Control method for air source heat pump unit

Abstract

The invention relates to the technical field of air source heat pump unit control, and discloses a control method for an air source heat pump unit. The method comprises the following steps: acquiring environment temperature and humidity data in real time, and performing noise filtering and time synchronization; generating a temperature prediction sequence based on the sliding window and the LSTM model; fluctuation characteristics are extracted through an optimization algorithm, and compressor control signals are generated; estimating the thickness of a frost layer in combination with a thermodynamic model and dynamically generating a defrosting instruction; closed-loop control parameters are generated through normalization processing and weight distribution; and finally, a system execution command is generated through multi-objective optimization, and compressor regulation and control and system health monitoring are achieved. The unit operation efficiency can be improved, the energy consumption can be reduced, and the defrosting control accuracy and the system stability can be enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of air source heat pump unit control, and particularly relates to a control method for an air source heat pump unit. BACKGROUND

[0002] In the technical field of air source heat pump unit control, the existing scheme consistent with the control method for an air source heat pump unit generally relies on single noise filtering and static temperature prediction model, and has limitations such as insufficient sensor data purification, insufficient temperature fluctuation feature extraction accuracy, and compressor frequency regulation lag. The existing method uses fixed threshold or linear regression model for temperature prediction, and in the scene of rapid environmental temperature fluctuation, prediction deviation accumulation and dynamic response lagging are prone to occur, and it is difficult to meet the stable implementation of compressor frequency dynamic regulation and energy consumption optimization. For the joint processing of environmental parameter acquisition, purified sensor data and predicted temperature sequence, the existing technology generally lacks a cooperative mechanism of composite filtering and entropy feature enhancement, and does not effectively integrate the dynamic modeling capability of long short-term memory model on time sequence features, resulting in a breakpoint in the generation link of predicted temperature sequence and compressor frequency instruction. In addition, in the conversion process of frequency optimization calculation and gradual adjustment parameters, the existing method is difficult to realize the closed-loop adaptation of gradient smoothing processing and device compatibility control, further aggravating the execution jitter of the compressor control signal and the energy loss in the scene of rapid temperature fluctuation. SUMMARY

[0003] The present application provides a control method for an air source heat pump unit to solve the problem of how to collect temperature and purified sensor data based on environmental parameters, and realize compressor frequency dynamic regulation and energy optimization in the scene of rapid environmental temperature fluctuation through composite filtering, entropy feature enhancement and LSTM model.

[0004] In order to solve the above technical problems, the present application provides a control method for an air source heat pump unit, comprising: Real-time collection of environmental temperature and humidity data, noise filtering, time stamp alignment and synchronization time synchronization by using multi-order filtering technology combined with statistical outlier rejection algorithm, and output of purified sensor data; extraction of statistical features and frequency energy features based on sliding window technology, generation of predicted temperature sequence by a pre-trained LSTM model; Based on the predicted temperature sequence, the fluctuation features are extracted by an improved particle swarm optimization algorithm, the frequency optimization calculation is performed according to the thermodynamic coupling coefficient, the gradient smoothing processing is performed by using a smooth transition function, and the device compatibility conversion is performed based on the Modbus protocol encapsulation function, and the compressor control signal is generated; The predicted temperature sequence and the compressor control signal are acquired, the frost thickness is estimated based on a physical mechanism-based thermodynamic model, the threshold is dynamically calculated and the safety range is checked in combination with the frost growth rate and historical defrosting cycle data, and a defrosting control instruction is generated; The normalization processing of the maximum minimum value normalization or Z-score standardization method is performed, the deviation feature extraction based on statistical analysis is performed, the energy consumption feedback coefficient is generated, the weight distribution processing is performed by using the multi-dimensional weight distribution algorithm, and the closed-loop control parameter is generated; Based on the closed-loop control parameter and the compressor control signal, the multi-objective optimization algorithm is used to perform multi-objective optimization operation, the joint control strategy is analyzed by using the rule engine and the priority sorting algorithm, and the system execution command is generated by encoding; The compressor frequency adjustment and the key performance indicator extraction are performed, the system health report is generated, and the sensor calibration parameter is detected.

[0005] Further, the multi-stage filtering technology is used in combination with the statistical outlier elimination algorithm to perform noise filtering, time stamp alignment and synchronization, and output purified sensor data, including: The noise filtering includes median filtering and weighted average filtering based on a moving window, and the outlying points exceeding the reasonable range are eliminated based on the standard deviation threshold; The multi-stage filtering technology is a composite filtering method combining median filtering and weighted average filtering; The statistical outlier elimination algorithm is based on a dynamic standard deviation threshold determination mechanism.

[0006] Further, the statistical features and frequency energy features are extracted based on the sliding window technology, and the predicted temperature sequence is generated by using the pre-trained LSTM model, including: The moving average line and the exponential weighted moving average method are used to capture the temperature change trend in a short period; The standardized feature parameters are dynamically adjusted according to the historical environmental data statistical distribution.

[0007] Further, the process of generating the predicted temperature sequence includes: The feature mapping and nonlinear relationship fitting are performed by using a multi-layer neural network structure or a tree model, and the predicted temperature sequence in a future short period is output; The weight parameters are dynamically adjusted to adapt to the real-time changes of the input data, and the uncertainty evaluation mechanism of the model is combined to give a confidence score to the prediction result.

[0008] Further, the process of extracting the fluctuation features based on the predicted temperature sequence by using the improved particle swarm optimization algorithm includes: The multi-objective optimization algorithm is used to perform multi-objective optimization operation on the compressor frequency; The multi-objective optimization algorithm includes a genetic algorithm, a particle swarm optimization algorithm or other heuristic algorithms, and comprehensively considers the compressor energy consumption, the starting frequency and the system response speed index to perform multi-objective optimization operation.

[0009] Further, the process of generating the compressor control signal includes the following steps: The gradient smoothing process is based on time series interpolation and weighted average method, and the progressive change curve of frequency adjustment is generated by calculating the difference between the current frequency instruction and the previous frequency parameter; the maximum frequency change gradient threshold is set to limit the maximum rate of frequency adjustment; during the smoothing process, the device feedback signal is monitored in real time, and the smoothing parameter is dynamically adjusted combined with the feedback information.

[0010] Further, the predicted temperature sequence and the compressor control signal are obtained, and the frost thickness estimation based on the thermodynamic model of the physical mechanism includes: The frost thickness estimation process includes comprehensive consideration of environmental temperature, humidity and compressor operating frequency, dynamic calculation of current frost thickness combined with historical defrosting records and frost layer attenuation model; The upper and lower limit boundary conditions of the frost thickness are set to prevent the estimated value from exceeding the physically reasonable range.

[0011] Further, the threshold dynamic calculation and safety range verification are performed combined with the frost layer growth rate and historical defrosting cycle data to generate the defrosting control instruction, which includes: The threshold dynamic calculation algorithm adopts a weight distribution mechanism to include the short-term volatility of environmental temperature prediction and the adjustment amplitude of compressor frequency into the threshold calculation model, forming a dynamic defrosting threshold with strong adaptability; The upper and lower limit boundaries of the threshold are set under the dual constraints of operation safety and energy consumption optimization.

[0012] Further, the expression of the predicted temperature sequence is generated through the pre-trained LSTM model, which includes:

[0013] wherein, is the purified temperature data; is the center index of the sliding window; is the sliding window size; is the adaptive weight of the first window; is the median filtering function; is the original temperature data; is the window half-width; is the index of the data in the window; Calculate frequency domain characteristics: in, Frequency domain energy characteristics; Index of frequency components; The number of main frequency components selected; For purification temperature data Frequency domain component vector after Fast Fourier Transform; For the first Power of each frequency component; The total power of all frequency components; Energy weighting coefficient; This represents the maximum amplitude value of the frequency domain component. Define the timing prediction output: in, To predict temperature series, from time arrive ; It is a long short-term memory network model; Let be the feature weight matrix; This is the matrix transpose. For bias terms; For the predicted duration; Temperature data after purification The arithmetic mean; Temperature data after purification The variance.

[0014] Furthermore, the expression for generating the compressor control signal includes: Define the calculation of frequency adjustment based on the improved particle swarm optimization algorithm: in, The optimized compressor frequency; ( ) represents the solution obtained using the particle swarm optimization algorithm; The frequency variable to be optimized; This refers to the current compressor frequency. The rate of temperature change; Thermodynamic coupling coefficient; and To optimize weights; The square of the L2 norm; Define a smooth transition function: in, The smoothed frequency function; It is a smoothing factor; is a slope coefficient; is a compressor frequency of a previous time; is a starting time stamp of frequency adjustment; represents a dynamic variable of a current time; is an integral variable; wherein, is a compressor control signal; is a cyclic redundancy check function; is a Modbus protocol encapsulation function; is a time stamp of instruction generation.

[0015] The key innovations of the present application include: (1) Through the composite filtering and entropy feature enhancement mechanism, the noise filtering processing of the purified sensor data is improved, and the accuracy and stability of the environmental parameter acquisition are improved.

[0016] (2) The long short-term memory model is used to infer the standardized feature parameters, enhance the dynamic modeling ability of the predicted temperature sequence generation, and ensure the accurate extraction of temperature fluctuation characteristics.

[0017] (3) In the frequency optimization calculation, the gradient smoothing processing and device compatibility conversion collaborative mechanism are introduced, the generation and execution link of the compressor frequency instruction are optimized, and the stability and compatibility of the control signal are ensured.

[0018] The main beneficial effects are as follows: (1) Through the composite filtering and entropy feature enhancement mechanism, the purified sensor data can more accurately reflect the environmental parameters after noise filtering processing, which improves the problem of insufficient sensor data purification, and makes the subsequent time series feature extraction more reliable.

[0019] (2) The application of the long short-term memory model improves the dynamic modeling ability of the standardized feature parameters in the predicted temperature sequence generation, ensures that the predicted temperature sequence generation link remains coherent in the scene of rapid environmental temperature fluctuation, and solves the problem of insufficient temperature fluctuation feature extraction accuracy.

[0020] (3) Through the gradient smoothing processing and device compatibility conversion collaborative mechanism, the compressor frequency instruction realizes signal stability and device compatibility in the generation and execution process, solves the problem of difficult gradient smoothing processing and device compatibility control in frequency optimization calculation, and reduces the execution jitter and energy loss of the compressor control signal. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1A flowchart of a control method for an air source heat pump unit is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0022] Embodiment one: reference Figure 1 A flowchart of a control method for an air source heat pump unit is provided for the embodiments of the present application, which flowchart can at least include steps S100-S600: S100, real-time acquisition of environmental temperature and humidity data, noise filtering and time synchronization of timestamp alignment and synchronization using multi-stage filtering technology combined with statistical outlier rejection algorithm, output of purified sensor data; extraction of statistical features and frequency energy features based on sliding window technology, generation of predicted temperature sequence through pre-trained LSTM model; S200, extraction of fluctuation features based on the predicted temperature sequence through the improved particle swarm optimization algorithm, frequency optimization calculation according to the thermodynamic coupling coefficient, gradient smoothing processing using a smooth transition function, and device compatibility conversion based on the Modbus protocol encapsulation function, generation of compressor control signals; S300, obtain the predicted temperature sequence and the compressor control signal, estimate the frost layer thickness based on the thermodynamic model of the physical mechanism, combine the frost layer growth rate and the historical defrosting cycle data, perform threshold dynamic calculation and safety range verification, and generate defrosting control instructions; S400, perform normalization processing of the maximum and minimum value normalization or Z-score standardization method, extract deviation features based on statistical analysis, generate energy consumption feedback coefficients, perform weight distribution processing using a multi-dimensional weight distribution algorithm, and generate closed-loop control parameters; S500, based on the closed-loop control parameters and the compressor control signals, perform multi-objective optimization operation using a genetic algorithm or a particle swarm optimization algorithm, analyze joint control strategies through a rule engine and a priority sorting algorithm, and encode system execution commands; S600, perform compressor frequency adjustment and key performance indicator extraction, generate a system health report, and detect abnormalities to generate sensor calibration parameters.

[0023] Step S100 at least includes steps S110-S130: S110, obtain temperature and humidity sensor data, perform noise filtering processing, and obtain purified sensor data; Specifically, the temperature and humidity data of the environment are collected in real time from the temperature and humidity sensor configured in the air source heat pump unit. The sensor collects data periodically through a preset sampling frequency. The collected raw sensor data contains noise and abnormal values that may exist. The collection process includes real-time monitoring of the sensor state. When an abnormal sensor or data loss is detected, an abnormal processing mechanism is triggered, the abnormal event is recorded, and a predefined compensation strategy is used for data filling to ensure data continuity and integrity. Further, the raw sensor data is used as input, and data preprocessing is performed according to a set noise filtering algorithm. Specifically, a multi-order filtering technique is combined with a statistical outlier elimination algorithm to identify and eliminate random noise and sudden abnormal values in the collected data. The multi-order filtering technique is a composite filtering method that combines median filtering and weighted average filtering. The statistical outlier elimination algorithm is based on a dynamic standard deviation threshold judgment mechanism. The noise filtering includes median filtering and weighted average filtering based on a moving window, combined with a standard deviation threshold judgment to eliminate outliers beyond a reasonable range, ensuring the stability and accuracy of the purified data. The filtered data is aligned and synchronized according to the timestamp to eliminate time offset in the data collection process, forming a purified sensor data set with consistent timing. The purified sensor data is used as the output field name of this step, which is used as input data for the subsequent step S120, ensuring the real-time and reliability of the environmental parameter data and providing basic data support for temperature trend prediction.

[0024] S120, extracting time sequence features from the purified sensor data, performing standardization processing, and generating standardized feature parameters; In step S120, the purification sensor data is inputted to firstly perform time series feature extraction. Specifically, a time window is constructed based on the purification sensor data, and statistical features of temperature and humidity are extracted by a sliding window technique, including but not limited to mean, variance, maximum, minimum, and change rate, etc. Further, a frequency domain analysis method such as Fast Fourier Transform (FFT) is used to mine periodicity features of the time series, and identify the periodic fluctuation rule of the environmental temperature. The feature extraction process also includes trend analysis of the data, and moving average and Exponential Weighted Moving Average (EWMA) methods are used to capture the temperature change trend in a short period. The extracted multi-dimensional features are standardized, and a normalization or Z-score standardization method is used to eliminate the influence of different feature dimensions, unify the numerical range of the features, and improve the generalization ability of the subsequent machine learning model. In the standardization process, the standardized feature parameters are dynamically adjusted according to the statistical distribution of the historical environmental data, to ensure the timeliness and adaptability of the feature parameters. The historical environmental data includes: running data in the last 30 days, original environmental parameters, data after purification, feature engineering data, equipment running state, and dynamically adjusted statistical parameters. The standardized feature parameters are outputted as the input features of step S130, and are used for subsequent temperature prediction model reasoning, to ensure that the temperature prediction model can perform efficient operation based on accurate and unified input features.

[0025] S130, machine learning model reasoning is performed on the standardized feature parameters to generate a predicted temperature sequence; In step S130, the standardized feature parameters are used as input features. Specifically, a pre-trained machine learning model is loaded. This model is constructed using supervised learning methods based on historical environmental temperature and humidity data and actual temperature measurements. Model types include, but are not limited to, Long Short-Term Memory (LSTM), Support Vector Machine (SVM), and Gradient Boosting Decision Tree (GBDT), with the best adaptability selected for inference. The model inference process includes batch processing of input features and time-series prediction. Internally, the model uses a multi-layer neural network structure or tree model to perform feature mapping and nonlinear relationship fitting, outputting a predicted temperature sequence for the next short period. During inference, the model dynamically adjusts weight parameters to adapt to real-time changes in input data. Combined with the model's uncertainty assessment mechanism, a confidence score is assigned to the prediction results to facilitate risk assessment by the subsequent control module. The predicted temperature sequence covers the temperature change trend over several future time steps, specifically in units of time steps, with resolution matching the sampling frequency. The predicted temperature sequence is passed as the output field name of this step to the input predicted data of step S210. It is used as the calculation input for compressor frequency dynamic control and is called in subsequent modules such as defrost trigger threshold optimization, forming the key data foundation for closed-loop control.

[0026] In another embodiment, in S110, a real-time data stream is received from a temperature and humidity sensor built into the air source heat pump unit. The sensor periodically generates data including temperature data at a preset sampling frequency of 5Hz. ( (raw temperature data) and humidity Raw sensor data ( (Original humidity data). To address impulse noise and drift anomalies in the data, a multi-stage filtering technique combined with a statistical outlier removal algorithm is employed: Formula ① defines a composite filter operator that fuses median filtering and weighted averaging.

[0027] in: The temperature data after purification; : The center index of the sliding window; The sliding window size (valued for a 15-second time window); For the first The adaptive weights of the window (range [0.6, 0.8]) are calculated by determining the standard deviation of the data within the window. Dynamic adjustment: ( (for preset thresholds) : No. Standard deviation of temperature data within the window; : Preset standard deviation threshold; Median filtering function; Raw temperature data; : Window half width; : Index of data within the window.

[0028] variable The data is generated by mapping the "ambient temperature" index of the (temperature and humidity sensor), and after being processed by formula ①, the output field "purification sensor data" is consumed by the "input data" of S120.

[0029] Furthermore, in S120, based on the purification sensor data from S110, a time window with a length of [missing information] is constructed. Extract the average temperature using a sliding window of seconds. ,variance And frequency domain energy characteristics. Formula ② uses the improved Shannon entropy to calculate frequency domain characteristics: in: Frequency domain energy characteristics; : Index of frequency components; The number of main frequency components selected; Purification temperature data Frequency domain component vector after Fast Fourier Transform (FFT); : No. Power of each frequency component; Total power of all frequency components; Energy weighting coefficient; : The maximum amplitude value of the frequency domain component.

[0030] Formula ② is obtained through training on the "temperature fluctuation pattern" index (historical operation database). The feature parameters are standardized by Z-score to generate the field "standardized feature parameters", which has a mean of 0 and a variance of 1, and is consumed by the "input features" of S130.

[0031] Furthermore, in S130, the standardized feature parameters are input into the pre-trained LSTM model for inference, and Equation ③ defines the temporal prediction output: in: Predicting temperature series, from time... arrive ; Long Short-Term Memory (LSTM) network model; : This is the feature weight matrix (3×64 dimensions); , , Standardization features derived from S120; This is the matrix transpose. For bias terms; The predicted duration is in minutes; Temperature data after purification The arithmetic mean; Temperature data after purification The variance.

[0032] The model parameters are obtained through training on "temperature-frequency correlation data" (historical temperature and humidity data and compressor logs). The output field "predicted temperature sequence" is consumed by the "input prediction data" of S210.

[0033] The technical effect of this section is to achieve noise suppression through composite filtering and entropy feature enhancement, and to complete high-precision temperature prediction by combining LSTM model, thus forming the data foundation for closed-loop control.

[0034] Step S200 includes at least steps S210-S230: S210. Extract fluctuation features from the predicted temperature sequence, perform frequency optimization calculations, and generate frequency adjustment instructions. In step S210, the predicted temperature sequence, as input prediction data, is first received by the control system and preprocessed to extract fluctuation characteristics. Specifically, the predicted temperature sequence covers the trend of environmental temperature changes over several future time steps. This sequence is segmented using a time window sliding technique. For the temperature data within each time window, its statistical fluctuation index is calculated, including but not limited to root mean square fluctuation amplitude, maximum rate of change, and number of local extrema. Further, combined with time series analysis methods, the autocorrelation function and partial autocorrelation function are used to calculate the periodic and trend characteristics of the temperature sequence, identifying the frequency components and amplitude changes of temperature fluctuations. During the fluctuation characteristic extraction process, the system filters abnormal fluctuations based on preset thresholds, eliminating abnormal fluctuation data caused by sensor errors or sudden environmental disturbances, ensuring the accuracy and representativeness of the fluctuation characteristics. This processing link also includes normalization processing of the fluctuation characteristics, unifying the dimensions of fluctuation indicators in different time periods to facilitate input compatibility with subsequent optimization calculation modules. The fluctuation characteristics serve as input parameters for frequency optimization calculation, entering the frequency optimization calculation stage, which performs frequency optimization calculations based on the thermodynamic coupling coefficient. The frequency optimization calculation is based on fluctuation characteristics and combined with the compressor operating state model. A multi-objective optimization algorithm is used to perform multi-objective optimization calculations on the compressor frequency. This multi-objective optimization algorithm includes Genetic Algorithm (GA), Particle Swarm Optimization (PSO), or other heuristic algorithms. It comprehensively considers indicators such as compressor energy consumption, start-up frequency, and system response speed to perform multi-objective optimization calculations and calculate the optimal frequency adjustment scheme. During the multi-objective optimization calculation process, the system monitors the compressor's current frequency and load status in real time and dynamically adjusts the optimization weights based on temperature fluctuation characteristics to ensure the timeliness and rationality of the frequency adjustment command. Furthermore, the frequency optimization calculation module has boundary constraints to limit the upper and lower limits of the compressor frequency adjustment, preventing the frequency from exceeding the equipment's safe operating range. The optimization result generates a compressor frequency adjustment command, which includes a specific frequency value and adjustment timing information. This compressor frequency command serves as the output field name for this step and is passed to the input command in step S220 for subsequent frequency command smoothing processing. Simultaneously, this command is also indirectly called by the defrost trigger threshold optimization module in subsequent step S300, forming a control closed loop.

[0035] S220: Perform gradient smoothing on the compressor frequency command to generate progressive adjustment parameters; In step S220, the compressor frequency command, as an input command, first enters the gradient smoothing processing module. Specifically, this module receives the frequency adjustment command from step S210. Based on the frequency value and adjustment timing in the command, the gradient smoothing processing uses a gradient smoothing algorithm to process the data, preventing sudden changes in the frequency command that could lead to compressor instability. The gradient smoothing processing is based on time series interpolation and weighted averaging. By calculating the difference between the current frequency command and the previous frequency parameter, it generates a gradual frequency adjustment curve. This process includes setting a maximum frequency change gradient threshold to limit the maximum rate of frequency adjustment, preventing excessively rapid frequency adjustment from causing mechanical shocks or a surge in energy consumption. During the smoothing process, the system monitors equipment feedback signals in real time, such as compressor speed, load current, and vibration sensor data. It dynamically adjusts the smoothing parameters based on the feedback information to ensure the smoothness and safety of the frequency adjustment process. Furthermore, when a significant deviation is detected between the frequency adjustment command and the current equipment state, an anomaly handling mechanism is triggered, recording the anomaly and activating a backup frequency adjustment strategy to prevent control command failure. The gradient smoothing processing generates gradual adjustment parameters, which include the filtered frequency adjustment curve and corresponding timestamp information. The progressive frequency parameter is passed as the output field name of this step to the input parameter of step S230 for subsequent device compatibility conversion and control signal generation. At the same time, this parameter is also used as a reference for the system closed-loop execution monitoring module in step S600 to realize real-time evaluation of the frequency adjustment execution effect.

[0036] S230: Perform device compatibility conversion on the asymptotic frequency parameters to generate compressor control signals; In step S230, the progressive frequency parameter is used as an input parameter. Specifically, a device compatibility conversion is first performed. This conversion process formats and encodes the frequency parameter for different models and communication protocols of the compressor controller in the air source heat pump unit, ensuring the compatibility and correct parsing of the control signal. Specifically, the numerical information in the progressive frequency parameter is mapped to the instruction format supported by the compressor controller, including the frequency setpoint, adjustment step size, and execution time point. During the conversion process, the system automatically selects the corresponding communication protocol, such as Modbus, CAN bus, or proprietary protocol, based on the device configuration file, and completes the encapsulation and verification of the instruction. Furthermore, the device compatibility conversion module integrates a security verification mechanism to verify the validity of the converted control signal, including instruction integrity verification, range legality judgment, and timing consistency confirmation, to prevent abnormal instructions from causing equipment failure. After the conversion is completed, an executable compressor control signal is generated and sent to the compressor control unit through the control bus. The compressor control signal contains specific frequency setting instructions and execution time control information. The compressor control signal, as the output field name of this step, is transmitted to the input control signal of step S310 for linkage control of the defrost trigger threshold optimization module. At the same time, it is called by the control parameter joint calculation module of step S500 to participate in multi-objective optimization calculation, form a joint control strategy, and constitute a closed-loop execution link for the control of the air source heat pump unit.

[0037] In another embodiment, in S210, based on the predicted temperature sequence output in S130, a sliding window is used to calculate the fluctuation amplitude index. Specifically, fluctuation features are extracted based on the predicted temperature sequence using an improved particle swarm optimization algorithm, and formula ④ defines the calculation of the frequency adjustment amount based on the improved particle swarm optimization algorithm: in: Optimized compressor frequency; ( ): Solved using the particle swarm optimization algorithm; : Frequency variables to be optimized; Current compressor frequency; : Rate of temperature change (obtained from the gradient of the "predicted temperature sequence", dimension °C / s); Thermodynamic coupling coefficient; and To optimize weights; It is the square of the L2 norm.

[0038] The formula uses a particle swarm optimization algorithm to find the optimal value within the frequency range [30Hz, 60Hz], and the output field "compressor frequency command" is consumed by the "input command" of S220.

[0039] Furthermore, in S220, the frequency command is subjected to gradient smoothing, and formula ⑤ defines the smooth transition function: in: : Smoothed frequency function; It is a smoothing factor; The slope coefficient; The compressor frequency at the previous moment; : The start timestamp of the frequency adjustment; Represents the dynamic variable at the current moment; it is the upper limit of the integration operation and marks the time point for calculating the smoothing frequency instruction. : Integral variable.

[0040] The integration limit is determined by the "Maximum Adjustment Rate" indicator in the "Device Response Characteristics Table". The output field "Asymptotic Frequency Parameter" is consumed by the "Input Parameters" of S230.

[0041] Furthermore, in S230, the asymptotic frequency parameters are encoded into device instructions, and formula ⑥ defines the protocol conversion function: in: : Compressor control signal; Cyclic Redundancy Check (CRC) function; CRC uses the CRC-16-IBM standard, and the checksum is defined by the "Communication Protocol Specification"; Modbus protocol encapsulation functions; : The timestamp when the instruction was generated.

[0042] The output field "Compressor Control Signal" is consumed by the "Input Control Signal" of S310.

[0043] Technical benefits of this section: Through multi-objective optimization and protocol conversion, smooth execution of frequency commands is achieved, ensuring the safety and responsiveness of equipment control.

[0044] Step S300 includes at least steps S310-S330: S310: Obtain the predicted temperature sequence and compressor control signal, estimate the frost thickness, and generate real-time frost data; In step S310, the predicted temperature sequence and the compressor control signal are used as inputs. Specifically, the system first receives the predicted temperature sequence from step S130 and the compressor control signal from step S230 through a communication interface. The predicted temperature sequence includes the trend of ambient temperature changes over several future time steps, and the compressor control signal includes the current compressor frequency setpoint and adjustment timing information. The system performs synchronization alignment processing on the input data, matching the predicted temperature sequence and the compressor control signal according to the timestamp to eliminate time offsets caused by differences in data acquisition frequency or communication delays, forming a time-consistent joint dataset. Further, based on the physical structural parameters and operating conditions of the air source heat pump unit, the system performs correlation analysis on the temperature changes and compressor frequency adjustments in the joint dataset. Specifically, it uses a multivariate regression model or a thermodynamic model based on physical mechanisms to estimate the frost thickness and the frost formation rate on the unit surface and heat exchanger. The frost thickness estimation process includes a comprehensive consideration of ambient temperature, humidity, and compressor operating frequency, combined with historical defrosting records and a frost decay model, to dynamically calculate the current frost thickness. During this process, the system sets upper and lower boundary conditions for frost thickness to prevent estimated values ​​from exceeding physically reasonable ranges. When abnormal data input is detected, such as a sudden change in the predicted temperature sequence or an abnormal control signal, an anomaly handling program is triggered to record the abnormal event and use historical averages for estimation compensation, ensuring the continuity and stability of frost state data. The estimation result is formatted as real-time frost data, including frost thickness value, estimation confidence level, and timestamp information. This real-time frost data serves as the output field name for this step, passed to the input data in step S320 for subsequent dynamic threshold calculation, and is indirectly called in the real-time energy consumption feedback analysis module in step S400 to achieve coordinated optimization of defrosting control.

[0045] S320. Extract key indicators from real-time frost data, perform dynamic threshold calculation, and generate dynamic defrosting thresholds. In step S320, the real-time frost data is used as input data and first enters the key indicator extraction module. Specifically, based on the real-time frost data, combined with the frost growth rate and historical defrost cycle data, the system calculates the key indicators for defrost triggering. The real-time frost data includes the difference between the current frost thickness and the threshold, the frost growth trend, and the remaining time of the defrost cycle. Further, the dynamic threshold calculation process employs a dynamic threshold calculation algorithm. Specifically, based on the fluctuation characteristics of the predicted temperature sequence and the adjustment trend of the compressor control signal, the defrost trigger threshold is adjusted in real time. The dynamic threshold calculation process incorporates the short-term fluctuations of the predicted ambient temperature and the adjustment amplitude of the compressor frequency into the threshold calculation model through a weight allocation mechanism, forming a highly adaptable dynamic defrost threshold. During the dynamic threshold calculation process, the system considers the dual constraints of equipment operation safety and energy consumption optimization, setting upper and lower limits for the threshold to prevent excessively low thresholds from causing frequent defrosting or excessively high thresholds from causing frost accumulation. The dynamic threshold calculation also incorporates energy consumption feedback data, adjusting the sensitivity of the threshold by adjusting the feedback coefficient in real time. An abnormal data processing mechanism operates synchronously. When real-time frost layer data shows abnormal fluctuations or the calculation results deviate from historical trends, the system records anomaly logs and uses smoothing filtering technology to correct the threshold, ensuring the stability and accuracy of the defrost trigger parameters. Finally, the dynamic defrost threshold is output in structured parameter form, including the defrost trigger threshold value, adjustment timestamp, and status identifier. This dynamic defrost threshold serves as the output field name for this step, passed to the input parameters of step S330 for subsequent safety range verification and defrost control command generation. It also provides a threshold reference for the energy consumption feedback analysis module in step S400, achieving multi-module data interaction and control closed loop.

[0046] S330: Verify the safety range of the dynamic defrost threshold and generate a defrost control command; In step S330, the dynamic defrost threshold, as an input parameter, is first processed by the safety range verification module. Specifically, the system compares the upper and lower limits of the dynamic defrost threshold according to preset equipment safety standards and environmental adaptability requirements to verify whether it is within a reasonable range. The verification process includes boundary judgment of the threshold value, limitation of the rate of change, and trend consistency analysis with historical thresholds. If the dynamic defrost threshold exceeds the safety range, the system automatically triggers the threshold correction mechanism, recalculates the threshold using historical stable thresholds or a prediction correction model based on machine learning, and records verification anomalies and correction results. Further, the system combines real-time frost layer data and compressor control signals to perform multi-factor fusion judgment, comprehensively considering the current frost layer state, ambient temperature prediction, and equipment operating load to generate a defrost control command. This command specifically includes the defrost start time, defrost duration, and defrost mode parameters, conforming to the equipment control protocol requirements. The generation process of the defrost control command adopts a combination of rule engine and optimization algorithm to ensure the rationality and execution efficiency of the command. After the command is generated, the system formats and encodes the command content, adds a timestamp and check code, forming a complete defrost control command. The defrost control command, as the output field name of this step, is passed to the input command of step S410 for subsequent energy consumption feedback analysis and closed-loop control execution. At the same time, this command is called by the control parameter joint calculation module of step S500 to realize the dynamic adjustment and optimization of the defrost strategy.

[0047] Step S400 includes at least steps S410-S430: S410: Collect compressor operating energy consumption data, perform normalization processing, and generate standard energy consumption indicators; In step S410, the defrosting control command, as an input command, is first received by the system's energy consumption feedback acquisition module. Specifically, the system collects real-time energy consumption data during compressor operation through an energy consumption metering device installed on the air source heat pump unit compressor. This energy consumption metering device uses a high-precision current sensor and a voltage sampling module to periodically collect data according to a preset sampling frequency. The collected data includes information such as instantaneous power, cumulative energy consumption, and operating time. The acquisition process includes real-time monitoring of the sensor status. When a sensor malfunction or data loss is detected, an abnormality handling mechanism is automatically triggered, recording the abnormal event and using a predefined compensation algorithm for data interpolation or reconstruction to ensure the continuity and integrity of the energy consumption data. Further, the raw energy consumption data is used as input, and the system preprocesses the data according to a normalization processing algorithm. Specifically, it uses a method based on Min-Max Normalization or Z-score standardization to convert energy consumption data under different time periods and operating conditions into unified energy consumption standard data, eliminating differences in data dimensions and amplitudes. During the normalization process, the system dynamically adjusts the normalization parameters, combines historical energy consumption statistics, and filters out abnormal peaks and sudden data changes to ensure the stability and representativeness of the energy consumption standard data. The normalization process includes sliding window filtering and weighted averaging of the collected data to mitigate the impact of short-term data fluctuations on the normalization results. This processing link also includes synchronizing the timestamps of the normalization results to ensure the timing consistency between the energy consumption standard data and the defrost control commands. The energy consumption standard data, as the output field name of this step, is passed to the input data of step S420 for subsequent deviation feature extraction and feedback intensity calculation. Simultaneously, this data is indirectly called by the control parameter joint calculation module in step S500, realizing the energy consumption feedback basis for closed-loop control.

[0048] S420. Extract deviation characteristics from energy consumption standard data, calculate feedback intensity, and generate energy consumption feedback coefficient. In step S420, the energy consumption standard data is used as input data. Specifically, the system first extracts deviation features from the normalized energy consumption standard data. The process of extracting these deviation features employs statistical analysis methods, comparing and calculating real-time energy consumption standard data with historical energy consumption benchmark models. This includes calculating key indicators such as energy consumption deviation rate, energy consumption fluctuation amplitude, and energy consumption trend changes. The deviation feature extraction module dynamically captures the changing characteristics of energy consumption standard data at different time scales using a sliding time window technique, combining moving average and exponentially weighted moving average (EWMA) algorithms to identify abnormal energy consumption fluctuations and long-term trends. Furthermore, the system employs anomaly detection algorithms, such as outlier identification based on standard deviation thresholds and anomaly classifiers based on machine learning, to filter energy consumption deviation features, eliminating abnormal data caused by sensor errors or external interference, ensuring the accuracy of energy consumption feedback coefficient calculation. The deviation features also include periodic analysis of the energy consumption standard data, using frequency domain analysis methods such as Fourier transform (FFT) to identify periodic components in energy consumption fluctuations, assisting in the timing adjustment of feedback regulation. Subsequently, based on the aforementioned deviation characteristics and a preset feedback intensity calculation model, the system generates the energy consumption feedback coefficient. This model comprehensively considers the magnitude, rate of change, and historical feedback effects of the energy consumption deviation, employing a weighted linear combination or nonlinear mapping function to calculate the feedback adjustment coefficient reflecting the current energy consumption state. During the feedback intensity calculation process, the system sets upper and lower limits to prevent excessively large adjustment coefficients from causing control parameter oscillations or excessively small adjustment coefficients from causing feedback lag. The dynamic adjustment mechanism adjusts the weight parameters of the adjustment coefficient in real time based on the real-time changes in energy consumption feedback, improving the sensitivity and stability of the feedback response. An anomaly handling mechanism operates synchronously; when deviation characteristic data or calculation results are abnormal, the system records the abnormal event and uses smoothing filtering technology to correct the adjustment coefficient, maintaining the continuity and reliability of feedback adjustment. The energy consumption feedback coefficient, as the output field name of this step, is passed to the input coefficient of step S430 for subsequent weight allocation processing and closed-loop control parameter generation. It is also called by the control parameter joint calculation module in step S500 to participate in the optimization calculation of the multi-objective control strategy.

[0049] S430: Perform weight allocation processing on the energy consumption feedback coefficients to generate closed-loop control parameters; In step S430, the energy consumption feedback coefficient is used as an input coefficient. Specifically, the system first performs weight allocation processing. This processing module combines the defrosting control command from step S330 and the compressor control signal from step S230, and uses a multi-dimensional weight allocation algorithm to perform weight allocation processing, integrating the energy consumption feedback coefficient with the defrosting strategy and compressor frequency adjustment parameters. The multi-dimensional weight allocation algorithm coordinates the weight allocation of the three types of control parameters through a dynamic priority model, specifically including: weight vector construction, dynamic adjustment mechanism, boundary constraints, and anomaly handling. The weight allocation processing is based on a preset priority model, considering the energy efficiency of system operation, equipment safety, and defrosting effect, and dynamically adjusts the weights of various feedback parameters to form a comprehensive control weight vector. Specifically, the system uses weighted average, fuzzy logic control, or multi-objective optimization methods to achieve reasonable allocation and coordination of different feedback signals. During the weight allocation process, the system sets weight boundaries to avoid a single feedback signal excessively affecting the overall control parameters and ensure the balance of the control strategy. Furthermore, the system combines real-time operating status monitoring data to dynamically correct the weight allocation results, adapting to changes in ambient temperature and fluctuations in equipment performance. Subsequently, based on the weight allocation results, the system generates closed-loop control parameters. This generation process employs a multi-parameter fusion algorithm, mapping the energy consumption feedback coefficient, defrost control command, and compressor control signal to the closed-loop control parameter space, forming the input for subsequent joint calculations of control parameters. The closed-loop control parameters include key control indicators such as energy consumption adjustment weights, defrost trigger sensitivity, and compressor frequency adjustment amplitude, all accompanied by timestamps and status identification information. During generation, the system performs parameter consistency verification and range limitation to prevent control parameters from exceeding the safe operating range of the equipment. An anomaly handling mechanism monitors abnormal fluctuations in the closed-loop control parameters, records anomaly logs, and initiates a parameter rollback strategy to ensure system stability. The closed-loop control parameters, as the output field names of this step, are passed to the input parameters of step S510 for multi-objective optimization calculations in the joint calculation module of control parameters, and simultaneously fed back to the system closed-loop execution monitoring module in step S600 to achieve real-time tracking and adjustment of the closed-loop control effect.

[0050] Step 500 includes at least steps S510-S530: S510: Obtain closed-loop control parameters and compressor control signals, perform multi-objective optimization calculations, and generate a joint control strategy; In step S510, the closed-loop control parameters and compressor control signal are used as inputs. Specifically, the system first receives the closed-loop control parameters from step S430, which include key control indicators such as energy consumption adjustment weight, defrost trigger sensitivity, and compressor frequency adjustment amplitude. Simultaneously, it receives the compressor control signal output from step S230, which includes the frequency setpoint, adjustment step size, and execution time information. The input data undergoes preliminary verification to confirm data integrity and timestamp consistency, eliminating anomalies caused by communication delays or data loss. Subsequently, the system maps the closed-loop control parameters and compressor control signal to a multi-objective optimization calculation module. Specifically, a multi-objective optimization algorithm, such as a genetic algorithm, particle swarm optimization algorithm, or other heuristic optimization methods, is used to perform multi-objective optimization calculations. By balancing multiple objectives such as energy efficiency, equipment safety, defrosting effect, and system response speed, the optimal joint control strategy is calculated. The optimization algorithm first normalizes the input parameters to eliminate dimensional differences between different control indices. Then, it constructs the objective function and constraints, including the safe operating range of the compressor frequency, the upper and lower limits of the defrost trigger threshold, and the adjustment boundary of energy consumption feedback. During algorithm iteration, the system monitors the optimization status in real time, uses a fitness function to evaluate the performance of each generation of strategies, and dynamically adjusts the weight allocation to ensure the convergence and stability of the optimization process. Furthermore, for abnormal data or unreasonable strategies that occur during the optimization process, the system has an anomaly detection mechanism that automatically removes abnormal individuals and records anomaly event logs to ensure the reliability of the optimization results. After the multi-objective optimization operation is completed, a joint control strategy is generated, including energy consumption adjustment weights, defrost sensitivity, and compressor frequency adjustment amplitude. This strategy is output in structured data form, with timestamps and status identification information. The joint control strategy, as the output field name of this step, is passed to the input strategy in step S520 for parsing the parameter adjustment sequence. It is also called by the system closed-loop execution monitoring module in step S600 to achieve real-time tracking and feedback of the joint control strategy's execution effect.

[0051] S520: Parse the execution priority from the joint control strategy and generate a parameter adjustment sequence; In step S520, the joint control strategy serves as the input strategy. Specifically, the system first parses the strategy content, extracting the execution priority and interdependencies of each control indicator. Based on a preset priority model and control logic, the parsing module uses a rule engine and priority sorting algorithm to refine and break down the energy consumption adjustment weights, defrosting sensitivity, and compressor frequency adjustment range in the joint control strategy, generating a parameter adjustment sequence. This sequence is sorted according to execution priority, clearly defining the adjustment order and time window for each parameter to ensure the orderly execution of control commands. During the parsing process, the system dynamically adjusts the priority weights based on the current operating status of the equipment and historical adjustment records, avoiding frequent switching that could cause equipment load fluctuations. Furthermore, the parameter adjustment sequence includes the adjustment range, adjustment rate, and execution time point for each parameter, using a timestamp synchronization mechanism to maintain consistency with the equipment control timing. The system performs a consistency check on the parsing results, verifying the logical integrity and execution feasibility of the parameter adjustment sequence to prevent conflicting or out-of-range instructions from the equipment. An anomaly detection mechanism operates concurrently. For any abnormal priority configurations or parameter exceedances detected during the parsing process, a predefined correction strategy is triggered to automatically adjust parameters or revert to the previous stable configuration, and an anomaly log is recorded. The parameter adjustment sequence is generated in structured data format, containing adjusted parameters, execution priority, and timestamp information. This parameter adjustment sequence serves as the output field name for this step and is passed to the input sequence for step S530 for the specific generation of device instruction encoding. Simultaneously, it is invoked by the system closed-loop execution monitoring module in step S600 to track the execution status and effect of parameter adjustments.

[0052] S530: Encode the parameter adjustment sequence into device instructions to generate system execution commands; In step S530, the parameter adjustment sequence is used as the input sequence. Specifically, the system first executes device instruction encoding based on device compatibility and communication protocol requirements. The encoding module maps the adjustment parameters, execution priority, and timestamp information in the parameter adjustment sequence into a binary or hexadecimal code stream conforming to the instruction format of the air source heat pump unit controller, including an instruction header, parameter fields, checksum, and end character. During the encoding process, the corresponding communication protocol, such as Modbus, CAN bus, or a proprietary protocol, is automatically selected based on the device model and firmware version to complete the encapsulation and verification of the instruction. Furthermore, the encoding module integrates a security verification mechanism to perform integrity verification, range validity judgment, and timing consistency confirmation on the generated instruction, preventing abnormal instructions from causing device malfunctions or communication errors. After encoding, the instruction is sent to the compressor control unit and related actuators via the control bus, supporting segmented transmission and acknowledgment mechanisms to ensure reliable instruction delivery and execution. The system monitors the instruction transmission status and device feedback in real time. Combined with an anomaly detection mechanism, it automatically triggers retransmission or backup strategies for instruction transmission failures or execution anomalies, and records the abnormal events. The device instruction encoding generates a system execution command, which includes a complete control instruction code stream and execution timing information. This system execution command, as the output field name of this step, is passed to the input command of step S610 for instruction execution by the system closed-loop execution monitoring module. It also provides control basis for real-time device status monitoring in step S600, realizing the closed loop of closed-loop control.

[0053] Step S600 includes at least steps S610-S630: S610: Execute system commands to adjust compressor frequency and generate real-time equipment status; Step S610, which executes the system execution command to adjust the compressor frequency and generate equipment status data, specifically involves receiving the system execution command from step S530 as an input command. This command includes control instructions such as the compressor frequency setpoint, adjustment step size, and execution time point. The system transmits the system execution command to the compressor control unit within the air source heat pump unit via the control bus. The control unit performs frequency adjustment operations according to the instructions. Specifically, the compressor control unit parses the system execution command, extracts the frequency adjustment parameters and corresponding execution timing, and drives the compressor inverter to adjust its speed according to the command requirements, achieving dynamic changes in the compressor's operating frequency. During execution, the system monitors the compressor's operating status parameters in real time, including speed, load current, vibration signals, and temperature sensor data, and collects equipment status information through built-in sensors. The real-time equipment status data is fed back to the central control system through the control unit. The data acquisition module preprocesses the feedback signal, including signal filtering, outlier identification, and data synchronization processing, eliminating sensor noise and communication errors to ensure the accuracy and real-time performance of the real-time equipment status data. Furthermore, the system establishes a real-time equipment status database, recording the start and end times, adjustment range, and corresponding equipment operating parameters for each frequency adjustment, forming a detailed operation log. The anomaly detection module continuously monitors the real-time equipment status data, triggering an alarm mechanism and recording the abnormal event in cases of abnormal speed, overload, or vibration. The real-time equipment status, as the output field name of this step, is passed to the input data of step S620 for subsequent key performance indicator extraction and system health assessment. It also provides real-time operational data for the system closed-loop execution monitoring module in step S600, enabling dynamic feedback for closed-loop control.

[0054] S620: Extract key performance indicators from real-time device status and generate a system health report; Step S620, which extracts key performance indicators from the real-time equipment status and generates a system health report, specifically involves receiving the real-time equipment status output in step S610 as input data. The system first performs feature extraction and performance evaluation on the real-time equipment status data. The process of extracting key performance indicators includes calculating key parameters of compressor operation based on a predefined performance indicator system. These key performance indicators include, but are not limited to, average speed, speed fluctuation amplitude, average load current and its rate of change, vibration frequency and amplitude characteristics, and temperature rise rate of the temperature sensor. The extraction of key performance indicators employs time-domain and frequency-domain analysis methods, combined with sliding window technology to segment the data and extract performance characteristics at different time scales. Furthermore, the system combines historical equipment status data with statistical analysis and trend prediction models to identify long-term trends and potential anomalies in operating parameters. Based on the extracted key indicators, the performance evaluation module performs a health score on the equipment's operating status, using a multi-dimensional scoring model to comprehensively evaluate the compressor's mechanical, electrical, and thermal states. This scoring model combines empirical rules and machine learning algorithms to dynamically adjust scoring weights, adapting to equipment aging and changes in the operating environment. The health report generation module summarizes the key performance indicators and health scores to form a structured system health report. The report includes a description of the current equipment status, performance trend analysis, anomaly warnings, and maintenance recommendations. The report includes a timestamp and status identifier to ensure the timeliness and accuracy of the information. An anomaly detection mechanism operates concurrently, recording abnormal fluctuations in performance indicators and abnormal values ​​in the health score, triggering maintenance reminders and fault warnings. The system health report, as the output field name of this step, is passed to the input report in step S630 for subsequent anomaly detection processing and sensor calibration parameter generation. It also provides equipment health status information to the system closed-loop execution monitoring module in step S600, supporting closed-loop control status monitoring and adjustment.

[0055] S630: Perform anomaly detection processing on the system health report and generate sensor calibration parameters; Step S630 involves anomaly detection processing of the system health report to generate sensor calibration parameters. Specifically, the system health report output in step S620 is received as an input report. The system first conducts an in-depth analysis of the key performance indicators and anomaly warning information in the report. The anomaly detection module employs a multi-level detection strategy, combining statistical threshold judgment, machine learning anomaly identification, and a rule engine to detect abnormal fluctuations, drifts, and fault phenomena in sensor data during equipment operation. Specifically, by comparing the statistical distribution of key indicators in the current health report with historical normal operation data, the degree of indicator deviation is identified, and it is determined whether there is sensor error or failure. Further, the system analyzes the data consistency between different sensors and locates abnormal sensors through sensor redundancy comparison and cross-validation methods. The anomaly detection result triggers the calculation of sensor calibration parameters. The sensor calibration parameter generation module, based on the anomaly type and degree, uses an adaptive calibration algorithm to calculate the sensor's bias correction value, proportional correction coefficient, and dynamic adjustment parameters. The calibration process dynamically adjusts the calibration strategy in conjunction with equipment operating conditions and environmental changes to avoid over-calibration leading to data distortion. The system stores the sensor calibration parameters, along with timestamps and anomaly event records, in the sensor calibration parameter database, forming a sensor calibration history record. After the sensor calibration parameters are generated, they are fed back to the input parameters of step S110 via the communication interface to achieve calibration and adjustment of the temperature and humidity sensor data acquisition, forming a closed-loop control sensor data correction mechanism. An anomaly handling mechanism is executed simultaneously; for calibration failures or abnormal fluctuations, an anomaly log is recorded and a maintenance reminder is triggered to ensure the accuracy and reliability of the sensor data. The sensor calibration parameters, as the output field name of this step, are passed to the input parameters of step S110 for sensor data preprocessing in the environmental parameter acquisition and temperature prediction module, supporting sensor state optimization and stable system operation under closed-loop control.

Claims

1. A control method for an air source heat pump unit, characterized by, Comprise: Real-time acquisition of environmental temperature and humidity data, noise filtering using multi-stage filtering technology combined with statistical outlier rejection algorithm, time synchronization of timestamp alignment and synchronization, output of purified sensor data; Based on the sliding window technology, statistical features and frequency energy features are extracted, and the pre-trained LSTM model is used to generate the predicted temperature sequence; Based on the predicted temperature sequence, the improved particle swarm optimization algorithm is used to extract the fluctuation feature, the frequency optimization calculation is performed according to the thermodynamic coupling coefficient, the gradient smoothing processing is performed using the smoothing transition function, and the device compatibility conversion is performed based on the Modbus protocol encapsulation function, and the compressor control signal is generated; Obtain the predicted temperature sequence and the compressor control signal, estimate the frost thickness based on the physical mechanism of the thermodynamic model, combine the frost growth rate and the historical defrosting cycle data, perform threshold dynamic calculation and safety range verification, and generate defrosting control instructions; Perform normalization processing using the maximum and minimum value normalization or Z-score standardization method, extract bias features based on statistical analysis, generate energy consumption feedback coefficients, perform weight distribution processing using a multi-dimensional weight distribution algorithm, and generate closed-loop control parameters; Based on the closed-loop control parameters and the compressor control signal, a genetic algorithm or a particle swarm optimization algorithm is used to perform multi-objective optimization operation, a rule engine and a priority sorting algorithm are used to analyze the joint control strategy, and a system execution command is generated by encoding; Perform compressor frequency adjustment and key performance indicator extraction, generate system health report, and detect abnormal sensor calibration parameters.

2. The control method according to claim 1, characterized by, The noise filtering includes median filtering and weighted average filtering based on a moving window, and the outlying points exceeding the reasonable range are removed based on a standard deviation threshold; The multi-stage filtering technology is a composite filtering method combining median filtering and weighted average filtering; The statistical outlier rejection algorithm is based on a dynamic standard deviation threshold determination mechanism. Based on the sliding window technology, statistical features and frequency energy features are extracted, and the pre-trained LSTM model is used to generate the predicted temperature sequence, including:

3. The control method according to claim 1, characterized by, Use moving average and exponential weighted moving average methods to capture temperature trends in short periods; Adjust the standardized feature parameters dynamically according to the statistical distribution of historical environmental data. The process of generating the predicted temperature sequence includes:

4. The control method according to claim 1, characterized by, Map features and nonlinear relationships through a multi-layer neural network structure or tree model, output the predicted temperature sequence in the future short period; Adjust the weight parameters dynamically to adapt to the real-time changes of the input data, and give the prediction result a confidence score based on the uncertainty evaluation mechanism of the model. The process of extracting fluctuation features based on the predicted temperature sequence through the improved particle swarm optimization algorithm includes:

5. The control method according to claim 1, characterized by, Use a multi-objective optimization algorithm to perform multi-objective optimization operation on the compressor frequency; The multi-objective optimization algorithm includes genetic algorithm, particle swarm optimization algorithm or other heuristic algorithm, and comprehensively considers the compressor energy consumption, startup frequency and system response speed indicators to perform multi-objective optimization operation. ​ 6. The control method according to claim 1, characterized by, The process of generating the compressor control signal includes: The gradient smoothing process is based on time series interpolation and weighted average method, and the gradual change curve of frequency adjustment is generated by calculating the difference between the current frequency instruction and the previous frequency parameter; including setting the maximum frequency change gradient threshold to limit the maximum rate of frequency adjustment; during the smoothing process, real-time monitoring of device feedback signal, dynamic adjustment of smoothing parameters combined with feedback information.

7. The control method according to claim 1, characterized by, The process of estimating the frost thickness based on the thermodynamic model of physical mechanism includes: The frost thickness estimation process includes comprehensive consideration of environmental temperature, humidity and compressor operating frequency, dynamic calculation of current frost thickness combined with historical defrosting records and frost layer attenuation model; Set the upper and lower limit boundary conditions of frost thickness to prevent the estimated value from exceeding the physically reasonable range.

8. The control method according to claim 1, characterized by, Combined with the frost layer growth rate and historical defrosting cycle data, the threshold dynamic calculation and safety range verification are performed to generate the defrosting control instruction, including: The threshold dynamic calculation algorithm incorporates the short-term volatility of environmental temperature prediction and the adjustment amplitude of compressor frequency into the threshold calculation model through a weight distribution mechanism, forming a dynamic defrosting threshold with strong adaptability; Set the upper and lower limit boundary of the threshold under the dual constraints of operation safety and energy consumption optimization.

9. The control method according to claim 1, characterized by, The expression of the predicted temperature sequence generated by the pre-trained LSTM model includes: ； wherein, is the temperature data after cleaning; is the center index of the sliding window; is the sliding window size; is the first is the adaptive weight of the window; is the median filter function; is the original temperature data; is the window half-width; is the index of the data within the window; Calculate the frequency domain features: ； wherein, is a frequency domain energy feature; is an index of a frequency component; is a number of selected main frequency components; is a purified temperature data is a frequency domain component vector after fast Fourier transform; is a power of the th frequency component; is a total power of all frequency components; is an energy weight coefficient; is a maximum amplitude value of a frequency domain component; Define the time series prediction output: ； wherein, is a predicted temperature sequence from time to ; is a long short-term memory network model; is a feature weight matrix; is a matrix transpose; is a bias term; is a prediction horizon; is an arithmetic mean of the purified temperature data ; is a variance of the purified temperature data .

10. The control method according to claim 1, characterized by, The expression of the compressor control signal includes: Define the frequency adjustment amount calculation based on the improved particle swarm optimization algorithm: ； wherein, is the optimized compressor frequency; ( ) is solved by a particle swarm optimization algorithm; is the frequency variable to be optimized; is the current compressor frequency; is the temperature rate of change; is the thermodynamic coupling coefficient; and is the optimization weight; is the square of the L2 norm; Define the smoothing transition function: ； wherein, is the smoothed frequency function; is the smoothing factor; is the slope coefficient; is the compressor frequency at the previous time instant; is the start time stamp of the frequency adjustment; represents the dynamic variable at the current time instant; is the integral variable; ； wherein, is a compressor control signal; is a cyclic redundancy check function; is a Modbus protocol encapsulation function; is a timestamp for the instruction generation.

Images