidden layer output. If the element value in S is greater than or equal to 0.5, it is assigned to 1, otherwise it is assigned to 0. represents the intermediate vector of the u-th node, and Represent the update vectors of the u-th node and the v-th node respectively, represents the set of nodes that build edges with the u-th node, , and represent the third weight parameter, the fourth weight parameter and the fifth weight parameter respectively, represents the attention coefficient between the uth node and the vth node, MLP represents a multi-layer perceptron, and pile represents a stacking operation. Indicates stacking the intermediate vectors of N second nodes. Indicates that the intermediate vectors of the M first nodes are stacked, T indicates a transpose operation, and sigmoid indicates a sigmoid activation function.

[0093] It should be noted that the size of the node update vector is 1×16, and the third weight parameter can be designed as a matrix of 8×16. Then the vector size obtained by multiplying the third weight parameter by the transpose of the node update vector is 8×1. Similarly, the fourth weight parameter is also designed as a matrix of 8×16. Then the size of the intermediate vector of the node is 8×1. The intermediate vectors of the N second nodes are stacked to obtain a matrix of N×8, and the intermediate vectors of the M first nodes are stacked to obtain a matrix of M×8. The N×8 matrix is ​​multiplied by the transpose of the M×8 matrix, and the size of the first adjacency matrix obtained is N×M.

[0094] It should be noted that the weight parameters and bias parameters in the elevator scheduling plan generation model are all learnable parameters. In the process of training the elevator scheduling plan generation model, the difference between the first adjacency matrix and the second adjacency matrix output by the elevator scheduling plan generation model is specified as the loss function, and the gradient information of the weight parameters and the bias parameters is calculated by the chain rule. The weight parameters and the bias parameters are updated according to the gradient information combined with the gradient descent algorithm to minimize the loss function. Commonly used gradient descent algorithms include stochastic gradient descent, adaptive learning rate method, etc. The adaptive learning rate methods include AdaGrad, Adam, etc., which are not described here.

[0095] In one embodiment of the present invention, the calculation formula of the loss function of the elevator scheduling solution generation model is as follows: ;in and Represent the element values ​​of the nth row and the mth column of the first adjacency matrix and the second adjacency matrix respectively.

[0096] In one embodiment of the present invention, Figure 2 As shown, the second adjacency matrix is ​​generated by the Kingfisher optimization algorithm, including the following steps:

[0097] Step S201, randomly generate an initialization population consisting of K individuals that meet the constraints, and initialize the current number of iterations to 0;

[0098] Where K is a custom parameter, preferably, K is set to 10;

[0099] The constraints include: the load capacity assigned to the elevator cannot be greater than the elevator's maximum load capacity; the number of passengers assigned to the elevator cannot be greater than the elevator's maximum load capacity;

[0100] The coding of individuals is represented by the second adjacency matrix;

[0101] Step S202, calculating the fitness values ​​of K individuals through the objective function;

[0102] Step S203, generating a first random number in the range of 0 to 1, when the first random number is less than the first threshold, proceeding to step S204, otherwise proceeding to step S205;

[0103] The first threshold is a custom parameter. Preferably, the first threshold is set to 0.8;

[0104] Step S204, when the first random number is greater than or equal to the second threshold, the code of the individual is updated through the perching strategy, otherwise the code of the individual is updated through the hovering strategy;

[0105] The second threshold is a custom parameter. Preferably, the second threshold is set to 0.5;

[0106] Step S205, updating the individual code through the diving strategy;

[0107] Step S206, updating the individual code through the local escape strategy;

[0108] Step S207, the current iteration number is increased by 1. When the iteration termination condition is met, the code of the individual with the largest fitness value in the initialized population is output, otherwise, the process returns to step S202 to continue execution.

[0109] In one embodiment of the present invention, the calculation formula of the objective function is as follows: ; Where Fit represents the fitness value of the individual, Q represents the number of elements with a value of 1 in the first adjacency matrix, that is, the floor to which the elevator is assigned, Indicates the floor where the mth elevator is located at the current moment, represents the qth floor assigned to the mth elevator, represents the passenger flow of the qth floor assigned to the mth elevator, Indicates the maximum number of people that can be carried by the mth elevator. represents the unit energy consumption of the mth elevator, and denote the first weight coefficient and the second weight coefficient respectively, and The sum of the values ​​is 1, preferably, and Set them to 0.5 respectively.

[0110] In one embodiment of the present invention, the iteration termination conditions include: the current iteration number is greater than or equal to the maximum iteration number, where the maximum iteration number is a custom parameter, preferably, the maximum iteration number is set to 50; within three consecutive iterations, the rate of change of the maximum value of the fitness value is within a third threshold, where the third threshold is a custom parameter, preferably, the third threshold is set to 0.1.

[0111] In one embodiment of the present invention, the calculation formula of the Pied Kingfisher optimization algorithm includes:

[0112] Definition: 1≤i≤K, 1≤j≤K, i≠j, and the jth , represents the code of the i-th individual whose current iteration number is t+1, Indicates the code of the individual with the largest fitness value at the current iteration number t, and They represent the fitness values ​​of the i-th individual and the j-th individual with the current iteration number t, respectively. The maximum value of the fitness value of the individual in the current iteration t, Max represents the maximum number of iterations, and the calculation formula of the habitat strategy includes: ; ; ;in and Represent the second and third random numbers respectively, BF is a constant value, assigned to 8, represents a fourth random number ranging from 0 to 1, e represents a natural constant, and exp represents an exponential function with the natural constant as a base;

[0113] The calculation formula of the hover strategy includes: ; ; ;in Indicates a fifth random number ranging from 0 to 1;

[0114] The calculation formula of the diving strategy includes: ; ; ; ; ;in and They represent the sixth and seventh random numbers in the range of 0 to 1, respectively. and denote the first control parameter and the second control parameter respectively, and b denotes the control factor;

[0115] The calculation formula of the local escape strategy includes: ; ; ; where 1≤m≤K, 1≤n≤K, i≠m≠n, and They represent the codes of the mth individual and the nth individual with the current iteration number t, respectively. abs represents the absolute value. The eighth random number in the range of 0 to 1 is shown. Indicates the ninth random number in the range of 0 to 0.5.

[0116] It should be noted that during the iteration of the Pied Kingfisher optimization algorithm, the coding of each individual needs to be corrected, that is, the second adjacency matrix of each individual needs to be corrected. If the element value of the second adjacency matrix is ​​greater than or equal to 0.5, it is directly assigned to 1, otherwise it is assigned to 0.

[0117] In one embodiment of the present invention, when the element value of the first adjacency matrix is ​​1, the elevator corresponding to the column stops at the floor corresponding to the row, and when the element value of the first adjacency matrix is ​​0, the elevator corresponding to the column does not stop at the floor corresponding to the row.

[0118] In one embodiment of the present invention, Figure 3 As shown, the scheduling of 30 floors allocated to 6 elevators is completed by the elevator scheduling scheme generation model provided by the present invention and the single spotted kingfisher optimization algorithm respectively. It is found through observation that the speed of generating elevator scheduling schemes by the elevator scheduling scheme generation model provided by the present invention is obviously faster than the speed of generating elevator scheduling schemes by the single spotted kingfisher optimization algorithm.

[0119] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are protected by the present embodiment.

Claims

1. A group control peak period dynamic partition energy-saving elevator dispatching method, characterized in that: The following steps are involved: Step S101, collecting the operation data of M elevators and the passenger flow of N floors at the current moment, and constructing graph structure data; Graph structure data includes: nodes, initial vectors of nodes, and edges between nodes; The nodes include: M first nodes and N second nodes; The M first nodes respectively establish mapping relationships with the M elevators; The N second nodes respectively establish mapping relationships with the N floors; The initial vectors of the M first nodes are represented by the operation data of the elevator; The initial vectors of the N second nodes are represented by the flow of people on the floor; Step S102, constructing an elevator dispatching plan generation model; The input of the elevator scheduling plan generation model is graph structure data; The output of the elevator scheduling plan generation model is the first adjacency matrix; The element value of the first adjacency matrix is ​​represented by 0 or 1. The element value of the mth column of the nth row of the first adjacency matrix is ​​1, which means that the mth elevator stops at the nth floor, otherwise it means that the mth elevator does not stop at the nth floor, where 1≤m≤M, 1≤n≤N; Step S103, generating a second adjacency matrix by using the Pied Kingfisher optimization algorithm as sample labels of training samples for training an elevator scheduling solution generation model; The representation of the second adjacency matrix is ​​the same as the representation of the first adjacency matrix; Step S104, inputting the graph structure data into the trained elevator scheduling solution generation model, and outputting a first adjacency matrix; Step S105, decoding the first adjacency matrix to obtain an elevator scheduling solution; The second adjacency matrix is ​​generated by the Pied Kingfisher optimization algorithm, including the following steps: Step S201, randomly generate an initialization population consisting of K individuals that meet the constraints, and initialize the current number of iterations to 0; Where K is a custom parameter; The constraints include: the load capacity assigned to the elevator cannot be greater than the elevator's maximum load capacity; the number of passengers assigned to the elevator cannot be greater than the elevator's maximum load capacity; The coding of individuals is represented by the second adjacency matrix; Step S202, calculating the fitness values ​​of K individuals through the objective function; Step S203, generating a first random number in the range of 0 to 1, when the first random number is less than the first threshold, proceeding to step S204, otherwise proceeding to step S205; The first threshold is a custom parameter; Step S204, when the first random number is greater than or equal to the second threshold, the code of the individual is updated through the perching strategy, otherwise the code of the individual is updated through the hovering strategy; The second threshold is a custom parameter; Step S205, updating the individual code through the diving strategy; Step S206, updating the individual code through the local escape strategy; Step S207, the current iteration number is increased by 1. When the iteration termination condition is met, the code of the individual with the largest fitness value in the initialized population is output, otherwise, the process returns to step S202 to continue execution; The objective function is calculated as follows: ； Where Fit represents the fitness value of the individual, Q represents the number of elements with a value of 1 in the first adjacency matrix, that is, the floor to which the elevator is assigned. Indicates the floor where the mth elevator is located at the current moment, represents the qth floor assigned to the mth elevator, represents the passenger flow of the qth floor assigned to the mth elevator, Indicates the maximum number of people that can be carried by the mth elevator. represents the unit energy consumption of the mth elevator, and denote the first weight coefficient and the second weight coefficient respectively, and The sum of the values ​​is 1; the calculation formula of the Pied Kingfisher optimization algorithm includes: Definition: 1≤i≤K, 1≤j≤K, i≠j, and They represent the codes of the i-th and j-th individuals with the current iteration number t, respectively. represents the code of the i-th individual whose current iteration number is t+1, Indicates the code of the individual with the largest fitness value at the current iteration number t, and They represent the fitness values ​​of the i-th individual and the j-th individual with the current iteration number t, respectively. The maximum value of the fitness value of the individual in the current iteration t, Max represents the maximum number of iterations; the calculation formula of the habitat strategy includes: ； ； ； in and Represent the second and third random numbers respectively, BF is a constant value, assigned to 8, represents the fourth random number ranging from 0 to 1, e represents a natural constant, and exp represents an exponential function with the natural constant as the base; the calculation formula of the hovering strategy includes: ； ； ； in Indicates the fifth random number with a value range of 0 to 1; the calculation formula of the diving strategy includes: ； ； ； ； ； in and They represent the sixth and seventh random numbers in the range of 0 to 1, respectively. and They represent the first control parameter and the second control parameter respectively, and b represents the control factor; the calculation formula of the local escape strategy includes: ； ； ； Where 1≤m≤K, 1≤n≤K, i≠m≠n, and They represent the codes of the mth individual and the nth individual with the current iteration number t, respectively. abs represents the absolute value. Indicates the eighth random number ranging from 0 to 1. It represents the ninth random number with a value range of 0 to 0.

5. During the iteration of the Pied Kingfisher optimization algorithm, the code of each individual is corrected, and if the element value of the second adjacency matrix is ​​greater than or equal to 0.5, it is directly assigned a value of 1, otherwise it is assigned a value of 0.

2. A group control peak period dynamic partition energy-saving elevator dispatching method according to claim 1, characterized in that: The operating data of the elevator include: the floor where the elevator is located, the direction of the elevator, the current load, the maximum load, the current number of passengers, the maximum number of passengers, the number of historical maintenance times, the number of historical accidents and the unit energy consumption. The directions of the elevator include: up, down and stay, which are represented by real number codes. The maximum load and the maximum number of passengers are set by the elevator manufacturer. The unit energy consumption is represented by the power consumption of the elevator when it travels one floor in an empty state.

3. A group control peak period dynamic partition energy-saving elevator dispatching method according to claim 1, characterized in that: The edges between the nodes include: the edge constructed between the first node and the second node indicates that the floor that establishes a mapping relationship with the second node generates a dispatch instruction for the elevator that establishes a mapping relationship with the first node.

4. A group control peak period dynamic partition energy-saving elevator dispatching method according to claim 1, characterized in that: The elevator dispatching scheme generation model includes: First hidden layer, second hidden layer, third hidden layer; The first hidden layer inputs the initial vector of the first node and outputs the first update vector; The second hidden layer inputs the initial vector of the second node and outputs the second update vector; The first update vector and the second update vector have the same number of dimensions; The initial vector of the node of the graph structure data is represented by the update vector; The third hidden layer inputs the graph structure data and outputs the first adjacency matrix.

5. A group control peak period dynamic partition energy-saving elevator dispatching method according to claim 4, characterized in that: The calculation formula for the first hidden layer includes: ； ； Where 1≤i≤H, H represents the number of dimensions of the initial vector of the first node, H is equal to 9, represents the first update vector of the first hidden layer output, Represents the value of the i-th dimension of the initial vector of the first node, and Respectively represent the mean and standard deviation of the initial vector of the first node, It means to concatenate the values ​​of H dimensions, NormInit means the normalized vector, represents the first weight parameter, represents the first bias parameter, ReLU represents the ReLU activation function; the calculation formula of the second hidden layer is as follows: ;in represents the second update vector for the output of the second hidden layer, The initial vector of the second node representing the input of the second hidden layer, represents the second weight parameter, Represents the second bias parameter.

6. A group control peak period dynamic partition energy-saving elevator dispatching method according to claim 4, characterized in that: The calculation formula for the third hidden layer includes: ； ； ； Where S represents the first adjacency matrix of the third hidden layer output. If the element value in S is greater than or equal to 0.5, it is assigned to 1, otherwise it is assigned to 0. represents the intermediate vector of the u-th node, and Represent the update vectors of the u-th node and the v-th node respectively, represents the set of nodes that build edges with the u-th node, , and represent the third weight parameter, the fourth weight parameter and the fifth weight parameter respectively, represents the attention coefficient between the uth node and the vth node, MLP represents a multi-layer perceptron, and pile represents a stacking operation. Indicates stacking the intermediate vectors of N second nodes. Indicates that the intermediate vectors of the M first nodes are stacked, T indicates a transpose operation, and sigmoid indicates a sigmoid activation function.

7. A group control peak period dynamic partition energy-saving elevator dispatching method according to claim 1, characterized in that: The iteration termination conditions include: the current iteration number is greater than or equal to the maximum iteration number, where the maximum iteration number is a custom parameter; within three consecutive iterations, the rate of change of the maximum value of the fitness value is within a third threshold, where the third threshold is a custom parameter.

Images