Multi-split air conditioner cluster intelligent scheduling algorithm based on building load prediction
By using a hybrid model combining LSTM and XGBoost for load forecasting and dynamic scheduling of multi-split air conditioning systems, the shortcomings of multi-split air conditioning systems in load forecasting and scheduling are addressed, achieving efficient, stable, and highly adaptable energy management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHONGSHAN CHANGHONG ELECTRIC
- Filing Date
- 2025-12-31
- Publication Date
- 2026-05-19
AI Technical Summary
Existing multi-split air conditioning systems are one-sided in load forecasting and lack forward-looking scheduling, resulting in serious energy waste. They also lack global optimization and adaptability, and cannot cope with the problems of fluctuating population density and regional functional differences in commercial buildings.
A building load forecasting algorithm based on a hybrid LSTM and XGBoost model is adopted. It combines multi-dimensional data to forecast load, dynamically allocate equipment load, coordinate equipment start-up and shutdown timing, and achieve optimal energy consumption through closed-loop optimization.
It achieves high-precision load forecasting, reduces energy consumption by 18%, improves operational stability and adaptability, and meets the scheduling needs of complex scenarios in commercial buildings.
Abstract
Description
Technical Field
[0001] This invention relates to the field of air conditioning cluster control technology, specifically to an intelligent scheduling algorithm for multi-split air conditioning clusters based on building load prediction. Background Technology
[0002] Commercial building multi-split air conditioning systems consist of multiple indoor and outdoor units, covering various areas such as offices, conference rooms, and exhibition halls. Their operational efficiency directly impacts the building's overall energy consumption. Existing technologies suffer from the following core deficiencies: Load forecasting is one-sided and dispatching lacks foresight: Traditional dispatching relies solely on real-time load data without taking into account environmental parameters, personnel flow, and historical load patterns, resulting in delayed dispatching decisions—equipment operates under overload during high-load periods and many devices idle during low-load periods, leading to serious energy waste (more than 25% higher than reasonable dispatching). The scheduling mode is fragmented and lacks global optimization: it often adopts independent control of a single device or simple linkage of partitions without considering the overall energy consumption of the cluster. For example, new devices are blindly started in high-load areas instead of prioritizing scheduling devices with high energy efficiency, resulting in unbalanced resource allocation. Poor adaptability and inability to cope with dynamic scenarios: Commercial buildings have characteristics such as large fluctuations in personnel density and differences in regional functions (such as temporary high load in meeting rooms and continuous medium load in office areas), which make it difficult for traditional fixed scheduling strategies to adapt, further aggravating energy waste and equipment wear and tear.
[0003] While existing technologies include load forecasting or energy-saving algorithms for individual air conditioners, a global intelligent scheduling solution that integrates multi-dimensional data for multi-split air conditioner clusters has yet to emerge. Therefore, this invention proposes an intelligent scheduling algorithm based on building load forecasting to address the aforementioned technical shortcomings. Summary of the Invention
[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting, which solves the aforementioned problems.
[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: an intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting, comprising the following steps: S1. Collect building environment parameters, personnel density, equipment operating parameters and historical load data, and perform preprocessing; S2. Train a building load forecasting model based on a hybrid LSTM and XGBoost model, and output the load forecast values for the next 1-4 hours. S3. Based on the prediction results and the equipment energy efficiency ratio and health status, dynamically allocate the load of each area to the cluster equipment; S4. Coordinate the start-up and shutdown sequence of equipment to avoid simultaneous high-frequency start-up or shutdown. S5. Optimize scheduling strategies based on closed-loop operation feedback data to achieve optimal energy consumption.
[0006] Preferably, the parameters collected in step S1 include indoor and outdoor temperature and humidity, light intensity, personnel density, equipment operating status, energy consumption data and historical load data. The environmental parameter collection frequency is 5 times / minute, and the equipment parameter collection frequency is 10 times / minute.
[0007] Preferably, in step S1, the data preprocessing uses a moving average filtering algorithm to remove abnormal data, and the data normalization process unifies the parameters of different dimensions to the 0-1 range.
[0008] Preferably, the load forecasting model described in step S2 has a forecasting accuracy of ≥88% and a short-term forecasting error of ≤5%.
[0009] Preferably, the dynamic load allocation in step S3 includes: equipment priority sorting, regional load adaptation and load migration optimization; wherein, high-load areas prioritize the coordinated operation of efficient equipment, and low-load areas reduce the frequency of equipment to 40%-60% of the rated frequency.
[0010] Preferably, the start-stop timing coordination in step S4 includes: the start interval between adjacent devices is ≥30 seconds, and the frequency is gradually increased from 5Hz to the target frequency during start-up; during shutdown, low-energy-efficiency devices are shut down first, and then the frequency is gradually reduced.
[0011] Preferably, the closed-loop optimization in step S5 aims to reduce the total energy consumption of the cluster by ≥18% compared with traditional distributed control, while maintaining regional temperature fluctuations ≤±1℃; the model training data is automatically updated every morning to optimize feature weights.
[0012] Preferably, the algorithm is adaptable to multiple scenarios such as commercial building office areas, conference rooms, and exhibition halls, and supports flexible division and dynamic adaptation of load units.
[0013] (III) Beneficial Effects Compared with existing technologies, this invention provides an intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting, which has the following advantages: 1. This intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting provides high-precision load forecasting through multi-dimensional data fusion: breaking through the limitations of traditional single-parameter forecasting, it integrates environmental, personnel, and historical data to build a hybrid AI model with a prediction accuracy of ≥88%, providing forward-looking support for scheduling decisions and solving the pain point of "lag" in traditional scheduling.
[0014] 2. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting has a global optimization scheduling logic for the cluster: abandoning the independent control mode of single equipment, with the goal of minimizing the total energy consumption of the cluster, it achieves optimal resource allocation through equipment priority sorting, dynamic load migration, and start-stop timing coordination, reducing energy consumption by ≥18% compared to traditional distributed control.
[0015] 3. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting dynamically adapts to complex scenarios in commercial buildings: The algorithm supports flexible division of load units and can adapt to the load fluctuation characteristics of different functional areas such as office areas, conference rooms, and exhibition halls, solving the problem of poor scheduling adaptability caused by large personnel flow and complex load changes in commercial buildings. Detailed Implementation
[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0017] A smart scheduling algorithm for multi-split air conditioning clusters based on building load forecasting includes the following steps: S1. Multi-dimensional data acquisition and preprocessing: Data Collection Parameters: Four core data types are collected through a sensor network deployed in various areas of the commercial building: Environmental parameters: indoor and outdoor temperature and humidity, light intensity (collection frequency 5 times / minute, accuracy ±0.3℃, ±5%RH); Personnel parameters: The personnel density of each area is obtained through infrared sensors or video analysis (quantified as 0-100%, with a sampling frequency of 1 time / minute). Equipment parameters: Operating status of each air conditioner (start / stop, frequency, energy efficiency ratio), energy consumption data (current, voltage, power, collected 10 times / minute); Historical data: Load data, environmental data, and population flow patterns for the past 3 months (stored categorized by weekdays / weekends and peak / off-peak periods); Data preprocessing: The moving average filtering algorithm is used to remove abnormal data (such as sudden changes in instantaneous population density and sensor failure data). Through data normalization, the parameters of different dimensions are unified to the 0-1 range to provide standardized input for the load forecasting model. S2. Training of a multi-dimensional fusion building load prediction model: Constructing an integrated AI load prediction model encompassing feature extraction, model training, and accuracy optimization, specifically including: Feature engineering: Extract core features from the preprocessed data, including 12 key features such as "difference between ambient temperature and humidity, light intensity, population density, time period characteristics (such as peak from 9:00 to 11:00), and historical load during the same period"; Model architecture: A hybrid model combining LSTM (Long Short-Term Memory Network) and XGBoost is adopted. LSTM captures the load change patterns in the time series (such as the time period when the daily load peak occurs), and XGBoost strengthens the nonlinear correlation of multiple features (such as the quantitative relationship between personnel density and load). Model training and optimization: Using 3 months of historical data as the training set, the model hyperparameters (learning rate, number of hidden layer nodes) were adjusted through cross-validation to ensure that the model's prediction accuracy for building load in the next 1-4 hours is ≥88%, and the prediction error for short periods (within 1 hour) is ≤5%. S3, Cluster Global Dynamic Load Allocation Strategy: Based on load forecasting results, and with the goal of "minimizing total energy consumption of the cluster and balancing equipment load," the operating tasks of each air conditioning unit are dynamically allocated: Equipment Priority Ranking: The current energy efficiency ratio (COP) of each air conditioner is calculated in real time, and combined with the equipment health status (such as running time and fault records), a priority list is generated—equipment with high energy efficiency ratio and good health status takes the lead in undertaking high loads; Regional load adaptation: The building is divided into several load units (e.g., each office is a unit). High load units (load ≥ 70%) prioritize the coordinated operation of multiple high-efficiency devices, while low load units (load ≤ 30%) only retain 1-2 devices to operate at reduced frequency (frequency reduced to 40%-60% of the rated frequency) to avoid inefficient idling of multiple devices. Load migration optimization: When the load in a certain area drops suddenly (such as when people leave a conference room), the algorithm automatically migrates the remaining load of the air conditioner in that area to the high-efficiency equipment in the surrounding high-load area, while shutting down redundant equipment to reduce ineffective energy consumption; S4. Equipment start-up and shutdown sequence coordination mechanism: To avoid grid impact and energy waste caused by multiple devices starting or stopping at high frequencies simultaneously, a "staggered start / stop - smooth load transition" strategy is designed: Startup coordination: During high-load periods (such as the morning rush hour at 9:00), start up the equipment in sequence according to its priority, with an interval of ≥30 seconds between the start-up of adjacent equipment, and gradually increase the frequency from low frequency (5Hz) to the target frequency during startup to reduce startup current loss; Shutdown coordination: During low-load periods (such as 22:00 at night), first shut down low-efficiency equipment, and then gradually reduce the load on high-efficiency equipment to avoid regional temperature fluctuations caused by sudden shutdowns. At the same time, reduce the number of equipment start-ups and shutdowns (≤2 times / unit per day) to extend equipment life. S5, Energy Consumption Closed-Loop Optimization: The algorithm collects real-time cluster operation data (total energy consumption, energy consumption per device, regional temperature deviation) and compares it with the preset energy-saving target (energy consumption reduction of ≥18% compared to traditional distributed control): If energy consumption exceeds the target value, the load allocation strategy will be dynamically adjusted (e.g., further increasing the load ratio of high-efficiency equipment). If the temperature deviation in the area is ≥±1℃, fine-tune the operating parameters of the corresponding equipment to ensure a balance between comfort and energy saving; The model training data is automatically updated every day at midnight to optimize feature weights and improve the adaptability of load forecasting and scheduling.
[0018] Example 1: Hardware compatibility requirements: Sensing Units: Deploy temperature and humidity sensors, light sensors, and personnel density sensors to cover all load units in the building and ensure comprehensive data collection; AI computing unit: Deploys an embedded AI chip, supports LSTM and XGBoost hybrid model computing, with a computing latency of ≤200ms, meeting real-time scheduling requirements; Communication unit: Based on IoT communication protocols (such as MQTT), it enables interconnection and interoperability of sensors, AI chips, and air conditioner controllers, with a communication latency of ≤50ms, ensuring fast command transmission; Control Unit: The main control module of the multi-split air conditioner supports receiving scheduling commands output by the AI chip, enabling precise execution of frequency adjustment, start / stop control, and load distribution.
[0019] Software operation process: Initialization phase: Load historical load database, equipment parameters (energy efficiency ratio, rated load), building area division information, and complete model initialization and sensor calibration; Data acquisition and preprocessing stage: Collect multi-dimensional data at a set frequency, and input the data into the load forecasting model after filtering and normalization. Load forecasting phase: The AI chip runs a hybrid model and outputs load forecast values for each region for the next 1-4 hours; Scheduling decision-making phase: Based on the prediction results, generate a list of equipment priorities, a load allocation scheme, and a start-up and shutdown sequence plan; Command execution and feedback phase: Send scheduling commands to the air conditioning controller, collect operational feedback data in real time, and adjust parameters through the closed-loop optimization module; Model update phase: The model is updated daily at midnight using the previous day's running data to optimize prediction accuracy and scheduling strategy.
[0020] Typical scenario implementation example: Commercial building office area scene (weekdays 9:00-18:00) Load forecast: The model combines historical data (peak load 80%), real-time population density (90%), and outdoor temperature (32℃) to predict that the load will remain at 75%-85% for the next 2 hours; Dispatch strategy: Prioritize starting the 3 outdoor units with the highest energy efficiency ratio, and distribute the load of high-load areas in the office area (such as open office areas) to these 3 units, while only 1 unit in low-load areas (such as corridors) is kept running at a reduced frequency; Start-up and shutdown coordination: At 12:00, the personnel density drops to 60% and the predicted load drops to 50%. First, shut down one low-efficiency outdoor unit, and then reduce the frequency of the remaining equipment from 50Hz to 35Hz. Implementation results: During this period, the total energy consumption of the cluster was reduced by 22% compared with traditional decentralized control, and the regional temperature was stabilized at 24-26℃, meeting the comfort requirements.
[0021] Commercial building conference room scenario (temporary high load, duration 2 hours) Load forecast: The model detected a sudden increase in meeting room occupancy to 100% (30 people) using personnel density sensors. Based on historical meeting load data, the model predicts the load will increase from 30% to 90%. Dispatch strategy: Temporarily relocate part of the load of two high-efficiency outdoor units in the surrounding office area to the conference room, where they will operate in conjunction with the original outdoor unit to avoid overloading a single device; Start-up and shutdown coordination: After the meeting, the personnel density dropped to 0, the predicted load returned to 30%, and the two temporarily scheduled devices were gradually shut down to restore the original low-load operation of the equipment; Implementation results: During high-load periods, the equipment did not overload, energy consumption was reduced by 19% compared to the traditional "blindly starting new equipment" solution, and the equipment operation stability was improved by 40%.
[0022] Specific implementation examples: Taking a multi-split air conditioning cluster (including 10 outdoor units and 50 indoor units) in a 10-story commercial building (building area of 5000㎡) as an example, the algorithm of this invention was used to conduct a 30-day field test verification, and the results are as follows: Load forecasting accuracy: The average load forecasting accuracy for each region within 30 days reached 89.5%, with a maximum of 92%, meeting the design target of ≥88%. Energy consumption performance: The total energy consumption of the cluster is reduced by 20.3% compared with the traditional distributed control scheme, exceeding the energy saving target of ≥18%; Operational stability: The number of times the equipment is overloaded has been reduced from 12 times / month in the traditional solution to 1 time / month, the number of start-ups and shutdowns has been reduced by an average of 35%, and the equipment failure rate has been reduced by 50%; Comfort: Temperature fluctuation in each area is ≤ ±0.8℃, and the user satisfaction score reaches 90 points (out of 100), which is 15 points higher than the traditional solution.
[0023] The experimental results show that the algorithm of this invention can effectively solve the problems of high energy consumption and disordered scheduling of multi-split air conditioning clusters in commercial buildings, taking into account energy saving, stability and comfort, and adapting to the complex operation scenarios of commercial buildings.
[0024] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A smart scheduling algorithm for multi-split air conditioning clusters based on building load forecasting, characterized in that, Includes the following steps: S1. Collect building environment parameters, personnel density, equipment operating parameters and historical load data, and perform preprocessing; S2. Train a building load forecasting model based on a hybrid LSTM and XGBoost model, and output the load forecast values for the next 1-4 hours. S3. Based on the prediction results and the equipment energy efficiency ratio and health status, dynamically allocate the load of each area to the cluster equipment; S4. Coordinate the start-up and shutdown sequence of equipment to avoid simultaneous high-frequency start-up or shutdown. S5. Optimize scheduling strategies based on closed-loop operation feedback data to achieve optimal energy consumption.
2. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting as described in claim 1, characterized in that: The parameters collected in step S1 include indoor and outdoor temperature and humidity, light intensity, personnel density, equipment operating status, energy consumption data and historical load data. The environmental parameter collection frequency is 5 times / minute, and the equipment parameter collection frequency is 10 times / minute.
3. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting according to claim 1, characterized in that: In step S1, the data preprocessing uses a moving average filtering algorithm to remove outlier data, and data normalization processes unify the parameters of different dimensions to the 0-1 range.
4. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting as described in claim 1, characterized in that: The load forecasting model described in step S2 has a forecasting accuracy of ≥88% and a short-term forecasting error of ≤5%.
5. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting according to claim 1, characterized in that: The dynamic load allocation in step S3 includes: equipment priority sorting, regional load adaptation, and load migration optimization; In high-load areas, high-efficiency equipment is prioritized for coordinated operation, while in low-load areas, equipment operates at 40%-60% of its rated frequency.
6. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting according to claim 1, characterized in that: The start-stop timing coordination mentioned in step S4 includes: the start interval between adjacent devices is ≥30 seconds, and the frequency is gradually increased from 5Hz to the target frequency during start-up; during shutdown, low-energy-efficiency devices are shut down first, and then the frequency is gradually reduced.
7. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting according to claim 1, characterized in that: The closed-loop optimization described in step S5 aims to reduce the total energy consumption of the cluster by ≥18% compared to traditional distributed control, while maintaining regional temperature fluctuations ≤±1℃; the model training data is automatically updated every morning to optimize feature weights.
8. The intelligent scheduling algorithm for multi-split air conditioning clusters based on building load forecasting according to claim 1, characterized in that: The algorithm is adaptable to various scenarios such as commercial building office areas, conference rooms, and exhibition halls, and supports flexible division and dynamic adaptation of load units.