Accurate refrigerant flow distribution method and energy-saving control system for multi-split air conditioner system

Abstract

The invention discloses a refrigerant flow accurate distribution method and an energy-saving control system for an air conditioner multi-split system, and the method comprises the steps: a multi-source sensing layer collects system operation parameters, carries out the fusion processing, and outputs system state data to an AI decision layer and a working condition adaptation layer; the AI decision-making layer realizes load advanced prediction and multi-objective optimization solution based on system state data, and outputs initial optimal control parameters to the working condition adaptation layer; the working condition adaptation layer identifies a current operation working condition, adjusts a multi-objective optimization weight, corrects an initial optimal control parameter, and outputs an optimal control parameter adaptive to the current working condition to the execution control layer and the energy-saving collaboration layer; the execution control layer drives an execution mechanism to act based on the optimal control parameters so as to adjust the refrigerant flow, and feeds back actual operation parameters of the execution mechanism to the energy-saving cooperation layer; and the energy-saving collaboration layer monitors the energy consumption of the system in real time, dynamically adjusts the multi-objective optimization weight, feeds back the multi-objective optimization weight to the AI decision-making layer, and carries out feedback correction on a prediction result and an optimization result of the AI decision-making layer at the same time.

Description

Technical Field

[0001] This invention relates to the field of air conditioning multi-split system control technology, specifically to a method for precise refrigerant flow distribution and an energy-saving control system for air conditioning multi-split systems. Background Technology

[0002] Multi-split air conditioning systems, with their structural advantage of connecting multiple indoor units to a single outdoor unit, enable independent temperature and humidity control in different indoor spaces, making them widely used in commercial buildings and residential settings. During the operation of a multi-split system, the rational distribution of refrigerant flow is crucial for ensuring system efficiency and indoor comfort, while energy-saving control is a key requirement under the current trend of low-carbon development. The synergistic optimization of both is the core objective for improving the overall performance of the system.

[0003] Existing multi-split air conditioning systems mostly use a feedback regulation mode based on real-time load for refrigerant flow distribution. This mode has obvious lag, resulting in large fluctuations in indoor temperature and failing to meet users' precise comfort requirements. Meanwhile, its control parameters are mostly fixed settings, which cannot be adapted to different operating conditions such as low temperature heating, high temperature cooling, and partial load. Under extreme or special conditions, it is prone to problems such as low refrigerant distribution efficiency and surge in energy consumption. In addition, most existing technologies only focus on matching refrigerant flow with load, neglecting the overall optimization of system energy consumption. This makes it difficult to achieve a balance between distribution accuracy and energy-saving effect, and fails to meet the current high requirements for efficient and energy-saving air conditioning systems.

[0004] The aforementioned technical issues collectively result in significant deficiencies in the operational accuracy, adaptability to operating conditions, and energy efficiency of existing multi-split air conditioning systems, thus limiting their application experience and promotional value.

[0005] Therefore, there is an urgent need to develop a method for precise refrigerant flow distribution and an energy-saving control system for multi-split air conditioning systems to solve the problems in existing technologies. Summary of the Invention

[0006] The purpose of this invention is to provide a method for precise refrigerant flow allocation and an energy-saving control system for multi-split air conditioning systems. This method enables precise, predictive allocation of refrigerant flow and energy-saving coordinated control under all operating conditions. It is also simple in structure and easy to use, thereby solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for precise refrigerant flow allocation and an energy-saving control system for multi-split air conditioning systems include a multi-source sensing layer, an AI decision-making layer, an operating condition adaptation layer, an execution control layer, and an energy-saving coordination layer. Each layer works in sequence to achieve predictive and precise refrigerant flow allocation and energy-saving control across all operating conditions. The specific coordination logic is as follows: The multi-source sensing layer collects system operating parameters and performs fusion processing, then outputs system status data to the AI ​​decision-making layer and the operating condition adaptation layer; the AI ​​decision-making layer performs load prediction and multi-objective optimization based on the system status data, outputting initial optimal control parameters to the operating condition adaptation layer; the operating condition adaptation layer identifies the current operating condition, adjusts the multi-objective optimization weights based on the identification results, corrects the initial optimal control parameters, and outputs optimal control parameters adapted to the current operating condition to the execution control layer and the energy-saving coordination layer; the execution control layer drives the actuator to adjust the refrigerant flow based on the optimal control parameters and feeds back the actual operating parameters of the actuator to the energy-saving coordination layer; the energy-saving coordination layer monitors system energy consumption in real time, dynamically adjusts the multi-objective optimization weights and feeds them back to the AI ​​decision-making layer, while also providing feedback correction to the AI ​​decision-making layer's prediction and optimization results.

[0008] By adopting the above technical solutions, the dual objectives of predictive and precise allocation of refrigerant flow and energy-saving control under all operating conditions are achieved. Through the coherent input and output logic between each layer, the continuity and reliability of the control process are ensured, effectively improving the accuracy and energy efficiency of the multi-split air conditioning system.

[0009] As a further aspect of the present invention: the multi-source sensing layer includes a physical sensing unit and a data fusion unit. The physical sensing unit acquires core parameters by combining physical sensing with virtual sensing. The core parameters include physical sensing parameters and virtual sensing parameters. The data fusion unit has a built-in filtering algorithm to preprocess the raw parameters acquired by the physical sensing unit to remove noise interference and output accurate system status data.

[0010] By adopting the above technical solution, and through the full-dimensional parameter acquisition of physical sensing combined with virtual sensing, along with noise removal processing of filtering algorithms, accurate acquisition of system state data is achieved, providing high-quality data support for subsequent decision-making and control at each level, and ensuring the effectiveness of subsequent control logic.

[0011] As a further aspect of the present invention: the physical sensing parameters include indoor unit return air temperature, air outlet temperature, indoor relative humidity, outdoor ambient temperature, outdoor ambient humidity, compressor discharge pressure, compressor discharge temperature, suction pressure, electronic expansion valve opening, indoor occupancy density, compressor power consumption, and fan power consumption; the virtual sensing parameters are calculated based on the physical sensing parameters through a mechanism model, including the actual cooling and heating load of the indoor unit, the phase change position of the refrigerant in the pipeline, and the pipeline pressure loss.

[0012] By adopting the above technical solutions, the specific range of physical sensing parameters and the acquisition method of virtual sensing parameters were clarified, realizing comprehensive coverage and in-depth perception of the operating status of multi-unit systems. This ensures that the collected data can fully reflect the system operation and load requirements, and further improves the accuracy and comprehensiveness of system status data.

[0013] As a further aspect of the present invention: the AI ​​decision layer includes an AI load prediction unit and a multi-objective optimization unit. The AI ​​load prediction unit has a built-in load prediction model based on a long short-term memory network, which is used to output the predicted cooling and heating load values ​​of each indoor unit within a preset time period based on system status data. The multi-objective optimization unit has a built-in optimization algorithm, which is used to construct and solve multi-objective optimization logic based on the load prediction values ​​and the dynamic weights output by the operating condition adaptation layer, so as to obtain the initial optimal opening degree of the electronic expansion valve of each indoor unit and the initial optimal operating frequency of the compressor.

[0014] By adopting the above technical solution, the load prediction model constructed by the long short-term memory network realizes the early prediction of load. Combined with the optimization algorithm, the multi-objective optimization solution is completed, and the initial optimal control parameters are output. This provides a basis for the parameter correction of the working condition adaptation layer, realizes the accurate conversion from data to initial control commands, and solves the problem of load response lag in traditional control.

[0015] As a further aspect of the present invention: the input features of the load forecasting model include fused historical system status data, seasonal features, work-rest patterns and predicted population density values. The seasonal features are encoded as preset values ​​corresponding to different seasons and transitional seasons. The work-rest patterns are divided into two categories: weekdays and weekends, and each category is further subdivided into different time periods.

[0016] By adopting the above technical solution and integrating historical system status data, seasonal characteristics, work and rest patterns, and personnel density prediction values, the prediction accuracy and adaptability of the load prediction model are improved, ensuring the reliability of the cooling / heating load prediction results of each indoor unit within the preset time period, and providing accurate load basis for subsequent optimization solutions.

[0017] As a further aspect of the present invention: the core of the multi-objective optimization logic is to construct an optimization objective by allocating error and total system energy consumption. The allocation error is the difference between the actual refrigerant flow and the required refrigerant flow, and the total system energy consumption is the sum of the compressor energy consumption and the fan energy consumption. The proportion of allocation error and total system energy consumption in the optimization objective is adjusted by dynamic weighting.

[0018] By adopting the above technical solution, an optimization target is constructed by integrating the allocation error and the total system energy consumption, and the ratio of the two is adjusted by dynamic weighting. This achieves synergistic optimization of refrigerant flow allocation accuracy and system energy-saving effect, avoiding the problem of pursuing allocation accuracy or energy-saving effect alone while ignoring the other target, and improving the overall operating efficiency of the system.

[0019] As a further aspect of the present invention: the operating condition adaptation layer includes an operating condition identification unit and a parameter correction unit. The operating condition identification unit has a built-in clustering algorithm to extract outdoor temperature, indoor load rate and operating mode as clustering features, and divides the operating conditions into multiple typical operating conditions. The parameter correction unit presets dynamic weights and control parameter correction coefficients under different operating conditions based on the operating condition identification results, fine-tunes the initial optimal control parameters, and outputs the optimal control parameters adapted to the current operating condition.

[0020] By adopting the above technical solution, the operating conditions are accurately identified through clustering algorithms. The initial optimal control parameters are fine-tuned by combining dynamic weights and control parameter correction coefficients under different operating conditions. This achieves accurate adaptation of control parameters to the current operating conditions, ensuring that the system can operate stably and efficiently under different operating conditions.

[0021] As a further aspect of the present invention: the typical operating conditions include low-temperature heating conditions, high-temperature cooling conditions, standard operating conditions, partial load conditions, and transitional season conditions. Each type of typical operating condition is divided by a preset outdoor temperature range, indoor load rate range, and operating mode.

[0022] By adopting the above technical solution and by pre-setting the outdoor temperature range, indoor load rate range, and operating mode classification standards, comprehensive coverage and precise definition of common operating conditions of multi-split air conditioning systems are achieved. This provides a clear basis for parameter correction of the operating condition adaptation layer and ensures the pertinence and effectiveness of operating condition adaptation.

[0023] As a further aspect of the present invention: the execution control layer includes a drive circuit and an actuator. The actuator consists of a stepper electronic expansion valve, a variable frequency compressor, and a brushless DC fan. The drive circuit adopts a pulse width modulation adjustment method to convert the optimal control parameters into a drive signal for the actuator to identify, thereby driving the actuator to move and achieve precise distribution of refrigerant flow.

[0024] By adopting the above technical solution, the optimal control parameters are converted into drive signals that the actuator can recognize through the pulse width modulation drive circuit, which drives the stepping electronic expansion valve, the variable frequency compressor and the brushless DC fan to work together, thereby achieving precise regulation of refrigerant flow. This transforms upper-level decisions into actual execution actions and ensures the accurate allocation of targets.

[0025] As a further aspect of the present invention: the energy-saving collaborative layer includes an energy consumption monitoring unit and a weight adjustment unit. The energy consumption monitoring unit is linked with the power metering module of the multi-source sensing layer to collect the total energy consumption of the system in real time. The weight adjustment unit is used to compare the real-time total energy consumption of the system with a preset energy consumption threshold. If the threshold is exceeded, the multi-objective optimization weight is dynamically adjusted and fed back to the AI ​​decision layer. At the same time, the real-time load is compared with the predicted load, and the real-time energy consumption is compared with the predicted energy consumption. If the deviation exceeds the preset range, the load prediction model is fine-tuned and the optimization parameters are re-solved.

[0026] By adopting the above technical solution, the total energy consumption of the system is collected in real time by the energy consumption monitoring unit. The multi-objective optimization weights are dynamically adjusted in combination with the preset energy consumption threshold. At the same time, the deviation between the prediction results and the optimization results is corrected, so as to realize the real-time control of the system energy consumption and the continuous optimization of the control accuracy, ensuring the stability of the energy-saving effect and the dynamic accuracy of the control process under all operating conditions.

[0027] Compared with the prior art, the beneficial effects of the present invention are: By employing a five-layer collaborative architecture and load prediction, the lag problem of traditional feedback regulation is solved, the accuracy of refrigerant distribution under different operating conditions is improved, and indoor comfort is ensured. At the same time, through multi-objective optimization and dynamic weight adjustment, the synergistic improvement of distribution accuracy and energy-saving effect is achieved, realizing the predictive and precise distribution of refrigerant flow. By intelligently identifying operating conditions and dynamically correcting parameters, the system adapts to the needs of different operating conditions. At the same time, with the help of real-time monitoring and deviation correction of the energy-saving collaboration layer, the system can operate stably and efficiently under various operating conditions, significantly improving the reliability and energy-saving stability of the system and possessing full-condition adaptive capability.

[0028] Other features and advantages of the present invention will be disclosed in detail in the following detailed description and accompanying drawings. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall structure of a method for precise refrigerant flow allocation and an energy-saving control system for a multi-split air conditioning system according to an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] In this embodiment of the invention, the method for precise refrigerant flow allocation and the energy-saving control system for a multi-split air conditioning system are described in [reference needed]. Figure 1 As shown, the system employs a five-layer architecture comprising a multi-source sensing layer, an AI decision-making layer, an operating condition adaptation layer, an execution control layer, and an energy-saving collaboration layer. This architecture enables precise prediction and allocation of refrigerant flow in multi-split air conditioning systems and energy-saving control across all operating conditions. The functions, inputs, outputs, and collaboration logic of each layer are as follows: 1. Multi-source sensing layer Function: Collects all-dimensional parameters during the operation of a multi-unit air conditioning system, performs data preprocessing and fusion, and outputs accurate system status data.

[0032] Composition: Includes a physical sensing unit and a data fusion unit.

[0033] The physical sensing unit consists of a high-precision temperature sensor, a pressure sensor, a humidity sensor, an infrared millimeter-wave personnel sensor, a power metering module, and a data acquisition card; the data fusion unit has a built-in Kalman filter algorithm.

[0034] Parameter acquisition method: A combination of physical and virtual sensing is used to collect 12 core parameters. Physical sensing parameters include indoor unit return air temperature, outlet air temperature, indoor relative humidity, outdoor ambient temperature, outdoor ambient humidity, compressor discharge pressure, compressor discharge temperature, suction pressure, electronic expansion valve opening, indoor occupancy density, compressor power consumption, and fan power consumption. Virtual sensing parameters are calculated based on physical parameters through a mechanistic model, including the actual cooling / heating load of the indoor unit, the phase change position of the refrigerant in the pipeline, and pipeline pressure loss.

[0035] The accuracy requirements for each physical sensor parameter are as follows: the measurement accuracy of indoor unit return air temperature and air outlet temperature is ±0.1℃; the measurement accuracy of indoor relative humidity is ±2%RH; the measurement accuracy of compressor exhaust pressure is ±0.01MPa; the measurement accuracy of indoor personnel density is ±1 person; the sampling frequency of the data acquisition card is set to 10Hz to ensure the real-time nature of parameter acquisition.

[0036] Data fusion process: The data fusion unit receives the raw parameters collected by the physical sensing unit and preprocesses them using a Kalman filter algorithm. This algorithm is used to remove noise interference, such as temperature fluctuations caused by personnel movement, and finally outputs accurate system status data.

[0037] Input: Raw parameters acquired by the physical sensing unit; Output: The fused system status data is transmitted to the AI ​​decision-making layer and the operating condition adaptation layer.

[0038] 2. AI Decision-Making Layer Function: Based on the fused system status data, it enables advance prediction of the dynamic load of each indoor unit, while completing multi-objective optimization solutions and outputting the initial optimal opening degree of the electronic expansion valve of each indoor unit and the initial optimal operating frequency of the compressor.

[0039] Composition: Includes an AI load prediction unit and a multi-objective optimization unit.

[0040] The AI ​​load forecasting unit incorporates a load forecasting model based on LSTM (Long Short-Term Memory) network; the multi-objective optimization unit incorporates a particle swarm optimization algorithm.

[0041] 2.1 Construction and Training Process of AI Load Prediction Model Model structure: The LSTM model includes an input layer, a hidden layer, and an output layer.

[0042] The input layer has 12 nodes, corresponding to the 12-dimensional feature parameters; the hidden layer has 3 layers, with 64 nodes in each layer. The number of output layer nodes is the number of indoor units, and the corresponding output is the load forecast value for each indoor unit.

[0043] Input features: fused historical data, seasonal features, work-rest patterns, and predicted population density. Historical data uses system status data from the same time period over the past 7 days; seasonal features are coded as values ​​from 0 to 4, corresponding to spring, summer, autumn, winter, and transitional seasons, respectively; work-rest patterns are divided into weekdays and weekends, each further subdivided into working hours and rest periods.

[0044] The training process specifically includes: Data preparation: Collect operational data of multi-split air conditioning systems for different apartment types and climate zones, with a data volume of no less than 500,000 records.

[0045] The data is preprocessed, including missing value imputation and outlier removal. Missing values ​​are imputed using linear interpolation, and outliers are identified and removed using the 3σ criterion.

[0046] Dataset partitioning: Divide the dataset into training set, validation set, and test set in a 7:2:1 ratio.

[0047] In one feasible embodiment, the training set includes 350,000 data points for model training; the validation set includes 100,000 data points for adjusting model parameters; and the test set includes 50,000 data points for evaluating the final performance of the model.

[0048] Model pre-training: The LSTM model is pre-trained based on the preprocessed training set data, with the training batch size set to 64 and the number of iterations set to 100.

[0049] The Adam optimizer, with a learning rate of 0.001, was used to optimize the model parameter update process; the mean squared error loss function was used to measure the difference between the model's predicted values ​​and the actual load values.

[0050] Model fine-tuning: Input the validation set data into the pre-trained model and adjust the model parameters based on the loss value of the validation set.

[0051] If the loss value on the validation set does not decrease after 10 consecutive iterations, then stop fine-tuning and obtain the basic model.

[0052] Model adaptation: For specific application scenarios, a small amount of on-site operational data is collected to further fine-tune the basic model and improve the prediction accuracy of the model in specific scenarios.

[0053] Model evaluation: Input the test set data into the trained model and calculate the prediction error.

[0054] The model is considered to have passed training when the prediction error of the indoor unit's cooling / heating load for the next 5-15 minutes is ≤ ±5%.

[0055] Input to the AI ​​load prediction unit: fused system state data output from the multi-source sensing layer; Output: The predicted cooling / heating load values ​​of each indoor unit for the next 5-15 minutes are transmitted to the multi-objective optimization unit and the operating condition adaptation layer.

[0056] 2.2 Multi-objective optimization unit Function: Based on the load forecast value and the dynamic weights output by the operating condition adaptation layer, a multi-objective optimization function is constructed and solved to obtain the initial optimal opening degree of the electronic expansion valve of each indoor unit and the initial optimal operating frequency of the compressor.

[0057] Construction of multi-objective optimization function: The objective function is .

[0058] in, E represents the refrigerant flow allocation error, which is the difference between the actual allocated flow and the required flow; E represents the total system energy consumption, which is the sum of the compressor energy consumption and the fan energy consumption. , The weights are dynamic and are output by the working condition adaptation layer.

[0059] Solution process: The objective function is solved using the particle swarm optimization algorithm.

[0060] The particle population size is set to 50, the number of iterations is 30, the inertia weight is 0.7, the cognitive factor is 1.4, and the social factor is 1.4.

[0061] The algorithm searches for the opening degree of the electronic expansion valve of each indoor unit and the operating frequency of the compressor that minimize the objective function value, and uses them as the initial optimal control parameters.

[0062] Inputs: Load forecast values ​​output by the AI ​​load forecasting unit, and dynamic weights output by the operating condition adaptation layer. and ; Output: The initial optimal opening degree of the electronic expansion valve of each indoor unit and the initial optimal operating frequency of the compressor are transmitted to the operating condition adaptation layer.

[0063] 3. Working Condition Adaptation Layer Function: Enables intelligent identification of operating conditions, adjusts the dynamic weights of the multi-objective optimization function based on the identification results, corrects the initial optimal control parameters output by the AI ​​decision layer, and outputs the optimal control parameters adapted to the current operating conditions.

[0064] Composition: Includes a working condition identification unit and a parameter correction unit.

[0065] The working condition identification unit incorporates a fuzzy clustering algorithm.

[0066] Operating condition identification process: The operating condition identification unit receives system status data output by the multi-source sensing layer and extracts outdoor temperature, indoor load rate and operating mode as clustering features.

[0067] The indoor load rate is the ratio of the actual load to the rated load; the operating modes include three categories: cooling, heating, and dehumidification.

[0068] The operating conditions were divided into 5 typical categories using a fuzzy clustering algorithm, with the specific classification criteria as follows: Operating Condition 1 (Low Temperature Heating): Outdoor temperature ≤ 5℃, and operating mode is heating; Operating Condition 2 (High Temperature Cooling): Outdoor temperature ≥ 35℃, and the operating mode is cooling; Operating condition 3 (standard operating condition): 5℃ < outdoor temperature < 35℃, and the operating mode is cooling or heating; Operating Condition 4 (Partial Load): Indoor load rate < 30%, operating mode is any mode; Operating Condition 5 (Transitional Season): Outdoor temperature 15-25℃, and operating mode is ventilation or dehumidification.

[0069] Inputs to the operating condition identification unit: system status data output from the multi-source sensing layer and load prediction values ​​output from the AI ​​load prediction unit; Output: The identification result of the current operating condition is transmitted to the parameter correction unit.

[0070] Parameter correction process: Based on the working condition identification results, the parameter correction unit presets the dynamic weights α, β and control parameter correction coefficients for different working conditions.

[0071] The rules for setting dynamic weights are as follows: Under standard operating conditions Prioritize ensuring the accuracy of refrigerant flow distribution; under partial load conditions Prioritize reducing system energy consumption; under low-temperature heating, high-temperature cooling, and transitional season conditions. Balance the allocation accuracy and energy consumption.

[0072] The control parameter correction coefficient is used to fine-tune the initial optimal control parameters output by the AI ​​decision layer. For example, the correction coefficient for the electronic expansion valve adjustment step size is 0.5 under low temperature heating conditions, 1.0 under standard conditions, and 0.8 under partial load conditions.

[0073] The parameter correction unit corrects the initial optimal control parameters according to the correction coefficient corresponding to the current operating condition, so as to obtain the optimal control parameters that are suitable for the current operating condition.

[0074] Inputs: Initial optimal control parameters output by the AI ​​decision-making layer, and working condition identification results output by the working condition identification unit; Output: The optimal control parameters adapted to the current operating conditions are transmitted to the execution control layer and the energy-saving coordination layer.

[0075] 4. Execution Control Layer Function: Receives the optimal control parameters output by the operating condition adaptation layer, drives the actuator to move, and achieves precise adjustment of refrigerant flow.

[0076] Composition: Includes drive circuit and actuator.

[0077] The actuator consists of a stepper electronic expansion valve, a variable frequency compressor, and a brushless DC fan; the drive circuit adopts PWM (pulse width modulation) regulation.

[0078] Working process: The drive circuit receives the optimal control parameters output by the working condition adaptation layer, including the optimal opening degree of the electronic expansion valve of each indoor unit, the optimal operating frequency of the compressor, and the optimal speed of the fan.

[0079] The control signal is converted into a drive signal recognizable by the actuator through PWM regulation, which drives the stepper electronic expansion valve to adjust its opening, the variable frequency compressor to adjust its operating frequency, and the brushless DC fan to adjust its speed. The stepper electronic expansion valve has an adjustment accuracy of ±1 pulse, and the variable frequency compressor has a frequency adjustment range of 15-120Hz, thereby achieving precise distribution of refrigerant flow.

[0080] Input: The optimal control parameters adapted to the current working condition output by the working condition adaptation layer; Output: Drives the actuator to adjust the refrigerant flow, and feeds back the actual operating parameters of the actuator to the energy-saving coordination layer.

[0081] 5. Energy-saving synergy layer Function: Real-time monitoring of system energy consumption, dynamic adjustment of the weights of multi-objective optimization functions, feedback correction of the optimization results of the AI ​​decision layer, and ensuring the energy-saving effect of the system.

[0082] Composition: Includes an energy consumption monitoring unit and a weight adjustment unit.

[0083] Among them, the energy consumption monitoring unit is linked with the power metering module of the multi-source sensing layer to collect the power consumption of the compressor and fan in real time; the weight adjustment unit is integrated into the industrial-grade AI controller of the AI ​​decision layer.

[0084] Working process: The energy consumption monitoring unit collects the total energy consumption of the system in real time and compares it with the preset energy consumption threshold.

[0085] If the actual energy consumption exceeds the preset threshold, the weight adjustment unit dynamically adjusts the weight β of the multi-objective optimization function, increases the weight ratio of energy consumption optimization, and feeds the adjusted weight back to the multi-objective optimization unit of the AI ​​decision layer.

[0086] Meanwhile, the energy consumption monitoring unit compares the real-time energy consumption data with the predicted energy consumption data of the AI ​​decision layer and calculates the energy consumption deviation. If the deviation exceeds ±5%, the AI ​​decision layer is triggered to re-solve the optimization parameters.

[0087] In addition, the energy-saving collaboration layer will compare the real-time load data with the predicted values ​​of the AI ​​load prediction unit to calculate the prediction deviation. If the deviation exceeds ±5%, the LSTM model will be fine-tuned online to correct the prediction error.

[0088] Inputs: Actual operating parameters of the actuator fed back from the execution control layer, and energy consumption data collected by the multi-source sensing layer; Output: The adjusted weights of the multi-objective optimization function are transmitted to the AI ​​decision layer; the model fine-tuning signal is transmitted to the AI ​​load prediction unit.

[0089] 6. Overall System Control Flow Initialization: After the system starts up, the physical sensing unit of the multi-source perception layer begins to collect raw parameters, and the AI ​​decision layer loads the pre-trained LSTM model and the control parameter set for each working condition. Data fusion: The data fusion unit of the multi-source sensing layer uses the Kalman filter algorithm to preprocess the raw parameters, remove noise interference, and output accurate system status data; Load forecasting and operating condition identification: The AI ​​load forecasting unit of the AI ​​decision layer inputs the fused system status data into the LSTM model to obtain the load forecast values ​​of each indoor unit in the next 5-15 minutes; at the same time, the operating condition identification unit of the operating condition adaptation layer extracts the clustering features in the system status data and identifies the current operating condition through fuzzy clustering algorithm. Multi-objective optimization solution: The multi-objective optimization unit of the AI ​​decision layer obtains the initial optimal control parameters by using the particle swarm optimization algorithm based on the load forecast value and the weights initially set by the operating condition adaptation layer; the parameter correction unit of the operating condition adaptation layer corrects the initial optimal control parameters based on the current operating condition identification result to obtain the optimal control parameters adapted to the current operating condition. Execution regulation: The drive circuit of the execution control layer receives the optimal control parameters and drives the stepper electronic expansion valve, variable frequency compressor and brushless DC fan to achieve precise distribution of refrigerant flow; Energy-saving coordination and feedback correction: The energy consumption monitoring unit of the energy-saving coordination layer collects the total energy consumption of the system in real time. If the energy consumption exceeds the preset threshold, the weight of the multi-objective optimization function is dynamically adjusted and fed back to the AI ​​decision layer. At the same time, the real-time load and the predicted load, and the real-time energy consumption and the predicted energy consumption are compared. If the deviation exceeds the threshold, the LSTM model is fine-tuned and the optimization parameters are re-solved. Iterative loop: Repeat steps 2-6 to achieve continuous dynamic and precise control and energy-saving optimization.

[0090] This embodiment refers to a multi-split air conditioning system in a commercial office building. The system includes one outdoor unit and eight indoor units, covering eight independent office areas.

[0091] 1. Multi-source sensing layer configuration: The indoor unit return air temperature and air outlet temperature adopt PT1000 high-precision temperature sensor, the compressor exhaust pressure adopts diffused silicon pressure sensor, the indoor relative humidity adopts capacitive humidity sensor, the indoor occupant density adopts infrared millimeter wave sensor, the power metering adopts Hall sensor, and the sampling frequency of the data acquisition card is set to 10Hz.

[0092] 2. AI Decision Layer Configuration: An industrial-grade AI controller equipped with an ARM Cortex-A76 processor and 4GB of memory is used, incorporating a pre-trained LSTM load prediction model and a particle swarm optimization algorithm. The LSTM model has 12 input layer nodes, 3 hidden layers with 64 nodes each, and 8 output layer nodes. The particle swarm optimization algorithm has a population size of 50 and 30 iterations.

[0093] 3. Operating Condition Adaptation Layer Configuration: Five pre-defined operating condition classification standards and corresponding dynamic weights and parameter correction coefficients. For example, under low-temperature heating conditions... The electronic expansion valve adjustment step correction factor is 0.5; under partial load conditions. The compressor frequency regulation coefficient is 0.8.

[0094] 4. Execution control layer configuration: The actuator adopts a stepper electronic expansion valve, a variable frequency compressor and a brushless DC fan; the drive circuit adopts PWM regulation.

[0095] 5. Energy-saving coordination layer configuration: The preset system energy consumption threshold is 1.2 kW·h / hour, which is determined based on the historical operating data of the multi-split system of this office building. When the actual energy consumption exceeds this threshold, the weight adjustment unit will adjust the β value from 0.4 to 0.6.

[0096] System operation process: After initialization, the sensors in the multi-source sensing layer begin to collect parameters, including the return air temperature, outlet air temperature, indoor relative humidity, indoor occupancy density, outdoor ambient temperature and humidity, compressor discharge pressure, discharge temperature, suction pressure, electronic expansion valve opening, and power consumption of the compressor and fan for the eight indoor units.

[0097] The data fusion unit preprocesses the above parameters using the Kalman filter algorithm to remove noise such as temperature fluctuations caused by personnel movement and outputs system status data.

[0098] The AI ​​load prediction unit inputs the fused system status data into the LSTM model to obtain the predicted cooling load values ​​for the eight indoor units in the next 10 minutes; the operating condition identification unit extracts the outdoor temperature of 38℃, indoor load rate of 45%, and operating mode (cooling), and identifies the current operating condition as a high-temperature cooling condition through a fuzzy clustering algorithm.

[0099] The multi-objective optimization unit uses the particle swarm optimization algorithm to solve for the initial optimal opening degree of the electronic expansion valves of the eight indoor units and the initial optimal operating frequency of the compressor based on the predicted cooling load value and the initial weights α=0.5 and β=0.5 corresponding to the high-temperature cooling condition. The parameter correction unit fine-tunes the initial optimal control parameters based on the correction coefficient of the high-temperature cooling condition to obtain the optimal control parameters adapted to the current condition.

[0100] The drive circuit of the execution control layer receives the optimal control parameters, drives the electronic expansion valve of each indoor unit to adjust to the corresponding opening degree, and drives the compressor to adjust the operating frequency to the optimal value, so as to achieve precise distribution of refrigerant flow.

[0101] The energy consumption monitoring unit of the energy-saving collaboration layer collects the total energy consumption of the system in real time. If the energy consumption exceeds the preset threshold of 1.2 kW·h / hour, the weight β is adjusted to 0.6 and fed back to the multi-objective optimization unit. The multi-objective optimization unit re-solves to obtain new optimal control parameters to ensure that the system energy consumption is reduced to a reasonable range.

[0102] Meanwhile, if the real-time cooling load is compared with the predicted cooling load and the deviation exceeds ±5%, the LSTM model is fine-tuned online to ensure the accuracy of subsequent predictions.

[0103] Through the above implementation methods, the refrigerant flow distribution error of the multi-split air conditioning system in the commercial office building is controlled within ±2.5%, the overall energy consumption is reduced by 22% compared with the traditional system, the indoor temperature fluctuation range is ≤0.5℃, and the operating efficiency and comfort are significantly improved.

[0104] This invention provides a method for precise refrigerant flow allocation and an energy-saving control system for multi-split air conditioning systems. It can achieve predictive and precise allocation of refrigerant flow and energy-saving coordinated control under all operating conditions, with high reliability.

[0105] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0106] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for precise refrigerant flow distribution and an energy-saving control system for multi-split air conditioning systems, characterized in that, It includes a multi-source sensing layer, an AI decision-making layer, an operating condition adaptation layer, an execution control layer, and an energy-saving coordination layer. The multi-source sensing layer collects system operating parameters and performs fusion processing, then outputs system status data to the AI ​​decision-making layer and the operating condition adaptation layer. The AI ​​decision-making layer uses the system status data to predict load in advance and solve multi-objective optimization problems, and outputs the initial optimal control parameters to the operating condition adaptation layer. The operating condition adaptation layer identifies the current operating condition, adjusts the multi-objective optimization weights based on the operating condition identification results, corrects the initial optimal control parameters, and outputs the optimal control parameters adapted to the current operating condition to the execution control layer and the energy-saving coordination layer. The execution control layer drives the actuator to adjust the refrigerant flow based on the optimal control parameters and feeds back the actual operating parameters of the actuator to the energy-saving coordination layer. The energy-saving coordination layer monitors the system energy consumption in real time, dynamically adjusts the multi-objective optimization weights and feeds them back to the AI ​​decision layer, while also providing feedback correction to the prediction and optimization results of the AI ​​decision layer.

2. The method for precise refrigerant flow allocation and energy-saving control system of a multi-split air conditioning system according to claim 1, characterized in that, The multi-source sensing layer includes a physical sensing unit and a data fusion unit. The physical sensing unit collects core parameters by combining physical sensing with virtual sensing. The core parameters include physical sensing parameters and virtual sensing parameters. The data fusion unit has a built-in filtering algorithm to preprocess the raw parameters collected by the physical sensing unit to remove noise interference and output accurate system status data.

3. The method for precise refrigerant flow allocation and energy-saving control system of a multi-split air conditioning system according to claim 2, characterized in that, The physical sensing parameters include indoor unit return air temperature, air outlet temperature, indoor relative humidity, outdoor ambient temperature, outdoor ambient humidity, compressor discharge pressure, compressor discharge temperature, suction pressure, electronic expansion valve opening, indoor occupancy density, compressor power consumption, and fan power consumption. The virtual sensing parameters are calculated based on the physical sensing parameters through a mechanism model, including the actual cooling and heating load of the indoor unit, the phase change position of the refrigerant in the pipeline, and pipeline pressure loss.

4. The method for precise refrigerant flow allocation and energy-saving control system of a multi-split air conditioning system according to claim 1, characterized in that, The AI ​​decision-making layer includes an AI load prediction unit and a multi-objective optimization unit. The AI ​​load prediction unit has a built-in load prediction model based on a long short-term memory network, which is used to output the predicted cooling and heating load values ​​of each indoor unit within a preset time period based on system status data. The multi-objective optimization unit has a built-in optimization algorithm, which is used to construct and solve multi-objective optimization logic based on the load prediction values ​​and the dynamic weights output by the operating condition adaptation layer, to obtain the initial optimal opening degree of the electronic expansion valve of each indoor unit and the initial optimal operating frequency of the compressor.

5. The method for precise refrigerant flow allocation and energy-saving control system of a multi-split air conditioning system according to claim 4, characterized in that, The input features of the load forecasting model include fused historical system status data, seasonal features, work-rest patterns, and predicted population density. The seasonal features are encoded with preset values ​​corresponding to different seasons and transitional seasons. The work-rest patterns are divided into two categories: weekdays and weekends, and each category is further subdivided into different time periods.

6. The method for precise refrigerant flow allocation and energy-saving control system of a multi-split air conditioning system according to claim 4, characterized in that, The core of the multi-objective optimization logic is to construct optimization objectives by allocating errors and total system energy consumption. The allocation error is the difference between the actual refrigerant flow and the required refrigerant flow, and the total system energy consumption is the sum of the compressor energy consumption and the fan energy consumption. The proportion of allocation error and total system energy consumption in the optimization objectives is adjusted by dynamic weighting.

7. The method for precise refrigerant flow distribution and energy-saving control system of a multi-split air conditioning system according to claim 1, characterized in that, The operating condition adaptation layer includes an operating condition identification unit and a parameter correction unit. The operating condition identification unit has a built-in clustering algorithm that extracts outdoor temperature, indoor load rate and operating mode as clustering features to divide the operating conditions into multiple typical operating conditions. The parameter correction unit presets dynamic weights and control parameter correction coefficients for different operating conditions based on the operating condition identification results, fine-tunes the initial optimal control parameters, and outputs the optimal control parameters that adapt to the current operating conditions.

8. The method for precise refrigerant flow distribution and energy-saving control system of a multi-split air conditioning system according to claim 7, characterized in that, The typical operating conditions include low-temperature heating, high-temperature cooling, standard, partial load, and transitional season conditions. Each typical operating condition is divided by a preset outdoor temperature range, indoor load rate range, and operating mode.

9. The method for precise refrigerant flow allocation and energy-saving control system of a multi-split air conditioning system according to claim 1, characterized in that, The execution control layer includes a drive circuit and an actuator. The actuator consists of a stepper electronic expansion valve, a variable frequency compressor, and a brushless DC fan. The drive circuit uses pulse width modulation to convert the optimal control parameters into a drive signal for the actuator to identify, thereby driving the actuator to achieve precise distribution of refrigerant flow.

10. The method for precise refrigerant flow allocation and energy-saving control system of a multi-split air conditioning system according to claim 1, characterized in that, The energy-saving collaborative layer includes an energy consumption monitoring unit and a weight adjustment unit. The energy consumption monitoring unit is linked with the power metering module of the multi-source sensing layer to collect the total energy consumption of the system in real time. The weight adjustment unit is used to compare the real-time total energy consumption of the system with a preset energy consumption threshold. If it exceeds the preset threshold, the multi-objective optimization weight is dynamically adjusted and fed back to the AI ​​decision layer. At the same time, the real-time load and the predicted load, and the real-time energy consumption and the predicted energy consumption are compared. If the deviation exceeds the preset range, the load prediction model is fine-tuned and the optimization parameters are re-solved.

Images