Intelligent building elevator traffic scheduling method and system for energy-saving optimization

Abstract

The invention provides an intelligent building elevator traffic scheduling method and system for energy-saving optimization. The method comprises the steps that real-time data of building sensing equipment and an elevator monitoring system are collected; constructing a prediction model to obtain the flow thermal distribution of future floors; designing a scheduling strategy based on multi-target reinforcement learning, wherein the scheduling strategy comprises target floor distribution and a dynamic operation path of the elevator; and formulating an execution control strategy for energy consumption optimization to realize global energy consumption minimization, and generating an optimized control instruction. Through deep fusion of flow prediction, scheduling optimization and energy consumption management, the passenger experience is remarkably improved while the adaptability, flexibility and energy efficiency of the system are improved, and an intelligent and efficient elevator scheduling solution is provided for modern high-rise buildings.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent building elevator traffic scheduling, and in particular relates to an intelligent building elevator traffic scheduling method and system for energy-saving optimization. Background Art

[0002] With the acceleration of urbanization and the popularity of high-rise buildings, elevators have become the core equipment for vertical transportation in buildings. Traditional elevator dispatching systems usually use preset rules or simple algorithms (such as first-come, first-served or shortest response time) to manage elevator operations. These systems can operate effectively in low-complexity scenarios, but they show obvious limitations in modern complex buildings, especially office buildings, commercial centers or residential communities with dense traffic.

[0003] On the one hand, the concentrated distribution of passenger flow during peak hours and the sparse flow characteristics during off-peak hours make it difficult for traditional elevator scheduling strategies to strike a balance between efficiency and energy consumption. During peak hours, the uneven allocation of elevator resources often leads to excessive waiting times and frequent overload alarms, while during off-peak hours, a lot of energy is wasted due to the empty operation of elevators. This phenomenon not only reduces the travel experience of passengers, but also increases the operating costs of buildings. On the other hand, the dynamic behavioral characteristics of building crowds (such as time period fluctuations in traffic and hot spot effects on specific floors) make it difficult for traditional methods to perceive and respond to crowd changes in a timely manner, and scheduling decisions lag behind actual needs. In addition, the existing technology lacks load balancing optimization for long-term use of equipment, which leads to excessive use of some elevators, accelerated equipment aging, and increased maintenance costs.

[0004] Although some studies have tried to introduce optimization algorithms (such as genetic algorithms, ant colony algorithms) and data-driven methods (such as statistical analysis or simple prediction models) to improve scheduling strategies, most of them lack the deep integration of time-space flow characteristics and real-time data within buildings. This leads to the scheduling system still showing problems such as delayed response, lack of flexibility and limited energy saving effect when dealing with diverse demands and complex environments.

[0005] In summary, the core problems of the prior art can be summarized as follows:

[0006] The scheduling strategy has poor adaptability to complex traffic characteristics: it cannot respond to peak and trough traffic changes in real time, resulting in long waiting times and no-load operation problems.

[0007] Lack of accurate prediction of future traffic changes: Existing methods are usually based on static rules or simple predictions and cannot adjust scheduling strategies in advance.

[0008] Inefficient energy utilization: The existing system fails to effectively combine real-time demand and equipment status to optimize energy consumption management.

[0009] Unbalanced equipment utilization: Long-term unbalanced load distribution accelerates the aging of some equipment and increases maintenance costs. Summary of the invention

[0010] The purpose of this invention is to propose an intelligent building elevator traffic scheduling method and system for energy-saving optimization. Through the deep integration of traffic prediction, scheduling optimization and energy consumption management, the system adaptability, flexibility and energy efficiency are improved while significantly improving the passenger experience, providing a smart and efficient elevator scheduling solution for modern high-rise buildings.

[0011] In order to achieve the above object, a first aspect of the present invention provides an intelligent building elevator traffic scheduling method for energy saving optimization, the method comprising:

[0012] S1. Collect real-time data from building sensor equipment and elevator monitoring systems and perform preprocessing. Perform feature extraction based on time windows based on the preprocessed data to obtain a weighted feature matrix. Use the physical proximity between floors to construct an association regularization term to enhance the floor association features of the weighted feature matrix. Convert the weighted feature matrix into a time series feature matrix based on a time window to capture historical trends and obtain a feature time series matrix.

[0013] S2. Build a prediction model based on the characteristic time series matrix to obtain the flow and thermal distribution of future floors. At the same time, build an optimization objective function based on the predicted value of the prediction model and the actual prediction error to train the prediction model. Use the characteristic time series matrix as the input of the trained prediction model to output the flow and thermal distribution within K time steps in the future period to identify high-demand floors and time periods.

[0014] S3. Based on the characteristic time series matrix and the flow thermal distribution within K time steps in the future period, a scheduling strategy based on multi-objective reinforcement learning is designed, including the target floor allocation and dynamic operation path of the elevator;

[0015] The reinforcement learning model based on multi-objective reinforcement learning is designed as follows:

[0016] State: Construct the state vector s based on the characteristic time series matrix and the flow thermal distribution in K time steps in the future period t =[H(t+k),D(t)];

[0017] Action: Scheduling decision vector a t , including elevator target floor allocation and elevator start and stop control;

[0018] Reward function:

[0019] R t=-α·WaitTime-β·Energy-γ·Imbalance+δ·Priority

[0020] Among them, WaitTime represents the average waiting time of all passengers in the system; Energy represents the total energy consumption, including the energy consumption of elevator operation and no-load operation; Imbalance represents the difference in load balance between elevators, calculated as the load standard deviation; Priority represents the priority response degree of high-traffic floors, encouraging requests from hot floors to be processed in a timely manner; α, β, γ, δ represent weight coefficients, controlling the importance of each target;

[0021] Using a multi-objective reinforcement learning model based on policy optimization Directly learn the optimal policy π * , the state s t Mapped to action a t :

[0022]

[0023] Among them, π represents the mapping strategy from state to action; π * represents the optimal strategy; γ represents the discount factor, which controls the importance of future rewards;

[0024] S4. According to the scheduling strategy and characteristic timing matrix and combined with the energy consumption state matrix, an energy consumption optimized execution control strategy is formulated to minimize the global energy consumption and generate optimized control instructions.

[0025] Furthermore, the real-time data includes:

[0026] Floor call matrix C(t): C(t)[i,j] represents the call intensity from floor i to floor j at time t;

[0027] Elevator state matrix S(t): describes the real-time position, load and direction of each elevator, S k (t) = [pos k ,load k ,dir k ],in:

[0028] POS k Indicates the current location floor of elevator k;

[0029] load k Indicates the load ratio;

[0030] dir k Indicates the running direction, with values ​​of {-1,0,1}, representing downlink, stationary, and uplink respectively;

[0031] Temporal feature T(t): represents context information;

[0032] The real-time data is combined into the basic feature matrix D(t) to express the state of the building elevator system at the current moment:

[0033] D(t)=[C(t),S(t),T(t)];

[0034] The preprocessing comprises:

[0035] Designing a dynamic anomaly detection model For outliers that may appear in elevator and floor call data, historical expected values ​​are used to eliminate outliers, expressed as:

[0036]

[0037] in, It is a dynamic expected value calculated based on historical data. 2 Represents the second norm, which is used to measure the degree of deviation of the feature matrix.

[0038] Furthermore, the association regularization term is constructed as follows:

[0039]

[0040] in, represents the physical proximity weight of floors i and j; is the weight matrix of floor i, is the weight matrix of floor j, and α is the weight coefficient of the regularization term.

[0041] Furthermore, the prediction model P has the following structure:

[0042] H(t+k)=P(D seq )

[0043] Among them, H(t+k) is the predicted flow thermal distribution of the future floor, D seq is the characteristic time series matrix;

[0044] The optimization objective function also includes local smoothing regularization and dynamic weighting terms:

[0045] The local smoothing regularization Ensure the continuity of thermal distribution between adjacent floors and reduce unreasonable prediction mutations, which can be expressed as:

[0046]

[0047] Among them, W i,j represents the physical proximity weight of floors i and j; β controls the regularization strength;

[0048] The dynamic weighting term Used to highlight the importance of high traffic floors, expressed as:

[0049]

[0050] Among them, ρ i is the flow weight of floor i, which is dynamically adjusted to the normalized value of the past flow mean, so that the highest demand floor has the most significant impact on the optimization; H true,i (t+k) is the real flow, H i (t+k) is the predicted flow rate.

[0051] Furthermore, the optimization objective function , expressed as:

[0052]

[0053] Item 1 is the prediction error, which measures the difference between the model prediction H(t+k) and the actual flow H true The gap of (t+k);

[0054] The second term is the local smoothing regularization Ensure the continuity of traffic distribution between adjacent floors;

[0055] The third item is the dynamic weighted item Make the optimization more focused on prediction accuracy on high traffic floors.

[0056] Furthermore, the S3 further includes:

[0057] In the decision-making process, the dynamic partitioning and priority scheduling strategies are designed in combination with the predicted future floor flow and thermal distribution H(t+k):

[0058] The dynamic partitioning: according to the predicted future floor flow thermal distribution H(t+k), the service area of ​​the elevator is dynamically adjusted, and high-flow floors are preferentially allocated to exclusive elevators, thereby reducing the no-load energy consumption of cross-zone operation;

[0059] The priority scheduling strategy: within the partition, give higher priority to the high-traffic floors to ensure quick response to requests from hot floors. i Calculated as:

[0060] P i =H i (t+k)·ρ i

[0061] Among them, ρ i It is the inverse proportional weight of the historical response delay. The longer the delay, the higher the priority.

[0062] Furthermore, the energy consumption state matrix includes operating power consumption, no-load energy consumption of the elevator, and braking energy recovery.

[0063] Furthermore, the S4 specifically includes:

[0064] S41. Construct an energy consumption optimization model to minimize global energy consumption;

[0065] S42. Based on the energy consumption optimization model, a dynamic start-stop mechanism is proposed to reduce the no-load energy consumption during the valley period, including:

[0066] Based on the predicted future floor flow thermal distribution and characteristic time series matrix, the idle elevator is judged to be dormant, and the dormant decision is defined as:

[0067]

[0068] Among them, ∈ represents the flow threshold, which controls the sensitivity of dormancy judgment; T threshold Indicates the minimum idle time threshold, used to avoid frequent starts and stops;

[0069] When the traffic in a certain area exceeds the dynamic threshold δ, the neighboring dormant elevators are woken up first to reduce the waiting time during high demand;

[0070] S43. Design an energy recovery distribution model to use the braking energy for elevator operation or other building equipment;

[0071] S44, generating optimized control instructions by combining the scheduling strategy and the energy consumption optimization model, including:

[0072] Elevator target floor and path planning;

[0073] Dynamic start-stop control instructions, including sleep and wake-up decisions;

[0074] Develop energy allocation plans to optimize energy consumption and utilization.

[0075] Furthermore, the energy consumption optimization model is expressed as:

[0076] ε=TotalEnergy-λ·EnergyRecovery

[0077] Among them, TotalEnergy represents the total energy consumption of the system, including the elevator operation power consumption and no-load operation power consumption, and ε is the flow threshold:

[0078]

[0079] in, and Respectively represent the unit power consumption of the elevator in operation and without load; and is the running and no-load time of elevator k; EnergyRecovery represents the total energy recovered by the system through braking energy; λ: the effective utilization coefficient of recovered energy, which is used to quantify the contribution of energy recovery to the system;

[0080] The energy recovery allocation model is expressed as:

[0081]

[0082] Among them, EnergyRedistribution is the total recovered energy, It represents the recovered energy generated by elevator k braking; η represents the energy recovery efficiency, which is used to control the distribution ratio of recovered energy.

[0083] In another aspect of the present invention, a smart building elevator traffic dispatching system for energy saving optimization is provided, the system comprising:

[0084] A real-time data acquisition unit is used to collect real-time data from building sensor equipment and elevator monitoring systems and perform preprocessing. Based on the preprocessed data, feature extraction based on a time window is performed to obtain a weighted feature matrix. The physical proximity between floors is used to construct an associated regular term to enhance the floor association features of the weighted feature matrix. The weighted feature matrix is ​​converted into a time series feature matrix based on a time window to capture historical trends, thereby obtaining a feature time series matrix.

[0085] The flow prediction unit is used to build a prediction model based on the characteristic time series matrix to obtain the flow thermal distribution of future floors. At the same time, the optimization objective function is built based on the predicted value of the prediction model and the actual prediction error to train the prediction model. The characteristic time series matrix is ​​used as the input of the trained prediction model to output the flow thermal distribution within K time steps in the future period, which is used to identify high-demand floors and time periods.

[0086] The scheduling strategy generation unit is used to design a scheduling strategy based on multi-objective reinforcement learning based on the characteristic time series matrix and the flow thermal distribution within K time steps in the future period, including the target floor allocation and dynamic operation path of the elevator;

[0087] The reinforcement learning model based on multi-objective reinforcement learning is designed as follows:

[0088] State: Construct the state vector s based on the characteristic time series matrix and the flow thermal distribution in K time steps in the future period t =[H(t+k),D(t)];

[0089] Action: Scheduling decision vector a t , including elevator target floor allocation and elevator start and stop control;

[0090] Reward function:

[0091] R t =-α·WaitTime-β·Energy-γ·Imbalance+δ·Priority

[0092] Among them, WaitTime represents the average waiting time of all passengers in the system; Energy represents the total energy consumption, including the energy consumption of elevator operation and no-load operation; Imbalance represents the difference in load balance between elevators, calculated as the load standard deviation; Priority represents the priority response degree of high-traffic floors, encouraging requests from hot floors to be processed in a timely manner; α, β, γ, δ represent weight coefficients, controlling the importance of each target;

[0093] Using a multi-objective reinforcement learning model based on policy optimization Directly learn the optimal policy π * , the state s t Mapped to action a t :

[0094]

[0095] Among them, π represents the mapping strategy from state to action; π * represents the optimal strategy; γ represents the discount factor, which controls the importance of future rewards;

[0096] The scheduling strategy optimization unit is used to formulate an energy-optimized execution control strategy based on the scheduling strategy and the characteristic timing matrix in combination with the energy consumption state matrix to minimize the global energy consumption and generate optimized control instructions.

[0097] The beneficial technical effects of the present invention are at least as follows:

[0098] The present invention introduces a flow prediction model based on spatiotemporal convolutional neural network (ST-CNN), which can analyze the flow distribution in the building in real time and accurately predict the demand heat map of each floor in the future. Compared with traditional methods, this module can not only capture the flow fluctuation characteristics of the time period, but also identify the hot spot effects of different floors, providing high-precision prediction support for scheduling optimization. Through flow prediction, the system can adjust the elevator partition and operation strategy in advance, thereby greatly reducing waiting time and overload problems.

[0099] In order to cope with complex traffic distribution and diversified needs, the present invention designs a multi-objective optimization scheduling framework based on deep reinforcement learning (such as PPO). This module takes minimizing waiting time, energy consumption and equipment load imbalance as optimization goals, and dynamically generates the optimal scheduling strategy based on real-time call requests, elevator status and traffic prediction results. Through adaptive learning, the system can balance response efficiency and load distribution during peak periods, and optimize elevator start and stop strategies during trough periods, thereby achieving efficient and energy-saving operation.

[0100] To further reduce energy consumption and equipment maintenance costs, the present invention combines traffic prediction and scheduling optimization, and introduces energy consumption management and equipment health monitoring mechanisms. The system can dynamically adjust the elevator start and stop strategy during off-peak periods to avoid unnecessary no-load operation, while recovering braking energy and allocating it to high-demand areas first. In addition, by monitoring the long-term load distribution of the equipment, the system can balance the frequency of equipment use, extend equipment life, and reduce maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0102] Figure 1 The present invention is a flow chart of an intelligent building elevator traffic scheduling method for energy saving optimization. DETAILED DESCRIPTION

[0103] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0104] like Figure 1 As shown, an embodiment of the present invention provides an intelligent building elevator traffic scheduling method for energy saving optimization, the method comprising the following steps:

[0105] S1. Collect real-time data from building sensor equipment and elevator monitoring systems and preprocess them. Perform feature extraction based on time windows based on the preprocessed data to obtain a weighted feature matrix. Use the physical proximity between floors to construct association regularization terms to enhance the floor association features of the weighted feature matrix. Convert the weighted feature matrix into a time series feature matrix based on a time window to capture historical trends and obtain a feature time series matrix.

[0106] Specifically, real-time call requests, elevator status and time features are collected to form a feature matrix for use in subsequent steps.

[0107] Furthermore, the following data is collected using building sensor equipment and elevator monitoring systems:

[0108] Floor call matrix C(t): C(t)[i,j] represents the call intensity from floor i to floor j at time t, in units of call times.

[0109] Elevator state matrix S(t): describes the real-time position, load and direction of each elevator, S k (t) = [pos k ,load k ,dir k ],in:

[0110] POS k Indicates the current floor of elevator k.

[0111] load k Indicates the load ratio (normalized to [0,1]).

[0112] dir k Indicates the running direction, and its value is {-1,0,1}, representing downlink, stationary and uplink respectively.

[0113] Time feature T(t): represents contextual information, such as timestamp (weekday / weekend, peak / trough, etc.), in one-hot encoding form.

[0114] Furthermore, the above data are combined into a basic feature matrix D(t) to express the state of the building elevator system at the current moment:

[0115] D(t)=[C(t),S(t),T(t)]

[0116] Furthermore, a dynamic anomaly detection model is designed for possible outliers in elevator and floor call data (such as non-existent floor requests or load data anomalies). Use historical expected values ​​to eliminate outliers and ensure data reliability:

[0117]

[0118] in, It is a dynamic expected value calculated based on historical data. 2 Represents the second norm, which is used to measure the degree of deviation of the feature matrix.

[0119] Furthermore, the cleaned data is used as input for the next step of feature extraction, reducing the interference of abnormal noise on modeling.

[0120] Furthermore, in order to capture the dynamic characteristics of building flow and elevator status, the feature matrix D(t) is mapped to the weighted feature matrix within the time window ΔT The weighted features are defined by the following formula:

[0121]

[0122] Where w(τ) is the time weight function, for example w(τ)=e -λ ( t-τ ), λ is the weight decay parameter, which is used to highlight the importance of data at more recent time points. Weighting matrix It provides dynamic distribution information in the time dimension, including key features such as call frequency changes, load fluctuations, and direction deviations.

[0123] Furthermore, in order to further improve the feature representation capability, the physical proximity between floors is used to construct an association regularization term. Define floor association regularization:

[0124]

[0125] in, Represents the physical proximity weight of floors i and j (e.g., the closer the elevator stops, the closer the elevator stops). The larger the value, the larger the regularization term. The smoothness between features of adjacent floors is enhanced and local sampling errors are reduced.

[0126] Furthermore, the weight matrix Further transformed into the time series feature matrix D seq , to capture historical trends:

[0127]

[0128] D seq The dimension is (ΔT, m, n), where ΔT is the time window length, m is the number of floors, and n is the feature dimension. This feature sequence is the direct input of the subsequent traffic prediction model and provides sufficient spatiotemporal dynamic information.

[0129] Further, output: the cleaned time series feature matrix D seq .

[0130] Function: This feature sequence plays an important role in patent scenarios. As the core input for subsequent steps (such as traffic prediction and scheduling optimization), it provides accurate data support for dynamic optimization.

[0131] Furthermore, through the above design, the scheme effectively integrates floor calls, elevator status and time feature information, and uses time weighting and floor association enhancement to overcome the problem of insufficient spatiotemporal dynamic perception in existing methods.

[0132] S2. Build a prediction model based on the characteristic time series matrix to obtain the flow and thermal distribution of future floors. At the same time, build an optimization objective function based on the predicted value of the prediction model and the actual prediction error to train the prediction model. Use the characteristic time series matrix as the input of the trained prediction model, and output the flow and thermal distribution within K time steps in the future period to identify high-demand floors and time periods.

[0133] Specifically, this step receives the output feature time series matrix D of step 1 seq , whose dimension is (ΔT,m,n), and includes dynamic features such as floor call intensity, elevator status and time context:

[0134] ΔT: time window length.

[0135] m: number of floors.

[0136] n: characteristic dimension (such as floor call frequency, load status, direction characteristics, etc.).

[0137] Each D seq [t,i,k] represents the value of floor i and feature dimension k at time point t. This matrix can capture the historical flow change trend of the floor and is the key input for future flow prediction.

[0138] Furthermore, in order to predict the future floor flow thermal distribution H(t+k), a prediction model combining the spatiotemporal dynamic characteristics is designed.

[0139]

[0140] Model design: Includes the following modules:

[0141] Temporal feature extraction: Model the dynamic changes within the time window ΔT and extract the change patterns in the key time periods.

[0142] Inter-floor correlation modeling: By capturing the spatial correlation between floors (such as the linkage relationship between adjacent floors), the traffic trend of hot spots can be predicted.

[0143] Attention mechanism: Dynamically assign weights to highlight the features of high-traffic floors.

[0144] Furthermore, in order to address special issues in patent scenarios (such as sudden changes in hot spots and uneven distribution of floor traffic), two targeted regularization items were designed during the optimization process:

[0145] Local smoothing regularization: ensures continuity of thermal distribution between adjacent floors and reduces unreasonable prediction mutations:

[0146]

[0147] Represents the physical proximity weight of floors i and j. The closer the distance, the greater the weight.

[0148] β controls the regularization strength.

[0149] Dynamic weighting term: In order to highlight the importance of high-traffic floors, a dynamic weighting term is added to the model's loss function:

[0150]

[0151] ρ i is the flow weight of floor i, which is dynamically adjusted to the normalized value of the past flow mean, so that the impact of high-demand floors on optimization is more significant.

[0152] Furthermore, considering the prediction error and the above regularization term, the optimization objective function is constructed

[0153]

[0154] The first term is the prediction error, which measures the gap between the model prediction and the actual traffic.

[0155] The second term is local smoothing regularization, which ensures the continuity of flow distribution between adjacent floors.

[0156] The third term is a dynamic weighting term, which makes the optimization pay more attention to the prediction accuracy of high-traffic floors.

[0157] Furthermore, the trained model Input D seq After that, the flow thermal distribution H(t+k) within K time steps in the future period is output:

[0158] The dimension is (K, m), where K is the number of prediction time steps and m is the number of floors.

[0159] Each H i (t+k) represents the traffic intensity of floor i at the future time t+k.

[0160] Furthermore, the output is: future traffic thermal distribution H(t+k), which is used to identify high-demand floors and time periods.

[0161] Function: Provide dynamic input for step S3 (scheduling optimization), support efficient scheduling decisions, reduce waiting time and improve elevator resource utilization.

[0162] Furthermore, through the above design, the prediction model of this step combines the spatiotemporal dynamic characteristics and introduces innovative regularization terms for the special scenario of floor traffic distribution. Local smooth regularization and dynamic weighting terms can effectively improve the prediction smoothness between floors and the optimization focus on high-traffic floors.

[0163] S3. Based on the characteristic time series matrix and the thermal distribution of traffic within K time steps in the future period, a scheduling strategy based on multi-objective reinforcement learning is designed, including the target floor allocation and dynamic operation path of the elevator.

[0164] Specifically, this step receives the future flow thermal distribution H(t+k) output from step 2 and the dynamic feature matrix D(t) generated in step 1, and combines the two to provide input for scheduling optimization:

[0165] H(t+k): Floor flow thermal distribution matrix in the next K time steps, each H i (t+k) represents the predicted traffic intensity of floor i.

[0166] D(t): Contains the real-time status information of the elevator (such as position, load, direction, etc.), providing support for real-time decision-making.

[0167] Furthermore, this step aims to optimize the elevator dispatching strategy through the reinforcement learning model to solve the problems of insufficient resource allocation during peak periods and energy waste during trough periods.

[0168] Furthermore, in order to achieve scheduling optimization, the core elements of the reinforcement learning model are designed:

[0169] State: system state vector s t =[H(t+k),D(t)], combining the future traffic prediction and the real-time status of the elevator to comprehensively describe the current operating environment of the system.

[0170] Action: Scheduling decision vector a t ,include:

[0171] Destination floor assignment for the elevator.

[0172] Elevator start and stop control (running or sleeping).

[0173] Reward function: Combined with multi-objective optimization, the following indicators are integrated:

[0174] R t =-α·WaitTime-β·Energy-γ·Imbalance+δ·Priority

[0175] WaitTime: The average waiting time of all passengers in the system.

[0176] Energy: total energy consumption, including energy consumption of elevator operation and no-load operation.

[0177] Imbalance: The difference in load balance between elevators, calculated as the load standard deviation.

[0178] Priority: The priority response level of high-traffic floors, encouraging requests from hot floors to be processed promptly.

[0179] α, β, γ, δ: weight coefficients that control the importance of each target.

[0180] Furthermore, a multi-objective reinforcement learning model based on policy optimization is used , directly learn the optimal strategy π * , the state s t Mapped to action a t :

[0181]

[0182] π: Mapping strategy from state to action.

[0183] π * : The optimal strategy.

[0184] γ: Discount factor, controlling the importance of future rewards.

[0185] Understandably, model design highlights:

[0186] Temporal and spatial feature fusion: Model input s t Including future traffic and real-time status, the spatiotemporal interaction features are extracted through neural networks.

[0187] Dynamic weight adjustment: Dynamically adjust the weight coefficient in the reward function for different time periods (such as peak periods and trough periods) to adapt to real-time needs.

[0188] Furthermore, in the decision-making process, H(t+k) is combined to design dynamic partitioning and priority scheduling strategies:

[0189] Dynamic zoning: According to the predicted traffic distribution H(t+k), the service area of ​​the elevator is dynamically adjusted, and high-traffic floors are preferentially allocated to exclusive elevators to reduce the no-load energy consumption of cross-zone operation.

[0190] Priority scheduling: Within a zone, high-traffic floors are given higher priority to ensure that requests from hot spots are responded to quickly. i Calculated as:

[0191] P i =H i (t+k)·ρ i

[0192] Among them, ρ i It is the inverse proportional weight of the historical response delay. The longer the delay, the higher the priority.

[0193] Furthermore, after the reinforcement learning model is trained, it generates a scheduling strategy in real time. t ,include:

[0194] Target floor path planning for each elevator ensures optimal allocation.

[0195] Dynamic start-stop control determines whether the elevator is enabled or disabled based on prediction and current status to save energy.

[0196] Furthermore, the output is: a scheduling optimization strategy A(t), including the target floor allocation and dynamic operation path of the elevator.

[0197] Function: The optimization strategy significantly reduces the waiting time, improves the response efficiency of hotspot floors, and reduces the energy consumption of no-load operation during the valley period, providing a direct decision-making basis for step S4 (energy consumption optimization and execution control).

[0198] Through the above design, this step introduces innovative priority weights and dynamic partitioning mechanisms, and combines the reinforcement learning framework to optimize the multi-objective scheduling strategy to accurately respond to the peak and valley operation challenges of building elevators.

[0199] S4. According to the scheduling strategy and characteristic timing matrix and combined with the energy consumption state matrix, an energy consumption optimized execution control strategy is formulated to minimize the global energy consumption and generate optimized control instructions.

[0200] Specifically, this step receives the scheduling strategy A(t) generated in step 3 and the dynamic feature matrix D(t) in step 1, and formulates an energy-optimized execution control strategy in combination with the energy consumption state matrix E(t):

[0201] A(t): The optimized scheduling strategy, including the target floor and path planning of each elevator, as well as the start and stop status.

[0202] D(t): Real-time status information of the elevator (such as position, load, direction, etc.), providing dynamic system characteristics.

[0203] E(t): Energy consumption state matrix, including operating power consumption, no-load energy consumption of the elevator, and braking energy recovery.

[0204] The goal of this step is to optimize the elevator energy consumption and generate control instructions in combination with the existing scheduling strategy.

[0205] Furthermore, in order to minimize the global energy consumption, an energy consumption optimization model ε is constructed, and the optimization objective is:

[0206] ε=TotalEnergy-λ·EnergyRecovery

[0207] TotalEnergy: The total energy consumption of the system, including the elevator operation power consumption and no-load operation power consumption:

[0208]

[0209] and Respectively represent the unit power consumption of the elevator when running and when not loaded.

[0210] and is the running and idle time of elevator k.

[0211] EnergyRecovery: The total energy recovered by the system through braking energy.

[0212] λ: Effective utilization coefficient of recovered energy, used to quantify the contribution of energy recovery to the system.

[0213] Furthermore, in order to reduce the no-load energy consumption during the off-peak period, a dynamic start-stop mechanism is proposed:

[0214] Sleep determination: Based on the predicted flow H(t+k) and real-time status D(t), the idle elevators are determined to be in sleep mode. The sleep decision is defined as:

[0215]

[0216] ∈: Traffic threshold, which controls the sensitivity of sleep determination.

[0217] T threshold : Minimum idle time threshold, used to avoid frequent starts and stops.

[0218] Wake-up mechanism: When the traffic in a certain area exceeds the dynamic threshold δ, the adjacent dormant elevators are woken up first to reduce the waiting time during high demand.

[0219] Furthermore, an energy recovery distribution model is proposed to use the braking energy for elevator operation or other building equipment:

[0220]

[0221] The recovered energy generated by elevator k braking.

[0222] η: Energy recovery efficiency, used to control the distribution ratio of recovered energy.

[0223] Furthermore, the scheduling strategy A(t) and the energy consumption optimization model ε are integrated to generate the optimized control instruction C exec (t), including:

[0224] Elevator target floor and path planning.

[0225] Dynamic start-stop control instructions, including sleep and wake-up decisions.

[0226] Develop energy allocation plans to optimize energy consumption and utilization.

[0227] Further, output: Execute control instruction C exec (t), including specific implementation plans for scheduling optimization and energy consumption optimization strategies.

[0228] Function: Through dynamic start-stop and energy recovery mechanisms, the total energy consumption of the system is reduced, while ensuring response efficiency during high-demand periods, providing support for energy conservation and efficient operation of buildings.

[0229] Through the above design, this step further optimizes energy consumption based on the scheduling strategy, combines dynamic start-stop and brake energy recovery mechanisms, reduces no-load energy consumption and makes full use of energy recovery resources.

[0230] In another embodiment of the present invention, an intelligent building elevator traffic dispatching system for energy saving optimization is provided, the system comprising:

[0231] A real-time data acquisition unit is used to collect real-time data from building sensor equipment and elevator monitoring systems and perform preprocessing. Based on the preprocessed data, feature extraction based on a time window is performed to obtain a weighted feature matrix. The physical proximity between floors is used to construct an associated regular term to enhance the floor association features of the weighted feature matrix. The weighted feature matrix is ​​converted into a time series feature matrix based on a time window to capture historical trends, thereby obtaining a feature time series matrix.

[0232] The flow prediction unit is used to build a prediction model based on the characteristic time series matrix to obtain the flow thermal distribution of future floors. At the same time, the optimization objective function is built based on the predicted value of the prediction model and the actual prediction error to train the prediction model. The characteristic time series matrix is ​​used as the input of the trained prediction model to output the flow thermal distribution within K time steps in the future period, which is used to identify high-demand floors and time periods.

[0233] The scheduling strategy generation unit is used to design a scheduling strategy based on multi-objective reinforcement learning based on the characteristic time series matrix and the flow thermal distribution within K time steps in the future period, including the target floor allocation and dynamic operation path of the elevator;

[0234] The reinforcement learning model based on multi-objective reinforcement learning is designed as follows:

[0235] State: Construct the state vector s based on the characteristic time series matrix and the flow thermal distribution in K time steps in the future period t=[H(t+k),D(t)];

[0236] Action: Scheduling decision vector a t , including elevator target floor allocation and elevator start and stop control;

[0237] Reward function:

[0238] R t =-α·WaitTime-β·Energy-γ·Imbalance+δ·Priority

[0239] Among them, WaitTime represents the average waiting time of all passengers in the system; Energy represents the total energy consumption, including the energy consumption of elevator operation and no-load operation; Imbalance represents the difference in load balance between elevators, calculated as the load standard deviation; Priority represents the priority response degree of high-traffic floors, encouraging requests from hot floors to be processed in a timely manner; α, β, γ, δ represent weight coefficients, controlling the importance of each target;

[0240] Using a multi-objective reinforcement learning model based on policy optimization Directly learn the optimal policy π * , the state s t Mapped to action a t :

[0241]

[0242] Among them, π represents the mapping strategy from state to action; π * represents the optimal strategy; γ represents the discount factor, which controls the importance of future rewards;

[0243] The scheduling strategy optimization unit is used to formulate an energy-optimized execution control strategy based on the scheduling strategy and the characteristic timing matrix in combination with the energy consumption state matrix to minimize the global energy consumption and generate optimized control instructions.

[0244] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0245] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a division of logical functions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0246] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0247] Although embodiments of the present invention have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A smart building elevator traffic scheduling method for energy saving optimization, characterized in that: The method comprises: S1. Collect real-time data from building sensor equipment and elevator monitoring systems and perform preprocessing. Perform feature extraction based on time windows based on the preprocessed data to obtain a weighted feature matrix. Use the physical proximity between floors to construct an association regularization term to enhance the floor association features of the weighted feature matrix. Convert the weighted feature matrix into a time series feature matrix based on a time window to capture historical trends and obtain a feature time series matrix. S2. Build a prediction model based on the characteristic time series matrix to obtain the flow and thermal distribution of future floors. At the same time, build an optimization objective function based on the predicted value of the prediction model and the actual prediction error to train the prediction model. Use the characteristic time series matrix as the input of the trained prediction model to output the flow and thermal distribution within K time steps in the future period to identify high-demand floors and time periods. S3. Based on the characteristic time series matrix and the flow thermal distribution within K time steps in the future period, a scheduling strategy based on multi-objective reinforcement learning is designed, including the target floor allocation and dynamic operation path of the elevator; The reinforcement learning model based on multi-objective reinforcement learning is designed as follows: State: Construct the state vector s based on the characteristic time series matrix and the flow thermal distribution in K time steps in the future period t =[H(t+k),D(t)]; Action: Scheduling decision vector a t , including elevator target floor allocation and elevator start and stop control; Reward function: R t =-a·WaitTime-b·Energy-c·Imbalance+d·Priority Among them, WaitTime represents the average waiting time of all passengers in the system; Energy represents the total energy consumption, including the energy consumption of elevator operation and no-load operation; Imbalance represents the difference in load balance between elevators, calculated as the load standard deviation; Priority represents the priority response degree of high-traffic floors, encouraging requests from hot floors to be processed in a timely manner; α, β, γ, δ represent weight coefficients, controlling the importance of each target; Using a multi-objective reinforcement learning model based on policy optimization Directly learn the optimal policy π * , the state s t Mapped to action a t : Among them, π represents the mapping strategy from state to action; π * represents the optimal strategy; γ represents the discount factor, which controls the importance of future rewards; S4. According to the scheduling strategy and characteristic timing matrix and combined with the energy consumption state matrix, an energy consumption optimized execution control strategy is formulated to minimize the global energy consumption and generate optimized control instructions.

2. The intelligent building elevator traffic dispatching method for energy saving optimization according to claim 1 is characterized in that: The real-time data includes: Floor call matrix C(t): C(t)[i,j] represents the call intensity from floor i to floor j at time t; Elevator state matrix S(t): describes the real-time position, load and direction of each elevator, S k (t) = [pos k ,load k ,dir k ],in: POS k Indicates the current floor of elevator k; load k Indicates the load ratio; dir k Indicates the running direction, with values ​​of {-1,0,1}, representing downlink, stationary, and uplink respectively; Temporal feature T(t): represents context information; The real-time data is combined into the basic feature matrix D(t0, which expresses the state of the building elevator system at the current moment: D(t)=[C(t),S(t),T(t)]; The preprocessing comprises: Designing a dynamic anomaly detection model For outliers that may appear in elevator and floor call data, historical expected values ​​are used to eliminate outliers, expressed as: in, It is the dynamic expected value calculated based on historical data. ∥·∥2 represents the binary norm, which is used to measure the degree of deviation of the feature matrix.

3. The intelligent building elevator traffic dispatching method for energy saving optimization according to claim 2 is characterized in that: The associated regularization term is constructed as follows: in, represents the physical proximity weight of floors i and j; is the weight matrix of floor i, is the weight matrix of floor j, and α is the weight coefficient of the regularization term.

4. The intelligent building elevator traffic dispatching method for energy saving optimization according to claim 1 is characterized in that: The prediction model P has the structure: H(t+k)=P(D seq ) Among them, H(t+k) is the predicted flow thermal distribution of the future floor, D seq is the characteristic time series matrix; The optimization objective function also includes local smoothing regularization and dynamic weighting terms: The local smoothing regularization Ensure the continuity of thermal distribution between adjacent floors and reduce unreasonable prediction mutations, which can be expressed as: Among them, W i,j represents the physical proximity weight of floors i and j; β controls the regularization strength; The dynamic weighting term Used to highlight the importance of high traffic floors, expressed as: Among them, ρ i is the flow weight of floor i, which is dynamically adjusted to the normalized value of the past flow mean, so that the highest demand floor has the most significant impact on the optimization; H true,i (t+k) is the real flow, H i (t+k) is the predicted flow rate.

5. The intelligent building elevator traffic dispatching method for energy saving optimization according to claim 4 is characterized in that: The optimization objective function It is expressed as: Item 1 is the prediction error, which measures the difference between the model prediction H(t+k) and the actual flow H true The gap of (t+k); The second term is the local smoothing regularization Ensure the continuity of traffic distribution between adjacent floors; The third item is the dynamic weighted item Make the optimization more focused on prediction accuracy on high traffic floors.

6. The intelligent building elevator traffic dispatching method for energy saving optimization according to claim 1 is characterized in that: The S3 further includes: In the decision-making process, the dynamic partitioning and priority scheduling strategies are designed in combination with the predicted future floor flow and thermal distribution H(t+k): The dynamic partitioning: according to the predicted future floor flow thermal distribution H(t+k), the service area of ​​the elevator is dynamically adjusted, and high-flow floors are preferentially allocated to exclusive elevators, thereby reducing the no-load energy consumption of cross-zone operation; The priority scheduling strategy: within the partition, give higher priority to the high-traffic floors to ensure quick response to requests from hot floors. i Calculated as: P i =H i (t+k)·ρ i Among them, ρ i It is the inverse proportional weight of the historical response delay. The longer the delay, the higher the priority.

7. The intelligent building elevator traffic dispatching method for energy saving optimization according to claim 1 is characterized in that: The energy consumption state matrix includes operating power consumption, no-load energy consumption of the elevator, and braking energy recovery.

8. The intelligent building elevator traffic dispatching method for energy saving optimization according to claim 1 is characterized in that: The S4 specifically includes: S41. Construct an energy consumption optimization model to minimize global energy consumption; S42. Based on the energy consumption optimization model, a dynamic start-stop mechanism is proposed to reduce the no-load energy consumption during the valley period, including: Based on the predicted future floor flow thermal distribution and characteristic time series matrix, the idle elevator is judged to be dormant, and the dormant decision is defined as: Among them, ∈ represents the flow threshold, which controls the sensitivity of dormancy judgment; T threshold Indicates the minimum idle time threshold, used to avoid frequent starts and stops; When the traffic in a certain area exceeds the dynamic threshold δ, the neighboring dormant elevators are woken up first to reduce the waiting time during high demand; S43. Design an energy recovery distribution model to use the braking energy for elevator operation or other building equipment; S44, generating optimized control instructions by combining the scheduling strategy and the energy consumption optimization model, including: Elevator target floor and path planning; Dynamic start-stop control instructions, including sleep and wake-up decisions; Develop energy allocation plans to optimize energy consumption and utilization.

9. The intelligent building elevator traffic dispatching method for energy saving optimization according to claim 8 is characterized in that: The energy consumption optimization model is expressed as: ε=TotalEnergy-λ·EnergyRecovery Among them, TotalEnergy represents the total energy consumption of the system, including the elevator operation power consumption and no-load operation power consumption, and ε is the flow threshold: in, and Respectively represent the unit power consumption of the elevator in operation and without load; and is the running and no-load time of elevator k; EnergyRecovery represents the total energy recovered by the system through braking energy; λ: the effective utilization coefficient of recovered energy, which is used to quantify the contribution of energy recovery to the system; The energy recovery allocation model is expressed as: Among them, EnergyRedistribution is the total recovered energy, It represents the recovered energy generated by elevator k braking; η represents the energy recovery efficiency, which is used to control the distribution ratio of recovered energy.

10. An intelligent building elevator traffic dispatching system for energy saving optimization, characterized in that: The system comprises: A real-time data acquisition unit is used to collect real-time data from building sensor equipment and elevator monitoring systems and perform preprocessing. Based on the preprocessed data, feature extraction based on a time window is performed to obtain a weighted feature matrix. The physical proximity between floors is used to construct an associated regular term to enhance the floor association features of the weighted feature matrix. The weighted feature matrix is ​​converted into a time series feature matrix based on a time window to capture historical trends, thereby obtaining a feature time series matrix. The flow prediction unit is used to build a prediction model based on the characteristic time series matrix to obtain the flow thermal distribution of future floors. At the same time, the optimization objective function is built based on the predicted value of the prediction model and the actual prediction error to train the prediction model. The characteristic time series matrix is ​​used as the input of the trained prediction model to output the flow thermal distribution within K time steps in the future period, which is used to identify high-demand floors and time periods. The scheduling strategy generation unit is used to design a scheduling strategy based on multi-objective reinforcement learning based on the characteristic time series matrix and the flow thermal distribution within K time steps in the future period, including the target floor allocation and dynamic operation path of the elevator; The reinforcement learning model based on multi-objective reinforcement learning is designed as follows: State: Construct the state vector s based on the characteristic time series matrix and the flow thermal distribution in K time steps in the future period t =[H(t+k),D(t)]; Action: Scheduling decision vector a t , including elevator target floor allocation and elevator start and stop control; Reward function: R t =-a·WaitTime-b·Energy-c·Imbalance+d·Priority Among them, WaitTime represents the average waiting time of all passengers in the system; Energy represents the total energy consumption, including the energy consumption of elevator operation and no-load operation; Imbalance represents the difference in load balance between elevators, calculated as the load standard deviation; Priority represents the priority response degree of high-traffic floors, encouraging requests from hot floors to be processed in a timely manner; α, β, γ, δ represent weight coefficients, controlling the importance of each target; Using a multi-objective reinforcement learning model based on policy optimization Directly learn the optimal policy π * , the state s t Mapped to action a t : Among them, π represents the mapping strategy from state to action; π * represents the optimal strategy; γ represents the discount factor, which controls the importance of future rewards; The scheduling strategy optimization unit is used to formulate an energy-optimized execution control strategy based on the scheduling strategy and the characteristic timing matrix in combination with the energy consumption state matrix to minimize the global energy consumption and generate optimized control instructions.

Images