A method for calculating and controlling the energy of a split-type air conditioner

Abstract

This invention discloses a method for calculating and controlling the energy consumption of a split-type air conditioner, relating to the field of intelligent control technology for HVAC systems. The method includes: inputting a multi-source environmental equipment dataset into an LSTM neural network to generate a refrigerant dryness prediction value and calculating the theoretical cooling capacity based on the refrigerant dryness prediction value; calculating a baseline energy efficiency ratio (EER) based on the theoretical cooling capacity and dynamically correcting it based on the refrigerant dryness prediction value to generate a corrected EER; constructing a reinforcement learning decision model, inputting the corrected EER and the multi-source environmental equipment dataset into the reinforcement learning decision model to generate an optimized parameter combination; adjusting the compressor frequency based on the optimized parameter combination, and triggering electronic expansion valve control based on the refrigerant dryness prediction value. This invention achieves high-precision indirect measurement by constructing an LSTM neural network to replace traditional sensors, improving adaptability to dynamic operating conditions. By dynamically correcting the EER based on the optimal value defined by the refrigerant type, it avoids the accumulation of errors caused by changes in operating conditions due to a fixed EER.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for heating, ventilation and air conditioning, and in particular to a method for calculating and controlling the energy consumption of split air conditioners. Background Technology

[0002] In recent years, with the deep integration of artificial intelligence and the Internet of Things (IoT) technologies, the intelligent control of air conditioning systems has gradually evolved from single-parameter adjustment to multi-dimensional dynamic optimization. Data-driven energy efficiency management technology has become a research hotspot, especially in the field of split air conditioners. Scholars and enterprises are committed to collecting environmental and equipment operation data through sensor networks and combining them with machine learning models to achieve energy consumption prediction and control strategy optimization. Research on the refrigerant circulation characteristics of air conditioning systems is deepening, and refrigerant dryness, as a key parameter affecting refrigeration efficiency, has made dynamic monitoring and compensation algorithms a core direction for improving system energy efficiency.

[0003] Existing methods mostly evaluate air conditioner performance based on static or seasonal energy efficiency ratios, but they do not fully consider the nonlinear impact of real-time changes in refrigerant dryness on cooling capacity and energy consumption. When refrigerant dryness deviates from the design value, there is a significant discrepancy between the theoretical cooling capacity and the actual output, leading to distorted energy efficiency ratio calculations. In addition, the intelligent control of split air conditioners often relies on preset rules or offline trained models, which are difficult to cope with real-time changes under complex operating conditions, resulting in generated action parameters that lack accurate responses to internal thermodynamic processes. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for calculating and controlling the energy consumption of a split air conditioner to solve the problems of insufficient refrigerant dryness compensation and insufficient response of control parameters.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for calculating and controlling the energy of a split-type air conditioner, comprising: inputting a collected multi-source environmental equipment dataset into an LSTM neural network to generate a refrigerant dryness prediction value and calculating a theoretical cooling capacity based on the refrigerant dryness prediction value; calculating a baseline energy efficiency ratio based on the theoretical cooling capacity and dynamically correcting it based on the refrigerant dryness prediction value to generate a corrected energy efficiency ratio; and constructing a reinforcement learning decision model, inputting the corrected energy efficiency ratio and the multi-source environmental equipment dataset into the reinforcement learning decision model to generate an optimized parameter combination.

[0008] The compressor frequency is adjusted based on the optimized parameter combination, and the electronic expansion valve control is triggered based on the refrigerant dryness prediction value. The temperature change rate and compressor power deviation are monitored in real time, and the transfer learning mechanism is triggered based on the monitoring results to update the LSTM neural network parameters and recalculate the refrigerant dryness prediction value and correct the energy efficiency ratio to generate the final optimized parameter combination.

[0009] As a preferred embodiment of the energy calculation and control method for split air conditioners described in this invention, the multi-source environmental equipment dataset includes indoor and outdoor temperatures, compressor current and voltage, evaporator inlet and outlet temperature difference, condenser inlet and outlet temperature difference, and user-set temperature.

[0010] As a preferred embodiment of the split-type air conditioner regulation energy calculation and control method of the present invention, the specific steps for generating the refrigerant dryness prediction value and calculating the theoretical cooling capacity based on the refrigerant dryness prediction value are as follows.

[0011] Define a six-dimensional input layer, a first LSTM layer, a second LSTM layer, and a fully connected layer to construct an LSTM neural network;

[0012] Input the multi-source environmental equipment dataset into the LSTM neural network to generate refrigerant dryness prediction values;

[0013] The compressor power is calculated by measuring the compressor current and voltage, and the theoretical cooling capacity is calculated by correlating this with the predicted refrigerant dryness value.

[0014] As a preferred embodiment of the energy calculation and control method for split-type air conditioners described in this invention, the calculation of the baseline energy efficiency ratio and the dynamic correction based on the refrigerant dryness prediction value to generate a corrected energy efficiency ratio are as follows:

[0015] Based on the theoretical cooling capacity and compressor power, a benchmark energy efficiency ratio is generated. The refrigerant type is identified according to the split air conditioner design manual, and the optimal value of refrigerant dryness is defined based on the refrigerant type.

[0016] Calculate the difference between the predicted refrigerant dryness value and the optimal refrigerant dryness value to generate the refrigerant dryness deviation, and define the attenuation coefficient according to the refrigerant type;

[0017] The baseline energy efficiency ratio is adjusted based on the refrigerant dryness deviation and attenuation coefficient to generate the corrected energy efficiency ratio.

[0018] As a preferred embodiment of the split-type air conditioning energy calculation and control method of the present invention, the specific steps for generating the optimized parameter combination are as follows:

[0019] The state space is defined based on the modified energy efficiency ratio and multi-source environmental equipment dataset;

[0020] Identify the controllable parameters of the split air conditioner according to the split air conditioner design manual, define the action space based on the controllable parameters of the split air conditioner, and define the output layer based on the drive interface;

[0021] A reinforcement learning decision model is constructed based on the state space, action space, and output layer;

[0022] The modified energy efficiency ratio and multi-source environmental equipment datasets are merged into a state vector through the state space, and the original action parameters are generated through the action space. The optimized parameter combination is then generated through the output layer.

[0023] As a preferred embodiment of the energy calculation and control method for split-type air conditioners described in this invention, the steps of adjusting the compressor frequency according to the optimized parameter combination and triggering electronic expansion valve control based on the refrigerant dryness prediction value are as follows:

[0024] Extract the values ​​of each parameter from the optimized parameter combination, convert the compressor frequency into a PWM signal that can be recognized by the split air conditioner control protocol, and transmit it to the inverter;

[0025] The allowable deviation threshold is defined according to the experimental calibration method. The deviation of refrigerant dryness is compared with the allowable deviation threshold, and the electronic expansion valve is triggered based on the judgment result.

[0026] As a preferred embodiment of the split-type air conditioner regulation energy calculation and control method of the present invention, the real-time monitoring of temperature change rate and compressor power deviation, and the triggering of a transfer learning mechanism based on the monitoring results, includes the following specific steps.

[0027] Collect real-time indoor temperature and compressor power, and calculate the temperature change rate and compressor power deviation;

[0028] Temperature change rate threshold and power deviation threshold are defined based on thermal comfort standards and compressor safety tolerance.

[0029] The triggering conditions for the transfer learning mechanism are determined based on the temperature change rate threshold and the power deviation threshold, and the transfer learning mechanism is activated based on the determination results.

[0030] As a preferred embodiment of the split-type air conditioning energy calculation and control method of the present invention, the specific steps for generating the final optimized parameter combination are as follows:

[0031] The parameters of the LSTM neural network are updated using the stochastic gradient descent method to regenerate the refrigerant dryness prediction value and update the corrected energy efficiency ratio.

[0032] The temperature change rate and compressor power deviation are merged into the multi-source environmental equipment dataset to generate a corrected multi-source environmental equipment dataset.

[0033] The updated modified energy efficiency ratio and the modified multi-source environmental equipment dataset are input into the reinforcement learning decision model to generate the final optimized parameter combination.

[0034] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the split-type air conditioning energy calculation and control method described in the first aspect of the present invention.

[0035] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the split-type air conditioning energy calculation and control method as described in the first aspect of the present invention.

[0036] The beneficial effects of this invention are as follows: by constructing an LSTM neural network to replace traditional sensors to achieve high-precision indirect measurement, the adaptability to dynamic operating conditions is improved. The energy efficiency ratio is dynamically corrected by combining the optimal value defined by the refrigerant type, avoiding the accumulation of errors caused by changes in operating conditions due to a fixed energy efficiency ratio. The transfer learning mechanism is triggered by dual monitoring of temperature change rate and compressor power deviation, and the LSTM parameters are updated online by using stochastic gradient descent, which improves the generalization ability of the LSTM network under sudden environmental changes. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of the energy calculation and control method for split air conditioning systems.

[0039] Figure 2 A flowchart for generating the corrected energy efficiency ratio.

[0040] Figure 3 A flowchart for constructing a reinforcement learning decision-making model.

[0041] Figure 4 A flowchart for triggering the transfer learning mechanism. Detailed Implementation

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0045] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for calculating and controlling the energy regulation of a split-type air conditioner, including the following steps:

[0046] S1. Input the collected multi-source environmental equipment dataset into the LSTM neural network to generate refrigerant dryness prediction values ​​and calculate the theoretical cooling capacity based on the refrigerant dryness prediction values.

[0047] S1.1 Collect indoor and outdoor temperatures, compressor current and voltage, evaporator inlet and outlet temperature difference, condenser inlet and outlet temperature difference, and user-set temperature, and perform preprocessing to generate a multi-source environmental equipment dataset;

[0048] It should be noted that the indoor unit evaporator surface temperature is collected at a sampling frequency of 1Hz using a PT100 temperature sensor, and the outdoor unit condenser ambient temperature is collected as the outdoor temperature. The original waveforms of the compressor three-phase current and DC bus voltage are captured at a sampling frequency of 10kHz using a current clamp and voltage probe, and the total harmonic distortion (THD) characteristics of the current are calculated in real time. The inlet and outlet pipe wall temperatures of the evaporator are measured using dual PT100 probes, and the inlet and outlet temperatures are subtracted from the outlet temperature to generate the evaporator inlet and outlet temperature difference. Dual PT100 probes are deployed at the inlet and outlet pipe walls of the condenser to generate the condenser inlet and outlet temperature difference. The user-set temperature value is read through the RS485 bus of the air conditioning controller. The collected sensor data are filtered by median filtering with a sliding window of 30 seconds to remove transient interference pulses. Linear interpolation is used to fill in missing data points caused by communication packet loss. Finally, Z-score standardization is performed on all feature fields to generate a multi-source environmental equipment dataset with unified dimensions.

[0049] S1.2 Define a six-dimensional input layer, a first LSTM layer, a second LSTM layer, and a fully connected layer, and construct an LSTM neural network;

[0050] It should be noted that a six-dimensional input layer is defined with 6 neurons to match the six-dimensional feature vectors in the multi-source environmental device dataset; a first LSTM layer is constructed and configured with 64 neuron units, and the hidden state output is passed to the second LSTM layer as a time-series memory carrier; 64 neuron units are deployed in the second LSTM layer and receive the hidden state sequence of the first LSTM layer, and the refrigerant state features across time steps are extracted through a gating mechanism; finally, the final time-step output vector of the second LSTM layer is input into a fully connected layer, and the fully connected layer uses the Sigmoid activation function to map the 64-dimensional feature vector to the refrigerant dryness prediction value in the interval of 0 to 1, thus completing the topology construction of the LSTM neural network;

[0051] It should also be noted that the gating mechanism is the core structure of the LSTM neural network, which controls the flow of information through the forget gate, input gate, and output gate. The measurement error tolerance of the PT100 temperature sensor and the maximum safe temperature difference allowed by the design of the split air conditioner evaporator define the threshold of the temperature difference between the inlet and outlet of the evaporator. The forget gate dynamically clears outdated state memories that exceed the threshold of the temperature difference between the inlet and outlet of the evaporator to avoid interfering with the current prediction. The input gate filters the harmonic components in the total harmonic distortion rate of the compressor current and writes them into the cell state to enhance the sensitivity to load changes. The cell state continuously remembers the condenser inlet and outlet temperature difference response within ten seconds after the compressor frequency is adjusted. The output gate suppresses the high-frequency fluctuation noise of the outdoor temperature and extracts only the dominant feature of the condenser inlet and outlet temperature difference to ensure the stability of the prediction.

[0052] The multi-source environmental equipment dataset was divided into a training set (70%), a validation set (15%), and a test set (15%). High-precision pressure sensors and platinum resistance temperature probes were installed in the evaporator and condenser pipes to simultaneously collect refrigerant saturation pressure and temperature. The refrigerant type of the current air conditioning unit was identified according to the split-type air conditioning design manual, and the true refrigerant dryness value was determined based on the refrigerant property table. The training set was input into an LSTM neural network, and the mean squared error was used as the loss function to measure the deviation between the predicted and true refrigerant dryness values. The expression is as follows:

[0053] ;

[0054] in, It is the loss function value, representing the overall error of the LSTM neural network in predicting the refrigerant dryness. This represents the total number of training set data records for a single split-type air conditioner laboratory prototype; This represents the training set data index, with a value range of 1- ; Indicates the first Predicted refrigerant dryness values ​​from the dataset; Indicates the first The actual values ​​of refrigerant dryness in the data set;

[0055] Using the Adam optimizer, with an initial learning rate of 0.001, a batch size of 32, and 100 training epochs, gradients are calculated using backpropagation to update the LSTM neural network weights until the maximum number of training epochs is reached. After training, validation and test sets are input into the LSTM neural network for validation and testing. The final LSTM neural network is then saved, with the expression:

[0056] ;

[0057] in, This indicates the predicted refrigerant dryness value output. This represents the Sigmoid activation function; This represents the weight matrix of the fully connected layer; Indicates the output gate vector; This represents the element-wise multiplication operator; Represents the cell state activation function; Indicates cell state; Indicates the bias term; It indicates a specific moment in time.

[0058] S1.3 Input the multi-source environmental equipment dataset into the LSTM neural network to generate the refrigerant dryness prediction value, calculate the compressor power through compressor current and voltage, and correlate it with the refrigerant dryness prediction value to calculate the theoretical cooling capacity.

[0059] It should be noted that an LSTM neural network is loaded, and the multi-source environmental equipment dataset is input into the LSTM neural network in a time series manner. Temporal features are extracted through a dual LSTM layer, and a refrigerant dryness prediction value in the 0-1 range is generated. Based on the compressor's three-phase current and DC bus voltage, the actual compressor power is calculated using an instantaneous power integration algorithm. The refrigerant property table is called along different paths, and the liquid and gas phase enthalpies are retrieved from the table based on the current condenser temperature. The latent heat of phase change of the evaporator is calculated based on these values. The compressor's discharge and suction enthalpies are retrieved from the refrigerant property table based on the discharge and suction pressure sensors, and the enthalpy difference between them is calculated. The actual compressor power and enthalpy difference are converted into refrigerant mass flow rate through power consumption-enthalpy difference equivalent conversion. The theoretical cooling capacity is calculated based on the refrigerant mass flow rate, latent heat of phase change, and refrigerant dryness prediction value. The expression is as follows:

[0060] ;

[0061] in, This indicates the theoretical cooling capacity, and the unit is kW. This indicates the refrigerant mass flow rate, measured in kg / s. It is the latent heat of phase change, and the unit is kJ / kg; This indicates the predicted value of refrigerant dryness.

[0062] S2. Calculate the baseline energy efficiency ratio based on the theoretical cooling capacity, and dynamically correct it according to the predicted value of refrigerant dryness to generate the corrected energy efficiency ratio.

[0063] S2.1. Generate a benchmark energy efficiency ratio based on theoretical cooling capacity and compressor power, and define the optimal value of refrigerant dryness according to the type of refrigerant;

[0064] It should be noted that the ratio of theoretical cooling capacity to compressor power is used to obtain the benchmark energy efficiency ratio under the current operating conditions. The benchmark energy efficiency ratio is used as a dynamic reference value to represent the energy efficiency status of the split air conditioner. Through a two-dimensional mapping table (refrigerant dryness optimal value mapping table), the current refrigerant type and benchmark energy efficiency ratio are used as indexes to match and output the corresponding optimal refrigerant dryness value in the refrigerant dryness optimal value mapping table.

[0065] S2.2 Calculate the difference between the predicted refrigerant dryness value and the optimal refrigerant dryness value, generate the refrigerant dryness deviation, and define the attenuation coefficient according to the refrigerant type;

[0066] It should be noted that the refrigerant dryness deviation is obtained by calculating the difference between the predicted refrigerant dryness value and the optimal refrigerant dryness value; the saturated gas phase density at the current discharge pressure is obtained by consulting the refrigerant property table using the current compressor discharge pressure and refrigerant type. The theoretical value of valve-controlled flow gain is derived based on the refrigerant flow momentum equation. Formula; Five operating condition combinations covering the typical operating range of split air conditioners were selected (including rated operating condition 40℃ condensing temperature / 10℃ evaporating temperature, high load operating condition 50℃ condensing temperature / 15℃ evaporating temperature, low load operating condition 35℃ condensing temperature / 5℃ evaporating temperature, high condensing and low evaporating operating condition 50℃ condensing temperature / 5℃ evaporating temperature, and low condensing and high evaporating operating condition 35℃ condensing temperature / 15℃ evaporating temperature), and the ambient temperature and compressor frequency were kept constant in each operating condition combination; an electronic expansion valve opening step response experiment was performed in each operating condition combination, increasing the electronic expansion valve opening step by 10%, and the refrigerant dryness response curve was recorded in real time, adjusting the theoretical value of valve control flow gain. Continue until there is no overshoot and the settling time is ≤30 seconds, then record the theoretical value of the valve-controlled flow gain at this point. As a calibration constant All five sets of operating parameters, refrigerant types, and their corresponding optimal values ​​are integrated into a three-dimensional mapping table. When the split air conditioner is running, the three-dimensional mapping table is queried by refrigerant type, collected condensing temperature, and evaporating temperature. The calibration constant closest to the five operating conditions is automatically selected using the Euclidean nearest neighbor matching algorithm. According to the calibration constant and saturated gas phase density Calculate the attenuation coefficient The expression is,

[0067] ;

[0068] in, Indicates the attenuation coefficient; Indicates the calibration constant; This represents the saturated gas phase density; This represents the theoretical value of the valve-controlled flow gain. This indicates the cross-sectional area of ​​the electronic expansion valve orifice, which is set at the factory by the electronic expansion valve manufacturer. This indicates the flow resistance coefficient inside the electronic expansion valve, which is set by the manufacturer when the electronic expansion valve is manufactured.

[0069] S2.3 Adjust the baseline energy efficiency ratio based on the refrigerant dryness deviation and attenuation coefficient to generate the corrected energy efficiency ratio.

[0070] It should be noted that the gain adjustment factor is obtained by performing a scalar multiplication operation on the refrigerant dryness deviation and the attenuation coefficient; the energy efficiency ratio sensitivity coefficient is calculated based on the compressor frequency using the sensitivity coefficient formula; and the corrected energy efficiency ratio is calculated based on the baseline energy efficiency ratio, the gain adjustment factor, and the energy efficiency ratio sensitivity coefficient, with the following expression:

[0071] ;

[0072] ;

[0073] ;

[0074] in, Indicates the corrected energy efficiency ratio; Indicates the benchmark energy efficiency ratio; Indicates the gain adjustment factor; This indicates the deviation of the refrigerant dryness. Indicates the energy efficiency ratio sensitivity coefficient; This indicates the compressor frequency.

[0075] S3. Construct a reinforcement learning decision model by inputting the corrected energy efficiency ratio and multi-source environmental equipment dataset into the reinforcement learning decision model to generate an optimized parameter combination.

[0076] S3.1 Define the state space based on the corrected energy efficiency ratio and multi-source environmental equipment dataset, identify the controllable parameters of the split air conditioner according to the split air conditioner design manual, define the action space based on the controllable parameters of the split air conditioner, and define the output layer based on the driver interface;

[0077] It should be noted that the corrected energy efficiency ratio, real-time collected indoor and outdoor temperatures, total harmonic distortion (THD) characteristics of the current, evaporator inlet and outlet temperature differences, and condenser inlet and outlet temperature differences are combined in a fixed order to form a 6-dimensional state vector. State space constraints are applied to each dimension of the state vector (corrected energy efficiency ratio 1.5 to 5.0, indoor temperature 16℃ to 32℃, outdoor temperature -7℃ to 42℃, THD characteristics 0% to 12%, evaporator inlet and outlet temperature differences 2℃ to 15℃, condenser inlet and outlet temperature differences 3℃ to 25℃), forming a closed bounded state space S. A linear normalization formula is used to map the corrected energy efficiency ratio and the multi-source environmental equipment dataset to the [-1,1] interval. According to the split air conditioner design manual, the controllable parameters of the split air conditioner are identified as the electronic expansion valve opening (0-100%) and the compressor frequency (10-120Hz). z); Based on the controllable parameters of the split air conditioner, the action space is defined as a two-dimensional continuous action vector [expansion valve opening increment, compressor frequency increment]. According to the flow control stability requirements and anti-liquid slugging protection mechanism of the electronic expansion valve, the expansion valve opening increment range is set to -5% to +5%. According to the torque response characteristics of the compressor motor and the mechanical stress limit of the bearing, the compressor frequency increment range is set to -10Hz to +10Hz. Based on the RS-485 drive interface protocol, the output layer is defined as two independent drive interfaces: electronic expansion valve drive interface and compressor drive interface. The electronic expansion valve drive interface converts the expansion valve opening increment into a 0-10V analog voltage signal to drive the electronic expansion valve through a linear mapping formula. The compressor drive interface converts the compressor frequency increment into a PWM duty cycle signal to drive the compressor inverter through a piecewise linear formula.

[0078] S3.2 Construct a reinforcement learning decision model based on the state space, action space, and output layer;

[0079] It should be noted that a policy network and a value network are constructed based on the state space and action space. The policy network takes a 6-dimensional normalized state vector as input and outputs the mean and standard deviation of a Gaussian distribution after passing through two fully connected hidden layers. The value network consists of an action value network and a state value network. The action value network takes an 8-dimensional concatenation of the state vector and action vector as input and outputs a scalar Q-value after passing through the same hidden layer. The state value network takes a 6-dimensional state vector as input and outputs a scalar V-value after passing through the same hidden layer. The weight matrices and bias vectors of the policy network and value network are initialized. The policy network uses a Gaussian distribution to sample the policy function of the actions, and the value network includes a state value function and an action value function. In each iteration of the training loop, the policy network randomly samples action outputs based on the current state space, executes physical control actions through the output layer interface, collects environmental feedback data, and stores the environmental feedback data in the experience replay buffer. It then randomly samples batches of samples from the buffer to calculate the temporal difference error, updates the state value network parameters by minimizing the value loss function, and simultaneously optimizes the Bellman equation to update the action value network parameters. Finally, it updates the policy network parameters through the policy gradient, synchronizing the target network parameters every 1000 steps, until the average reward for 10 consecutive training cycles converges to a stable variance threshold of less than 0.01, completing the training. The expression for the reinforcement learning decision model is...

[0080] ;

[0081] in, This indicates an optimized combination of parameters; State-action value function; Indicates the state of the environment; This represents the action vector.

[0082] S3.3. The modified energy efficiency ratio and multi-source environmental equipment datasets are merged into a state vector through the state space, and the original action parameters are generated through the action space. The optimized parameter combination is then generated through the output layer.

[0083] It should be noted that the corrected energy efficiency ratio is combined with real-time multi-source environmental equipment data in a fixed order to generate a 6-dimensional original state vector; predefined state space constraint rules are applied to perform boundary truncation on the original state vector; a linear normalization method is used to map the constrained original state vector to a standardized state in the interval [-1,1]; the standardized state is input into the trained policy network, and the normalized original action parameters [a1,a2] are output; inverse normalization calculation is performed based on the expansion valve opening increment range and the compressor frequency increment range to generate an optimized parameter combination [expansion valve opening increment, compressor frequency increment]; the optimized parameter combination is converted into a physical control signal through the output layer drive interface to execute closed-loop control.

[0084] S4. Adjust the compressor frequency according to the optimized parameter combination, and trigger the electronic expansion valve control based on the refrigerant dryness prediction value.

[0085] S4.1 Extract the values ​​of each parameter in the optimized parameter combination, convert the compressor frequency into a PWM signal that can be recognized by the split air conditioner control protocol, and transmit it to the inverter;

[0086] It should be noted that the compressor frequency increment is extracted from the optimized parameter combination; the current operating frequency of the compressor returned by the inverter is read through the Modbus-RTU protocol; the target frequency is calculated based on the compressor frequency increment and the current operating frequency; a piecewise linear transformation rule is applied: if the target frequency is ≤80Hz, the PWM duty cycle is (target frequency / 120)×100%; if the target frequency is >80Hz, the PWM duty cycle is 70%+0.75%×(target frequency-80); the PWM of the split air conditioner microcontroller unit is configured to generate a pulse signal with a 10kHz carrier frequency and a duty cycle parameter of Duty_cycle; the PWM control command is encapsulated in the Modbus-RTU protocol format through the RS-485 communication interface, and the Modbus-RTU protocol data frame is transmitted to the compressor inverter to perform speed control.

[0087] S4.2. Define the allowable deviation threshold according to the experimental calibration method, compare the refrigerant dryness deviation with the allowable deviation threshold, and trigger the electronic expansion valve control according to the judgment result.

[0088] It should be noted that the variable frequency split air conditioner under test was set up in an enthalpy difference laboratory conforming to ISO 5151 standards, and connected to a refrigerant circulation monitoring unit, vibration sensing element, and temperature and pressure acquisition unit; the liquid slugging critical point and superheating critical point were defined by the compressor body shear strength limit and the evaporator tube wall Biot number deterioration criterion, and the allowable deviation threshold was calculated to be ±0.05 based on the minimum absolute value of the two critical points; when the refrigerant dryness deviation... satisfy The expansion valve opening increment remains constant; when When this happens, a correction increment is generated. ,in Indicates the increment of the expansion valve opening; when When this happens, a correction increment is generated. Based on the judgment result, the correction increment will be input to the electronic expansion valve drive interface. Electronic expansion valve control is implemented, wherein, This indicates the opening value of the electronic expansion valve, which is obtained through decoding the valve position sensor signal.

[0089] S5. Real-time monitoring of temperature change rate and compressor power deviation, and triggering the transfer learning mechanism based on the monitoring results to update the LSTM neural network parameters and recalculate the refrigerant dryness prediction value and correct the energy efficiency ratio to generate the final optimized parameter combination.

[0090] S5.1. Collect real-time indoor temperature and compressor power, and calculate the temperature change rate and compressor power deviation;

[0091] It should be noted that the raw electrical signal of the indoor ambient temperature is collected by a PT1000 temperature sensor, processed into a digital temperature value by a 24-bit converter, and recorded in a time-series queue. Simultaneously, a Hall current probe is used to acquire the three-phase operating current waveform of the compressor, and the DC bus voltage is collected and read. The real-time power is obtained by multiplying the current and voltage every 5 seconds. The first and last temperature values ​​(earliest and latest recorded points) are extracted from the queue at 50-second intervals, and the temperature change rate is generated based on the ratio of the temperature difference to the time span of 45 seconds. Based on the compressor's operation under ISO 5151 standard conditions in the enthalpy difference laboratory, the compressor's rated power reference value is determined, and the compressor power deviation is obtained by calculating the real-time power and the compressor's rated power reference value.

[0092] S5.2 Define the temperature change rate threshold based on thermal comfort standards and the power deviation threshold based on compressor safety tolerance;

[0093] It should be noted that, based on the critical sensing equation for heat flux density of human skin, the maximum permissible rate of change of 0.003℃ / s corresponding to a thermal comfort quantification index of <10% is defined as the temperature change rate threshold; based on the thermal aging equation for motor stator winding insulation, ±10% of the compressor manufacturer's rated power is taken as the safety tolerance threshold boundary and the power deviation threshold.

[0094] S5.3 Determine the triggering conditions for the transfer learning mechanism based on the temperature change rate threshold and the power deviation threshold, and start the transfer learning mechanism based on the determination result;

[0095] It should be noted that the temperature change rate threshold and power deviation threshold are called, and the temperature change rate and compressor power deviation are read in real time. When it is detected that the absolute value of the temperature change rate is >0.003℃ / s and the compressor power deviation exceeds ±10% for 3 consecutive cycles, a 32-bit trigger command is sent to the interrupt controller. Upon receiving the 32-bit trigger command, the current compressor control thread is immediately interrupted, the compressor three-phase current waveform acquired by the Hall current probe is called, the evaporator wall temperature and exhaust pipe temperature are acquired by the PT1000 temperature sensor, and the compressor exhaust pressure is acquired by the pressure sensor.

[0096] The compressor three-phase current waveform is analyzed by FFT to extract the current frequency. The evaporator wall temperature is calculated by thermal resistance compensation to generate the evaporator refrigerant temperature. The actual superheat is calculated based on the exhaust pipe temperature and compressor exhaust pressure. The current frequency, evaporator refrigerant temperature and actual superheat are combined to generate a three-dimensional operating condition feature vector.

[0097] The historical optimization library is called, and the three-dimensional operating condition feature vector is compared with the historical operating condition feature vector stored in the historical optimization library using cosine similarity. When a historical operating condition feature vector with a similarity greater than 85% is matched, the historical PID gain parameter and historical compressor frequency limit are extracted from the historical operating condition feature vector. The real-time compressor power deviation and evaporator refrigerant temperature are called simultaneously. The new PID gain parameter is calculated based on the real-time compressor power deviation, and the new compressor frequency limit is calculated based on the evaporator refrigerant temperature.

[0098] The new PID gain parameters and new compressor frequency limits are transferred to the EEPROM control parameter table, and the compressor is driven to run synchronously. The compressor power deviation is continuously monitored within 30 seconds. If the compressor power deviation drops to within ±3%, the migration parameters are added to the historical optimization library. Otherwise, the parameters before migration are rolled back and the abnormal event code 0xE1 is recorded. If there is no situation where the absolute value of the temperature change rate is >0.003℃ / s and the compressor power deviation exceeds ±10% for 3 consecutive cycles, the real-time temperature change rate and compressor power deviation are reread.

[0099] S5.4. Update the LSTM neural network parameters using the stochastic gradient descent method, regenerate the refrigerant dryness prediction value, and update the corrected energy efficiency ratio.

[0100] It should be noted that when the transfer learning mechanism updates the compressor frequency limit, the maximum allowable speed range of the compressor changes. This change in speed limit directly affects the refrigerant circulation speed in the evaporator and condenser: increasing the compressor speed increases the refrigerant flow rate, while decreasing the speed reduces the refrigerant flow rate. This change in refrigerant flow rate alters the refrigerant evaporation rate in the evaporator and the refrigerant condensation efficiency in the condenser, causing changes in the evaporator inlet / outlet temperature difference and the condenser inlet / outlet temperature difference. The changed evaporator inlet / outlet temperature difference, condenser inlet / outlet temperature, indoor and outdoor temperatures, total harmonic distortion (THD) characteristics of the current, and the user-set temperature are used to generate a new six-dimensional input vector. After the LSTM neural network generates a refrigerant dryness prediction value based on this new six-dimensional input vector, the predicted refrigerant dryness value is compared with the optimal refrigerant dryness value, and the square of the difference is calculated as the loss. The backpropagation algorithm calculates the sensitivity of the loss to changes in the connection weights of the 128 neurons in the hidden layer, starting from the fully connected layers of the LSTM neural network, i.e., the contribution of each weight parameter to the loss. Following a preset learning rate of 0.001, each weight parameter is fine-tuned in the direction of reducing error, with the adjustment amount being the product of the sensitivity corresponding to each weight and the learning rate. During this process, 32 sets of continuous time-series data are used as a calculation batch each time, and the process is iterated 50 times to complete a single training iteration. The updated four types of weight parameters (input gate, forget gate, output gate, and memory cell weights) are permanently overwritten and stored in flash memory to complete the LSTM neural network parameter update. The updated LSTM neural network is then used to regenerate the refrigerant dryness prediction value, which is then substituted into the corrected energy efficiency ratio formula to regenerate the corrected energy efficiency ratio.

[0101] S5.5. Incorporate the temperature change rate and compressor power deviation into the multi-source environmental equipment dataset, and input the updated corrected energy efficiency ratio and the multi-source environmental equipment dataset into the reinforcement learning decision model to generate the final optimized parameter combination.

[0102] It should be noted that the temperature change rate and compressor power deviation are used as new features, and merged with the multi-source environmental equipment dataset (indoor and outdoor temperatures, total harmonic distortion of current, evaporator / condenser inlet and outlet temperature difference, and user-set temperature) to generate an eight-dimensional state vector; the updated corrected energy efficiency ratio is obtained and input into the reinforcement learning decision model along with the multi-source environmental equipment dataset; through... - A greedy strategy selects the optimal action vector; the optimal action vector is decoded by a fully connected hidden layer to generate the final optimized parameter combination of the expansion valve opening increment and the compressor frequency increment; the expansion valve opening increment is written to the electronic expansion valve control register, and the compressor frequency increment is written to the inverter address.

[0103] This embodiment also provides a computer device applicable to the energy calculation and control method for split air conditioners, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the energy calculation and control method for split air conditioners as proposed in the above embodiment.

[0104] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0105] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the method for calculating and controlling the energy regulation of a split-type air conditioner as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0106] In summary, this invention achieves high-precision indirect measurement by constructing an LSTM neural network to replace traditional sensors, thus improving adaptability to dynamic operating conditions. It also dynamically corrects the energy efficiency ratio by combining the optimal value defined by the refrigerant type, avoiding error accumulation caused by changes in operating conditions due to a fixed energy efficiency ratio. Furthermore, it improves the generalization ability of the LSTM network under sudden environmental changes by triggering a transfer learning mechanism through dual monitoring of temperature change rate and compressor power deviation, and by using stochastic gradient descent to update LSTM parameters online.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for calculating and controlling the energy of a split-type air conditioner, characterized in that: include, The collected multi-source environmental equipment dataset is input into the LSTM neural network to generate refrigerant dryness prediction values ​​and calculate the theoretical cooling capacity based on the refrigerant dryness prediction values. The multi-source environmental equipment dataset includes indoor and outdoor temperatures, compressor current and voltage, evaporator inlet and outlet temperature difference, condenser inlet and outlet temperature difference, and user-set temperature. The specific steps for generating the refrigerant dryness prediction value and calculating the theoretical cooling capacity based on the refrigerant dryness prediction value are as follows. Define a six-dimensional input layer, a first LSTM layer, a second LSTM layer, and a fully connected layer to construct an LSTM neural network; Input the multi-source environmental equipment dataset into the LSTM neural network to generate refrigerant dryness prediction values; The compressor power is calculated by using the compressor current and voltage, and the theoretical cooling capacity is calculated by correlating it with the predicted value of refrigerant dryness. The baseline energy efficiency ratio (EER) is calculated based on the theoretical cooling capacity, and then dynamically corrected according to the predicted refrigerant dryness value to generate the corrected EER. The specific steps are as follows: Based on the theoretical cooling capacity and compressor power, a benchmark energy efficiency ratio is generated. The refrigerant type is identified according to the split air conditioner design manual, and the optimal value of refrigerant dryness is defined based on the refrigerant type. Calculate the difference between the predicted refrigerant dryness value and the optimal refrigerant dryness value to generate the refrigerant dryness deviation, and define the attenuation coefficient according to the refrigerant type; The baseline energy efficiency ratio is adjusted based on the refrigerant dryness deviation and attenuation coefficient to generate the corrected energy efficiency ratio. A reinforcement learning decision-making model is constructed by inputting the corrected energy efficiency ratio and multi-source environmental equipment dataset into the model to generate an optimal parameter combination. The specific steps for generating the optimal parameter combination are as follows. The state space is defined based on the modified energy efficiency ratio and multi-source environmental equipment dataset; Identify the controllable parameters of the split air conditioner according to the split air conditioner design manual, define the action space based on the controllable parameters of the split air conditioner, and define the output layer based on the drive interface; A reinforcement learning decision model is constructed based on the state space, action space, and output layer; The modified energy efficiency ratio and multi-source environmental equipment datasets are merged into a state vector through the state space, and the original action parameters are generated through the action space. The optimized parameter combination is then generated through the output layer. The compressor frequency is adjusted according to the optimized parameter combination, and the electronic expansion valve is triggered based on the refrigerant dryness prediction value. The system monitors the temperature change rate and compressor power deviation in real time, and triggers a transfer learning mechanism based on the monitoring results to update the LSTM neural network parameters, recalculate the refrigerant dryness prediction value and correct the energy efficiency ratio, and generate the final optimized parameter combination.

2. The method for calculating and controlling the energy of a split-type air conditioner as described in claim 1, characterized in that: The steps for adjusting the compressor frequency based on the optimized parameter combination and triggering electronic expansion valve control based on the refrigerant dryness prediction value are as follows: Extract the values ​​of each parameter from the optimized parameter combination, convert the compressor frequency into a PWM signal that can be recognized by the split air conditioner control protocol, and transmit it to the inverter; The allowable deviation threshold is defined according to the experimental calibration method. The deviation of refrigerant dryness is compared with the allowable deviation threshold, and the electronic expansion valve is triggered based on the judgment result.

3. The method for calculating and controlling the energy of a split-type air conditioner as described in claim 1, characterized in that: The real-time monitoring of temperature change rate and compressor power deviation, and the triggering of a transfer learning mechanism based on the monitoring results, are described in the following steps. Collect real-time indoor temperature and compressor power, and calculate the temperature change rate and compressor power deviation; Temperature change rate threshold and power deviation threshold are defined based on thermal comfort standards and compressor safety tolerance. The triggering conditions for the transfer learning mechanism are determined based on the temperature change rate threshold and the power deviation threshold, and the transfer learning mechanism is activated based on the determination results.

4. The method for calculating and controlling the energy of a split-type air conditioner as described in claim 1, characterized in that: The specific steps for generating the final optimized parameter combination are as follows: The parameters of the LSTM neural network are updated using the stochastic gradient descent method to regenerate the refrigerant dryness prediction value and update the corrected energy efficiency ratio. The temperature change rate and compressor power deviation are merged into the multi-source environmental equipment dataset to generate a corrected multi-source environmental equipment dataset. The updated modified energy efficiency ratio and the modified multi-source environmental equipment dataset are input into the reinforcement learning decision model to generate the final optimized parameter combination.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the split air conditioning energy calculation and control method according to any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the split air conditioning energy calculation and control method according to any one of claims 1 to 4.

Images