An intelligent scheduling algorithm and system for elevator energy recovery based on multi-source data

By integrating multi-source data and using reinforcement learning mechanisms, the problems of single-objective optimization and insufficient self-learning ability in the elevator energy management system have been solved. This has enabled the optimal balance of multiple objectives and carbon asset accounting, thereby improving the overall efficiency and economic benefits of the elevator energy management system.

CN123121746APending Publication Date: 2026-11-17倪正荣
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611238725.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing elevator energy management systems suffer from problems such as single-objective optimization, lack of multi-source data fusion, insufficient static rule scheduling and self-learning capabilities, leading to shortened energy storage life, deterioration of power grid quality and insufficient carbon asset accounting.

Method used

By employing multi-source data fusion, multi-objective optimization models, and reinforcement learning self-learning mechanisms, and combining elevator operation, energy storage status, grid status, and environmental control data, the system achieves an optimal balance among multiple objectives, including economic benefits, energy storage lifespan, and grid friendliness. It also integrates fault early warning and carbon asset accounting functions.

Benefits of technology

It achieves an optimal balance among multiple objectives, including economic benefits, energy storage lifespan, and grid friendliness, improves dispatch efficiency, reduces fault risk, provides data support for carbon asset development, and enhances the overall energy efficiency and economic benefits of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

This invention discloses an intelligent scheduling algorithm and system for elevator energy recovery based on multi-source data, comprising five steps: multi-source data acquisition, data fusion feature extraction, multi-objective optimization scheduling decision-making, energy allocation execution, and self-learning of scheduling effects. It integrates four types of data sources: elevator operation, energy storage status, grid status, and environmental control data. A multi-objective optimization model is constructed with the goals of maximizing economic benefits, optimizing energy storage lifespan, and grid friendliness. An improved PSO algorithm is used to solve the model, and Q-learning reinforcement learning is employed for continuous optimization. Energy allocation is executed according to four priority levels: grid feedback, energy storage charging, local power supply, and peak-valley arbitrage. Fault early warning and carbon asset accounting functions are integrated. Compared with fixed-rule strategies, it improves energy efficiency by 5% to 10% and extends energy storage lifespan by more than 20%.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent energy management technology, specifically to an intelligent scheduling algorithm and system for elevator energy recovery based on multi-source data fusion and multi-objective optimization algorithms. Background Technology

[0002] Elevator energy recovery systems involve multiple energy flow nodes, including regenerative braking, thermoelectric power generation, hybrid energy storage, and multiple loads. Achieving optimal energy distribution among these nodes to maximize economic benefits, optimize energy storage lifespan, and minimize grid impact is a key challenge for improving the overall energy efficiency of the system.

[0003] Existing energy management methods have the following limitations:

[0004] 1. Single-objective optimization: Most methods only focus on economic benefits without considering energy storage life and grid friendliness, resulting in a shortened life of energy storage systems or a decline in grid power quality.

[0005] 2. Lack of multi-source data fusion: The system does not make full use of multi-source information such as elevator operation status, energy storage status, power grid status and environmental control data, resulting in insufficient basis for dispatching decisions.

[0006] 3. Static rule scheduling: Scheduling strategies based on fixed priority rules cannot adapt to dynamically changing operating conditions and are inefficient.

[0007] 4. Lack of self-learning ability: Unable to continuously optimize scheduling strategies based on operational experience, resulting in declining operational efficiency over the long term.

[0008] 5. Lack of carbon asset accounting: Carbon emission reduction data collection is not included in the energy management function, which cannot support the development of CCER carbon assets. Summary of the Invention

[0009] Purpose of the invention

[0010] The purpose of this invention is to provide an intelligent scheduling algorithm and system for elevator energy recovery based on multi-source data. Through multi-source data fusion, multi-objective optimization model and reinforcement learning self-learning mechanism, it achieves the optimal balance of multiple objectives such as economic benefits, energy storage life and grid friendliness, and integrates fault early warning and carbon asset accounting functions.

[0011] Technical solution

[0012] (The detailed technical solution is the same as described in the claims; repeated descriptions are omitted here. The actual application documents should elaborate on it in full.)

[0013] Beneficial effects

[0014] 1. Multi-objective optimal balance: Simultaneously optimize three objectives: economic benefits, energy storage lifespan, and grid friendliness, with adjustable weights to adapt to different application scenarios.

[0015] 2. Multi-source data fusion: It integrates four types of data sources: elevator operation, energy storage status, power grid status and environmental control data, so that scheduling decisions are comprehensive and accurate.

[0016] 3. Predictive Active Scheduling: Based on elevator operation status prediction, changes in energy flow are anticipated in advance to achieve proactive scheduling, which improves efficiency by 5% to 10% compared to passive response.

[0017] 4. Self-learning and continuous optimization: The Q-learning reinforcement learning algorithm continuously learns online and gradually optimizes the scheduling strategy. The longer the system runs, the higher the efficiency.

[0018] 5. Fault early warning and protection: The fault early warning function based on deviation analysis can detect component degradation in advance and adjust the scheduling strategy to reduce the risk of failure.

[0019] 6. Carbon Asset Data Support: Real-time collection of carbon emission reduction data provides accurate and reliable data support for the development of CCER carbon assets, achieving the dual benefits of "electricity saving + carbon trading". Attached Figure Description

[0020] Figure 1 This is a diagram of the overall architecture of the intelligent scheduling algorithm.

[0021] Figure 2 A flowchart for multi-source data fusion;

[0022] Figure 3 Flowchart for solving a multi-objective optimization model;

[0023] Figure 4 Decision tree for prioritizing energy allocation;

[0024] Figure 5 Diagram of the Q-learning self-learning framework;

[0025] Figure 6 This is a fault early warning logic diagram;

[0026] Figure 7 This is a flowchart for carbon asset data collection and accounting. Detailed Implementation

[0027] Example 1: Standard Building Application

[0028] System configuration: 15kW elevator, 50Wh / 100kW supercapacitor, 4.8kWh / 20kW lithium battery, 300W TEG array, 6 heat pipes + 15kg PCM.

[0029] Algorithm parameters: PSO particle count 50, maximum iterations 100, solution time <500ms. Q-learning learning rate 0.1, discount factor 0.95. Weights w1=0.5, w2=0.3, w3=0.2. Control period 1 second.

[0030] Actual test results: Average daily energy saving of 22 kWh (18 kWh from regenerative braking + 4 kWh from TEG generation), 1.8 kWh more than the fixed-rule scheduling strategy (an 8.9% improvement). Annual lithium battery degradation rate of 2.1% (40% lower than the 3.5% of the fixed strategy). Grid-connected power fluctuation reduced by 25%. Average daily carbon emission reduction of 13.8 kg CO2.

[0031] Example 2: Multi-elevator building networking

[0032] Three elevators operate in a network, with unified scheduling via a cloud platform. When one elevator has surplus braking energy, it prioritizes supplying other elevators in the building or common loads. Actual test results: Overall building energy saving rate increased by 12%, and energy storage utilization rate increased by 18%.

[0033] Example 3: Peak-Valley Arbitrage Model

[0034] In regions implementing peak-valley electricity pricing (peak price 1.2 yuan / kWh, off-peak price 0.3 yuan / kWh), the algorithm charges and stores energy during off-peak hours and releases it back to the grid during peak hours. The average daily arbitrage profit is 6.5 yuan per unit, resulting in an annual increase of 2,372 yuan per unit.

Claims

1. An intelligent scheduling algorithm and system for elevator energy recovery based on multi-source data, characterized in that, Includes the following steps: Step S1: Multi-source data acquisition. Real-time acquisition of four data sources: elevator operation data, energy storage system data, power grid data, and environmental control data. Elevator operation data includes speed, load, direction, and operating status. Energy storage system data includes supercapacitor SOC, lithium battery SOC, and temperature. Power grid data includes grid voltage, frequency, real-time electricity price, and power factor. Environmental control data includes machine room temperature, PCM temperature, and TEG output power. Step S2: Multi-source data fusion and feature extraction. Time synchronization, outlier filtering, and feature extraction are performed on the four types of data to construct a feature vector X = v, m, d, SOCsc, SOCbat, Tbat, Vgrid, fgrid, price, PF, Troom, Tpcm. Step S3: Energy dispatch decision based on a multi-objective optimization model. The optimization objectives are maximizing economic benefits, optimizing energy storage lifetime, and grid friendliness. Constraints include energy storage capacity, power, temperature, and grid connection. The optimal energy allocation strategy is then solved. Step S4: Energy allocation execution. Based on the dispatch decision results from Step S3, the power flow of the braking energy feedback subsystem, the thermal energy generation subsystem, and the energy storage system is controlled to execute energy allocation. Step S5: Dispatch effect evaluation and model self-learning. The dispatch effect is evaluated in real time, and the dispatch strategy parameters are continuously optimized through reinforcement learning algorithms.

2. The algorithm according to claim 1, characterized in that, The objective function of the multi-objective optimization model described in step S3 is: max F = w1 × f_economic + w2 × f_lifetime + w3 × f_grid-friendly, where f_economic is the economic benefit sub-objective, f_economic = pricesell × Esell _ pricebuy × Ebuy, price_sell is the electricity selling price, E_sell is the electricity fed back to the grid, price_buy is the electricity buying price, and E_buy is the electricity obtained from the grid; f_lifetime is the energy storage lifespan sub-objective, f_lifetime = -DODsc × ksc + DODbat × kbat, DOD is the depth of discharge, and k is the lifespan loss coefficient; f_grid_friendly is the grid-friendly sub-objective, f_grid_friendly = -|P_grid - P_grid_avg| / P_rated, P_grid is the grid-connected power fluctuation, and P_grid_avg is the average grid-connected power; w1, w2, and w3 are weighting coefficients, w1 + w2 + w3 = 1. Default values: w1=0.5, w2=0.3, w3=0.

2.

3. The algorithm according to claim 1, characterized in that, The constraints in step S3 include: energy storage capacity constraint: SOC_min ≤ SOCt ≤ SOC_max; power constraint: |P_sct| ≤ P_sc_max and |P_batt| ≤ P_bat_max; temperature constraint: T_batt ≤ T_bat_max; grid connection constraint: |P_gridt| ≤ P_grid_max and THD ≤ 5%; energy balance constraint: P_regen + P_teg + P_grid = P_load + P_charge + P_loss.

4. The algorithm according to claim 1, characterized in that, The energy allocation priority of the energy dispatch strategy described in step S3 is as follows: First priority: Braking energy is directly fed back to the grid (highest efficiency and shortest path), and is executed first when the grid is acceptable and the grid-connected power does not exceed the limit; Second priority: Energy storage system charging, with supercapacitors absorbing instantaneous peak power first and lithium batteries storing power smoothly; Third priority: Local power supply, with TEG generation and energy storage system releasing energy to supply lighting or control cabinet DC loads first; Fourth priority: Charging energy from the grid during periods of low grid electricity price and releasing energy storage to feed back to the grid or supply local loads during periods of high electricity price, thereby achieving peak-valley arbitrage.

5. The algorithm according to claim 1, characterized in that, The multi-objective optimization solution in step S3 adopts an improved particle swarm optimization algorithm (PSO): Particle encoding: each particle represents a set of energy allocation decision variables Psc, Pbat, Pgrid, Ploadlocal; fitness function: F = w1 × f_economic + w2 × f_lifetime + w3 × f_grid_friendly; particle update: v_it+1 = (w1 × r1 × pbesti - xi) + c2 × r2 × (gbest - xi), x_it + 1 = x_it + v_it+1; the inertia weight adopts an adaptive decay strategy: = _max - (max - min) × t / t_max; solution time constraint: the optimization solution time for each scheduling cycle does not exceed 500 milliseconds.

6. The algorithm according to claim 1, characterized in that, The model self-learning described in step S5 adopts the Q-learning reinforcement learning algorithm: State space S: Discrete representation of feature vector X; Action space A: Combination of energy allocation decision variables; Reward function R: R = × economic + × lifetime + × grid_friendly, which is the reward weight; Q value update: Qs, a Qs, a + _lr ×R + discount × maxQ(s', a') − Q(s, a); Learning rate _lr=0.1, discount factor _discount=0.95, -greedy exploration policy=0.1; The algorithm continuously learns online and gradually optimizes the scheduling policy parameters to maximize the long-term cumulative reward.

7. The algorithm according to claim 1, characterized in that, The algorithm also includes a fault warning function: establishing a normal operation model for each component based on historical operating data; and calculating in real time the deviation between the actual operating parameters of each component and the normal model. When the deviation exceeds the warning threshold, a fault warning message is output, and the energy scheduling strategy is adjusted to reduce the load on the faulty component. The warning types include TEG module degradation warning, supercapacitor aging warning, lithium battery capacity degradation warning, and heat pipe performance degradation warning.

8. The algorithm according to claim 1, characterized in that, The algorithm also includes carbon asset data collection and accounting functions: real-time statistics of electricity savings, electricity fed back to the grid, and TEG generation; calculation of carbon emission reduction based on the regional grid emission factor: CO2 = E × EF_grid, where E is the electricity savings and EF_grid is the regional grid emission factor; and generation of carbon emission reduction reports on a daily, monthly, quarterly, and annual basis to provide data support for the development of CCER carbon assets.

9. The algorithm according to claim 1, characterized in that, The algorithm achieves local real-time control through a PLC controller or edge computing gateway, with a control cycle of no more than 1 second; at the same time, it uploads the operating data to the cloud platform through an IoT communication module, and the cloud platform performs big data analysis, multi-device network management, and remote updates of strategy parameters.

10. The algorithm according to claim 1, characterized in that, The algorithm is applicable to elevator energy recovery systems with power ranges from 7.5kW to 30kW. It works in conjunction with the three-in-one system integration architecture, the dual-effect structure design of the single pipe, and the hybrid energy storage power distribution control method. The overall power saving rate of the system is 30% to 50%, and the additional energy efficiency improvement brought by the optimization of the scheduling strategy is 5% to 10%.