Multi-target coordinated noise reduction control method and system for compressors and fans in refrigeration equipment

By constructing a multiphysics prediction model and a dynamic risk map, the compressor frequency and fan speed of the refrigeration equipment are dynamically adjusted, solving the problems of noise superposition and resonance in the refrigeration equipment, and achieving optimization of noise, temperature and energy consumption, adapting to different operating conditions and structural changes.

CN123065183APending Publication Date: 2026-11-10QINGDAO DELLWARE REFRIGERATION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611437249.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-16
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In existing refrigeration equipment, the independent control of compressors and fans can easily lead to noise superposition and resonance, making it difficult to dynamically balance noise, temperature and energy consumption. Moreover, existing noise reduction measures are costly, have poor adaptability, and cannot adapt to different operating conditions and structural changes.

Method used

By collecting multi-source operational data, a multi-physics prediction model is constructed, a dynamic risk map is generated, a multi-objective collaborative optimization model is established, and the compressor frequency and fan speed are dynamically adjusted to avoid high-risk speed combinations and optimize noise, temperature and energy consumption.

Benefits of technology

It achieves proactive avoidance of high-risk speed combinations under different operating conditions, suppression of low-frequency hum and narrow-band tone, shortening of temperature recovery time, reduction of energy consumption, and adaptation to different structural changes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123065183A_ABST
    Figure CN123065183A_ABST
Patent Text Reader

Abstract

This invention relates to the field of equipment control optimization technology, specifically providing a multi-objective collaborative noise reduction control method and system for compressors and fans in refrigeration equipment. The method includes: collecting multi-source operating data of the refrigeration equipment to construct an offline historical multi-source time-series dataset covering multiple operating conditions; training a multi-physics prediction model based on this dataset to predict future acoustic, temperature, and energy consumption responses and output prediction uncertainties; constructing a dynamic risk map based on the compressor order frequency, fan blade passing frequency, and cabinet structure resonance band, calculating the risks of frequency overlap, low-difference-frequency beat vibration, and structural resonance; using the prediction results and risk map as evaluation and filtering criteria, constructing a multi-objective function including noise and sound quality objectives, temperature and refrigeration safety objectives, and energy consumption objectives; solving for the optimal control vector under refrigeration safety constraints and issuing it for execution. This invention achieves active noise avoidance and multi-scenario adaptive collaborative optimization, reducing overall noise and energy consumption while ensuring refrigeration performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of intelligent equipment control technology, and in particular relates to a multi-target collaborative noise reduction control method and system for compressors and fans of refrigeration equipment. Background Technology

[0002] Commercial refrigeration equipment is widely used in convenience stores, supermarkets, fresh food stores, bakeries, and restaurants, and is often deployed in high-traffic areas, requiring continuous 24 / 7 operation. The mechanical vibrations, electromagnetic excitations, aerodynamic noise, and refrigerant pulsations generated by compressors, evaporators, and condensers within the confined cabinet space propagate through sheet metal, glass doors, air ducts, and the floor, creating low-frequency humming, periodic tones, and start-stop shocks that affect the acoustic environment and customer comfort. This noise is not caused by a single component: changes in the variable-frequency compressor's speed cause migration of the fundamental frequency, harmonics, and electromagnetic bands; the fan blade frequency changes with speed; and the cabinet structure has a resonant frequency band. When the compressor's order is close to or falls within the structural resonance band, even if the sound pressure level of a single component is not high, it will still produce prominent low-frequency tones due to spectral superposition, beat vibration, and structural amplification. Door opening, restocking, increased ambient temperature, frost formation, and changes in air duct resistance further increase the compressor and fan speeds, dynamically altering noise contribution and resonance risk with operating conditions.

[0003] Existing control systems mostly employ temperature-controlled start / stop, independent compressor frequency conversion, or fixed fan speed, with some implementing PID regulation for the compressor, evaporator fan, and condenser fan separately. Noise reduction primarily relies on sound insulation cotton, vibration damping pads, duct rectification, blade optimization, and low-noise components. While these measures are effective at the design condition or individual component level, the independent control of the compressor and fan easily leads to high-risk speed combinations with similar order and blade frequency, resulting in flapping vibrations and narrow-band tones. Single-point or average temperature feedback struggles to balance temperature uniformity with rapid recovery after door opening or restocking. Reducing compressor frequency or fan speed can reduce noise in the short term but may lead to insufficient heat exchange, increased condensing pressure, cabinet temperature exceeding limits, and prolonged recovery time. Sound absorption and vibration reduction measures increase cost, size, and maintenance difficulty, and are difficult to adapt to aging, loosening, duct blockage, and cabinet type differences. Furthermore, the priorities for noise, temperature, and energy consumption differ during peak business hours, nighttime, and restocking / frost periods, making it difficult for existing fixed-parameter or single-objective controls to dynamically balance these factors, and lacking the ability to identify the noise contribution of each component based on online sound and vibration data. Summary of the Invention

[0004] To address the problems existing in the prior art, the present invention provides a multi-target coordinated noise reduction control method for compressors and fans in refrigeration equipment, comprising the following steps: Step S1: Collect multi-source operating data of the refrigeration equipment, generate state vectors and actual control vectors based on the multi-source operating data, and form an offline historical multi-source time-series dataset under different operating condition combinations; the multi-source operating data includes at least the cabinet zone temperature, refrigeration pressure, compressor and fan operating parameters, acoustic and vibration data; Step S2: Train a multiphysics prediction model based on the offline historical multi-source time series dataset. The multiphysics prediction model takes the historical multi-source state sequence and candidate control vector as input, obtains the prediction response vector and outputs the prediction uncertainty. The prediction response vector includes at least acoustic response index, temperature response index and energy consumption response index. Step S3: Based on the compressor rotation frequency, fan blade passing frequency and cabinet structure resonance band, calculate the frequency overlap risk, low difference frequency beat vibration risk, structural resonance excitation risk and low frequency tone risk, and establish a dynamic risk map of compressor frequency-evaporator fan speed-condenser fan speed. Step S4: Construct a multi-objective collaborative optimization model, using the prediction results and prediction uncertainty of the multi-physics prediction model as evaluation criteria, and the dynamic risk map as a priori filtering condition for candidate actions. Construct a multi-objective function that includes at least noise and sound quality objectives, temperature and cooling safety objectives, and energy consumption objectives, and solve for the optimal control vector under the constraints of equipment, cooling, and speed. Step S5: Deploy the multiphysics prediction model, dynamic risk map, and multi-objective collaborative optimization model to the edge control platform, periodically output the optimal compressor frequency, optimal evaporator fan speed, and optimal condenser fan speed, and send them down for execution.

[0005] Preferably, the multiphysics prediction model in step S2 includes: The thermal-cooling state coding module is used to extract the spatial distribution characteristics of temperature and the thermal dynamic characteristics of the refrigeration system based on the historical multi-source state sequence, and output the thermal-cooling state code. The acoustic vibration order coding module is used to extract sound pressure, sound quality and rotational order features based on the historical multi-source state sequence, and output acoustic vibration state code. A control action encoding module is used to encode the candidate control vector and output the control action code; The cross-modal coupling fusion module is used to fuse the thermal-cooling state code, acoustic-vibration state code, and control action code to output a unified coupling feature; The multi-task prediction module is used to output the prediction response vector and the prediction uncertainty based on the unified coupling feature.

[0006] Based on the above scheme, the establishment of the offline historical multi-source time series dataset includes: The cabinet internal zone temperature includes data from multiple temperature measuring points arranged at different heights and horizontal positions within the cabinet; the refrigeration pressure includes the compressor suction pipe pressure and the exhaust pipe pressure; the compressor and fan operating parameters include at least the actual operating frequency of the compressor, the actual speed of the evaporator fan, the actual speed of the condenser fan, the compressor current, and the total power of the unit; the acoustic and vibration data include at least the overall sound pressure level, pitch prominence, and component vibration characteristics. Based on the multi-source operation data, state vectors and actual control vectors are established. The state vectors and actual control vectors at L time points are used to form historical time series inputs, and the response vectors of the next H control cycles are used as the supervision targets. By changing the ambient temperature, cabinet load, door opening frequency, compressor frequency, evaporator fan speed, and condenser fan speed, the state vector, actual control vector, and predicted response vector are made to cover multiple operating conditions, forming an offline historical multi-source time series dataset.

[0007] Specifically, the loss function of the multiphysics prediction model is:

[0008] in, Predicting loss for sound pressure level and sound quality. To predict losses based on zoned temperature and cooling safety indicators, For power and energy consumption prediction losses, To predict the uncertainty calibration loss, Loss due to physical consistency constraints; , , , and For the corresponding weights.

[0009] Specifically, the establishment of the dynamic risk map includes: Calculate the k-th order rotational frequency based on the compressor rotational speed, calculate the blade passage frequency of the evaporator or condenser based on the number of blades and rotational speed, extract the electromagnetic sideband by combining the number of motor pole pairs, drive switching frequency or measured order spectrum, and calculate the harmonic frequency and speed sideband. The compressor order, fan blade frequency, and side band that meet the preset threshold are used to form the characteristic frequency set of the current candidate speed combination. After the equipment is first installed or repaired, or within a preset period, a restricted frequency sweep is performed to identify the resonant center frequency and bandwidth of the cabinet structure, forming a set of resonant bands and corresponding transfer coefficients. Based on the frequencies in the set of characteristic frequencies, calculate the frequency overlap risk, low-difference frequency beat vibration risk, and structural resonance risk, and calculate the comprehensive interference risk; By traversing the feasible sets of compressor frequency, evaporator fan speed, and condenser fan speed, the comprehensive interference risk of each candidate speed combination is calculated to form the dynamic risk map.

[0010] Specifically, solving the multi-objective function in step S4 includes: Based on the allowable range of change of the actual control vector at the previous moment, a local candidate action is generated, and combined with the supplementary candidate velocity in the dynamic risk map to obtain a candidate control vector; The candidate control vector and the current historical multi-source state sequence are input into the multiphysics prediction model to obtain the predicted response vector; The candidate control vectors are filtered according to constraints, prediction uncertainty thresholds and interference risk thresholds, and the remaining candidate control vectors are sorted non-dominated according to a multi-objective function to form a Pareto candidate set. The optimal control vector is obtained by selecting a compromise solution from the Pareto candidate set based on dynamic weights.

[0011] Based on the above scheme, the multi-objective function includes: Noise and sound quality targets: ;in, To predict the A-weighted equivalent sound pressure level within the window, For low-frequency sound pressure levels, For pitch prominence, To comprehensively address the risks of intervention, For normalization operators; Temperature and refrigeration safety objectives: ;in, To predict the relative temperature of M temperature measurement points within the prediction window to the target temperature The mean absolute deviation For the temperature standard deviation of the zone, To predict the highest temperature, The upper limit of the allowable temperature, For refrigeration safety risks; Overall energy consumption for the forecast window: ; For the predicted compressor power, For evaporator fan power and This refers to the power of the condenser fan; Motion smoothness target: ; Here, W is the normalized control vector, and W is the positive semi-definite weight matrix. A comprehensive evaluation function is established based on each objective: , , , , , For dynamic weights.

[0012] Preferably, the method further includes: Obtain the response vectors after executing the optimal compressor frequency, optimal evaporator fan speed, and optimal condenser fan speed, and write the multi-source state sequence, actual control vector, and actual response vector into the online incremental dataset; trigger the feature fine-tuner to update according to a preset period or number of samples.

[0013] Preferably, step S5 further includes scenario-based control: The operating mode is identified based on the current door status, ambient temperature, highest temperature inside the cabinet, and running time. The operating modes include regular business mode, door opening mode, replenishment recovery mode, nighttime stable mode, defrosting mode, and fault protection mode. In the door opening mode and replenishment recovery mode, the temperature weight is increased and the compressor and fan are allowed to accelerate rapidly within the low-risk speed range, while limiting the sound pressure peak and speed abrupt changes. In the aforementioned standard operating mode, select a speed combination with low volume, low beat vibration, and sufficient temperature margin; In the nighttime stable mode, low-frequency sound pressure and overall energy consumption are reduced; In the defrosting mode, regular temperature control optimization is paused and the system operates according to the defrosting logic. After defrosting is completed, the set of candidate control actions is restored.

[0014] On the other hand, the present invention provides a multi-objective coordinated noise reduction control system for a refrigeration equipment compressor and a fan, used to implement the multi-objective coordinated noise reduction control method described above, the system comprising: The data acquisition module is used to collect multi-source operating data during the operation of the refrigeration equipment. The multi-source operating data includes at least the cabinet zone temperature, refrigeration pressure, compressor and fan operating parameters, acoustic and vibration data. The offline dataset construction module is used to generate state vectors, actual control vectors, and future response vectors based on the multi-source operating data, forming offline historical multi-source time-series datasets under different operating condition combinations. The multiphysics prediction model module is used to take historical multi-source state sequences and candidate control vectors as inputs, predict the response vectors for multiple future control cycles and output the prediction uncertainty. The response vectors include at least acoustic response indicators, temperature response indicators and energy consumption response indicators. The dynamic risk map construction module is used to calculate the frequency overlap risk, low-difference frequency beat vibration risk, structural resonance excitation risk and low-frequency tone risk based on the compressor rotation frequency, fan blade passing frequency and cabinet structure resonance band, and to establish a dynamic risk map of compressor frequency-evaporator fan speed-condenser fan speed. The multi-objective collaborative optimization module is used to take the prediction results and prediction uncertainty of the multi-physics prediction model as the evaluation basis, take the dynamic risk map as the candidate action prior filtering condition, construct a multi-objective function that includes at least noise and sound quality objectives, temperature and cooling safety objectives and energy consumption objectives, and solve the optimal control vector under the constraints of equipment, cooling and speed. The closed-loop control execution module is used to send the optimal control vector to the compressor inverter, evaporator fan driver and condenser fan driver for execution, collect actual operation data to form an online incremental dataset, and update the multiphysics prediction model online based on the online incremental dataset.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By constructing a dynamic risk map, this invention can predict the risks of frequency overlap, low-difference frequency beat vibration, and structural resonance under different control vectors before speed adjustment, enabling the control system to actively avoid high-risk speed combinations and suppress the generation of low-frequency hum and narrowband tone from the source. 2. The multi-objective function simultaneously constrains the highest temperature inside the cabinet, the average temperature deviation, the standard deviation of the zone temperature, the exhaust temperature, and the superheat, avoiding sacrificing heat exchange capacity for noise reduction and shortening the temperature recovery time after the door is opened or restocked. 3. By utilizing the actual sound and vibration transmission characteristics of the cabinet and updating the risk map and lightweight parameters of the model with online incremental data, it can adapt to different structures such as vertical display cabinets, glass door cabinets, and workbench cabinets, as well as dust accumulation in air ducts, loose installations, and component performance degradation. 4. Select more efficient action combinations based on actual working conditions to reduce unnecessary high air volume, high condensing pressure and frequent speed changes, thereby reducing daily overall energy consumption while meeting temperature and noise constraints. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the refrigerator control system and sensor setup of the present invention; Figure 2 This is an overall flowchart of the control method of the present invention; Figure 3 This is a schematic diagram of the multiphysics prediction model of the present invention; Figure 4 This is a flowchart of the risk map construction process for the present invention; Figure 5 This is a flowchart of the multi-objective optimization process of the present invention; Figure 6 This is a diagram of the noise reduction closed-loop control architecture of the present invention; Figure 7 This is a comparison chart of the dynamic response of the highest temperature inside the cabinet between the present invention and other methods; Figure 8A comparison of the 1 / 3 octave band noise spectrum between the fixed parameter temperature control and the method of the present invention; Figure 9 This is a Pareto distribution diagram for the multi-objective optimization of noise and energy consumption in this invention. Detailed Implementation

[0017] The invention will be further described below with reference to specific embodiments.

[0018] This invention provides a multi-objective coordinated noise reduction control method for compressors and fans in refrigeration equipment, such as... Figure 1 As shown, the present invention includes the following steps: Step S1: Collect multi-source operating data of the refrigeration equipment, generate state vectors and actual control vectors based on the multi-source operating data, and form offline historical multi-source time series datasets under different operating condition combinations; the multi-source operating data includes at least the cabinet zone temperature, refrigeration pressure, compressor and fan operating parameters, acoustic and vibration data; This invention takes vertical glass door refrigerators, vertical open display cabinets, or workbench refrigerators as the implementation objects. The refrigeration equipment specifically includes a variable frequency compressor, condenser, condenser fan, throttling device, evaporator, evaporator fan unit, internal air supply and return ducts, controller, and cabinet frame.

[0019] Multi-source operating data is acquired through zone temperature sensors, temperature sensors, pressure sensors, vibration sensors, microphones, compressor inverters, fan drivers, door switches, ambient temperature and humidity sensors, and power meters. The multi-source operating data specifically includes the cabinet zone temperature, evaporator temperature, suction and exhaust pressure, ambient temperature and humidity, door status, compressor frequency and current, evaporator fan speed, condenser fan speed, overall sound pressure level and sound quality, component vibration, and power data.

[0020] Specifically, such as Figure 2As shown, four layers of internal temperature measuring points are set along the height of the cabinet, located at approximately 20%, 40%, 65%, and 85% of the effective height of the cabinet from the bottom. For each height layer, one temperature sensor is installed on the left, middle, and right sides along the width of the cabinet, for a total of 12 internal zone temperature sensors to obtain the internal zone temperature, used to characterize the temperature distribution in different height and lateral areas within the cabinet. One temperature sensor is installed at the return air vent and the supply air vent, and one temperature sensor is installed at the evaporator inlet and outlet. Pressure sensors are installed on the compressor suction pipe and exhaust pipe, and vibration sensors are installed on the compressor housing or mounting base, the evaporator fan shroud, the condenser fan bracket, and the cabinet frame. A main microphone is installed at the front of the cabinet, 0.5m~1.5m from the outer surface and 1.2m~1.5m above the ground, at a distance of 1m, and a reference microphone is installed inside the equipment or on the back. Data from the compressor inverter, fan driver, door switch, ambient temperature and humidity sensor, and power meter are simultaneously read. Furthermore, the aforementioned sensors and controllers are connected to the edge control platform via CAN, RS485, and Ethernet.

[0021] State vectors and actual control vectors are established based on multi-source operating data. The state vectors and actual control vectors at L time points are used to form historical time series inputs, and the response vectors of the next H control cycles are used as the supervision targets. Specifically, at time t, the state vector and control vector are constructed:

[0022]

[0023] in, For state vectors, This represents the temperature vector for each zone within the cabinet. This refers to the evaporator-related temperature. and These are the intake pressure and the exhaust pressure, respectively. Superheat; and These are ambient temperature and relative humidity, respectively. This refers to the gate status and heat load status; This refers to the actual operating frequency of the compressor. and These are the actual speeds of the evaporator fan and the condenser fan, respectively. and These are current and total power, respectively. and These are the A-weighted and C-weighted equivalent sound pressure levels, respectively. Tonal prominence; , and The vibration characteristics are for the compressor, evaporator fan, and condenser fan, respectively.

[0024]

[0025] in, For control vectors, , and These represent the target compressor frequency, target evaporator speed, and target condenser speed to be executed, respectively. The future response vector is obtained based on the state vector and control vector:

[0026] in, This is the response vector for the next H control cycles, which includes, in sequence, the overall sound pressure level and sound quality indicators, the temperature of each zone within the cabinet and its maximum temperature deviation, the exhaust temperature, the superheat, the compressor power, the evaporator fan power, the condenser fan power, and the overall energy consumption. It is important to note that this future response vector is derived from the state vector and the control vector, and corresponds to both.

[0027] An offline historical multi-source time-series dataset is formed based on the collected multi-source operational data, specifically: Acoustic and vibration data are acquired at sampling frequencies ranging from 6.4kHz to 25.6kHz, while temperature, pressure, rotational speed, current, and power data are acquired at sampling periods of 0.5 to 5 seconds. The edge control platform records the timestamps of each data point using a unified clock and frames the acoustic and vibration data into sliding windows ranging from 0.5 to 2 seconds, mapping the acoustic and vibration characteristics of each frame to the corresponding temperature, pressure, rotational speed, current, and power data within that time range. Simultaneously, resampling, interpolation, or window statistics are performed on data from different sampling frequencies. Missing values ​​are masked, and event tags are set for data related to sensor saturation, communication interruption, manual maintenance, and abnormal shutdown.

[0028] Furthermore, based on the main microphone signal, the system calculates the A-weighted equivalent sound pressure level, peak sound pressure level, C-weighted equivalent sound pressure level, low-frequency sound pressure level from 20Hz to 500Hz, 1 / 3 octave band sound pressure level, narrowband peak, tone prominence, loudness, or roughness, among other acoustic quality characteristics. For the vibration signal, it calculates the root mean square, peak value, kurtosis, order amplitude, and microphone-vibration coherence (acoustic-vibration coherence). Based on the real-time rotational speed, the frequency spectrum is converted into an order spectrum to extract the compressor order frequency, fan blade passing frequency, and octaves. For the cabinet temperature, the average, maximum, minimum, zone standard deviation, supply and return air temperature difference, and duration of exceeding limits are calculated. Based on the power meter or compressor and fan power models, the system calculates the comprehensive energy consumption per unit control cycle.

[0029] The historical time series input is constructed by taking the state vector and control vector from the current moment and the previous L moments, and the response vector from the next H control cycles is used as the supervision target. Among them, L and H are determined based on the thermal inertia of the refrigeration system, the control cycle, and the prediction error of the validation set. Specifically, L can be 12~60, and H can be 3~12.

[0030] Furthermore, while ensuring equipment safety and refrigeration temperature requirements, the system modifies ambient temperature, cabinet load, door opening frequency, compressor frequency, evaporator fan speed, and condenser fan speed to cover multiple operating conditions, including low load, rated load, high ambient temperature, frequent door opening, concentrated replenishment, evaporator frosting, defrosting recovery, and component aging. Layered random or Latin hypercube sampling is used for the compressor and fan speed combinations to increase sample density near frequencies, low difference frequencies, and structural resonances. Training, validation, and test sets are created based on equipment number, operating date, and operating condition, forming an offline historical multi-source time-series dataset. .

[0031] Step S1 establishes a multi-source synchronous acquisition of operating data for the compressor, evaporator fan and condenser fan. The data sources include existing operating records, controlled calibration tests and control vector sampling within a safe range.

[0032] Furthermore, in order to assess the comprehensive impact of different compressor-fan speed combinations (control vectors) on overall noise, temperature field, refrigeration safety, and energy consumption (response vectors) before the execution of control actions, this invention constructs a multiphysics prediction model oriented towards multiple sampling rates, multi-source coupling, and candidate action conditions, taking into account the characteristics of high sampling rate of acoustic and vibration signals but low sampling rate of thermal signals, rotational order shifting with rotational speed, significant lag in refrigeration response, candidate actions not yet actually executed, and inconsistent number of sensors for different cabinet types.

[0033] Step S2 involves training a multiphysics prediction model based on an offline historical multi-source time-series dataset. The model takes the historical multi-source state sequences and candidate control vectors from the dataset as input, predicts the response vector, and outputs the prediction uncertainty. The response vector includes at least acoustic response indices, temperature response indices, and energy consumption response indices. The multiphysics prediction model includes a thermo-cooling state encoding module, an acoustic-vibration order encoding module, a control action encoding module, a cross-modal coupling fusion module, and a multi-task prediction module.

[0034] like Figure 3 As shown, step S2 specifically includes: S2.1: First, from offline historical multi-source time series datasets... The model reads historical multi-source state sequences, actual control vectors, candidate control vectors, and future response vectors. Temperature, pressure, speed, power, and acoustic / vibration characteristics are Z-score standardized using the mean and standard deviation calculated solely from training set statistics. Logarithmic amplitude transformation is applied to the acoustic and vibration order spectra, and the same physical orders are aligned to a fixed-order grid based on the real-time speeds of the compressor and fan. Equipment configuration codes and effective masks are set for data with different cabinet types, numbers of sensors, or numbers of fans, enabling the same model to handle various commercial refrigeration equipment structures. Simultaneously, the validation and test sets use only the standardized parameters determined by the training set to avoid data leakage.

[0035] S2.1: Thermal-Refrigeration State Encoding Module, used to extract temperature spatial distribution characteristics and thermal dynamic characteristics of refrigeration system based on historical multi-source state sequences, and output thermal-refrigeration state codes; Specifically, the temperature of each zone within the cabinet is considered as a graph node. Connections are established between nodes based on the supply-return air path, spatial adjacency, and relationship with the same shelf level. Graph attention networks or graph convolutional networks are used to extract the spatial distribution features of temperature. These spatial distribution features, along with evaporator temperature, suction and exhaust pressure, superheat, environmental conditions, door status, and historical power, form the thermal-cooling time sequence input. The thermal-cooling time sequence is then processed through one-dimensional causal convolution, dilated convolution, and a gated cycle unit to extract short-term heat load changes, refrigeration system hysteresis, and temperature recovery features after door opening. The thermal-cooling state code is then output. It also sets up a first feature fine-tuner for lightweight online adaptation after deployment.

[0036] S2.3: Acoustic vibration order coding module, used to extract sound pressure, sound quality and rotation order features based on historical multi-source state sequences, and output acoustic vibration state codes; Specifically, the A-weighted equivalent sound pressure level, C-weighted equivalent sound pressure level, low-frequency sound pressure level, 1 / 3 octave band sound pressure level, tone prominence, compressor order frequency, fan blade frequency amplitude, vibration characteristics, and sound vibration coherence are combined to form the sound vibration time sequence input. Narrowband peak values, low-frequency envelopes, and modulation features are extracted using multi-scale one-dimensional convolution. Learnable weights are assigned to the compressor order frequency, evaporator fan blade frequency / octave, and condenser fan blade frequency / octave using order attention. Short-time sound pressure transitions during start-up, shutdown, acceleration / deceleration, and speed crossover processes are captured using time attention to obtain the acoustic vibration state code. A second feature adjuster is also provided to adapt to changes in installation location, cabinet structure, and sensor sensitivity.

[0037] S2.4: Control action encoding module, used to encode candidate control vectors and output control action codes; Specifically, the candidate control vectors are normalized according to the allowable range of each execution quantity: ,in, , and These are the target frequency of the candidate compressor, the target speed of the evaporator fan, and the target speed of the condenser fan, respectively. The changes in the candidate control vector relative to the actual control vector at the previous moment, the compressor order frequency, the fan blade passing frequency, and the minimum difference between each characteristic frequency are calculated and input into a fully connected residual network to obtain the control action code. .

[0038] S2.5: Cross-modal coupling fusion module, used to fuse thermal-cooling state codes. Acoustic and vibration state coding and control action coding Furthermore, by calculating the correlation weights between thermal state, acoustic vibration order, and candidate actions through cross-attention, a unified coupling feature is formed. ; The S2.6 multi-task prediction module is used to output the prediction response vector and prediction uncertainty based on the unified coupling characteristics. Specifically, Multiple task prediction heads are input, and each outputs a response vector for the next H control cycles. The task prediction heads include sound pressure and sound quality prediction heads, temperature and cooling safety prediction heads, and power and energy consumption prediction heads. The response vectors include A-weighted sound pressure level, peak sound pressure level, low-frequency sound pressure level, tone prominence, temperature of each zone, maximum temperature deviation, temperature standard deviation, exhaust temperature, superheat, compressor power, fan power, and overall energy consumption. The uncertainty of each prediction index is also output.

[0039] When the candidate control vector is outside the historical data coverage or the sensor status is abnormal, the corresponding uncertainty increases, which is used to subsequently narrow the action adjustment range or trigger a safety backoff.

[0040] The loss function of the multiphysics prediction model consists of sound pressure and sound quality prediction loss, temperature and cooling safety prediction loss, power and energy consumption prediction loss, prediction uncertainty calibration loss, and physical consistency constraint loss, specifically:

[0041] in, Predicting loss for sound pressure level and sound quality. To predict losses based on zoned temperature and cooling safety indicators, For power and energy consumption prediction losses, To predict the uncertainty calibration loss, Loss due to physical consistency constraints; , , , and For the corresponding weights.

[0042] Specifically, physical consistency constraints include: temperature, pressure, and power output meeting reasonable equipment boundaries; fan power varying with rotational speed satisfying an approximate cubic similarity law; within a localized area with identical heat load and other conditions, increasing the evaporative airflow should not cause a sudden drop in heat exchange capacity; compressor frequency, suction and discharge pressures, and power satisfy an empirical monotonic relationship; and the predicted total power of the entire unit is not less than the sum of the compressor and fan power. The values ​​are determined based on the mean square error between the predicted and actual values ​​of A-weighted equivalent sound pressure level, peak sound pressure level, low-frequency sound pressure level, and pitch prominence within the prediction window. The value is determined based on the mean square error between the predicted values ​​of temperature, maximum temperature deviation, exhaust temperature and superheat of each zone within the prediction window and the corresponding actual values. The value is determined based on the mean square error between the predicted values ​​of compressor power, evaporator power, condenser power, and overall energy consumption within the prediction window and their corresponding actual values. It is used to reflect the degree of credibility between the prediction uncertainty corresponding to each prediction indicator and the actual value.

[0043] The low-frequency noise and pitch prominence of commercial refrigeration equipment are closely related to the rotational order of the compressor and fan, as well as the transmission characteristics of the cabinet. Therefore, this invention constructs a dynamic risk map to quickly screen out high-risk speed combinations. This dynamic risk map is based on... The spectrum parameters are recursively corrected based on newly added data during the online operation phase, using limited frequency sweep data.

[0044] Step S3: Based on the compressor rotation frequency, fan blade passing frequency and cabinet structure resonance band, calculate the frequency overlap risk, low difference frequency beat vibration risk, structural resonance excitation risk and low frequency tone risk, and establish a dynamic risk map of compressor frequency-evaporator fan speed-condenser fan speed. like Figure 4 As shown, the establishment of a dynamic risk map includes: S3.1: Calculate the k-th order rotational frequency based on the compressor's rotational speed: ;in This refers to the compressor's rotational speed. For the j-th evaporator or condenser, calculate the blade passage frequency based on the number of blades and the rotational speed: ,in, For the number of leaves, Rotational speed; Further, the main electromagnetic sidebands are extracted by combining the number of motor pole pairs, drive switching frequency, or measured order spectrum, and the corresponding harmonics and speed sidebands are calculated; the compressor order, fan blade frequency, and sidebands with amplitudes higher than a preset threshold are combined to form the characteristic frequency set of the current candidate speed combination. The preset threshold is 6dB of local background noise or exceeds the 90th percentile of the amplitude of the corresponding frequency band of the training data. It can be adjusted according to the sensor noise floor and spectral resolution.

[0045] S3.2: Perform a restricted frequency sweep after the initial installation or maintenance of the equipment, or within a preset period, and record the frequency response of vibration and sound pressure level of the main microphone at each installation point; identify the resonant center frequencies of the cabinet base, back panel, air duct, glass door, and condenser bracket based on the vibration peak value, acoustic coherence, and sound pressure amplification factor. and bandwidth Forming a collection of resonance zones and corresponding transfer coefficient Limited frequency sweep allows the compressor and fan to change speeds in segments within a safe range. Simultaneously, when frequency sweeping without stopping is not required, the resonant band can be recursively updated using natural excitation data from daily speed changes.

[0046] S3.3: Based on any two frequencies in the characteristic frequency set and Calculate the frequency overlap risk separately and low-difference-frequency beat vibration risk :

[0047]

[0048] in, and The magnitude is the corresponding order value normalized to [0,1]. For frequencies close to the scale; ε is the threshold for beat vibration determination; ε is a positive number to prevent the denominator from being zero. and It can be determined based on spectral resolution, the range of perceptible modulation frequencies, and subjective listening tests, according to a specific implementation method. and 1Hz~5Hz is acceptable. A frequency of 0.1Hz to 0.5Hz can be selected; [The following is a separate, unrelated sentence:] Using... make It remains dimensionless. This risk indicator can identify periodic amplitude modulation and humming that may occur when the compressor order is close to the fan blade frequency, when the blade frequencies of two fans are close, or when the side bands are close to each other.

[0049] For each frequency in the characteristic frequency set The structural resonance risk is calculated based on its distance from the center of each resonance zone and the normalized transfer coefficient.

[0050] in, Let be the vibration amplitude at the i-th characteristic frequency, and let be the frequency of the cabinet structure near the r-th resonance band. The vibration amplification factor, δ r The half-width of the resonance band. Let r be the center frequency of the r-th resonance band.

[0051] For narrowband peaks in the 20Hz~250Hz range, the low-frequency tone risk is calculated based on the difference between the peak sound pressure level and the background sound pressure level in adjacent frequency bands. ; Will , , and After normalizing to [0,1], the overall intervention risk is calculated:

[0052] in, The weights are non-negative and satisfy the following conditions: Each weight can be determined based on the cabinet type, objective acoustic indicators, and subjective listening tests.

[0053] Traverse the feasible discrete grid or candidate set of compressor frequency, evaporator fan speed, and condenser fan speed, and calculate the comprehensive interference risk for each candidate speed combination. This forms a dynamic risk map.

[0054] When the deviation between the online measured sound pressure level and the risk prediction exceeds the 95th percentile of the offline verification error for 3 to 5 consecutive evaluation windows, the order amplitude, resonance band, and transmission coefficient are recursively corrected.

[0055] By utilizing the amplitude-squared coherence between the vibration signals from the compressor mounting point, evaporator fan shroud, and condenser fan bracket and the main microphone signal, and combining the order amplitude and transmission coefficient of each component, the noise contribution coefficients of the compressor, evaporator fan, and condenser fan are calculated. , and .

[0056] Step S4: Construct a multi-objective collaborative optimization model, using the prediction results and prediction uncertainty of the multi-physics prediction model as the evaluation basis, and the dynamic risk map as the prior filtering condition for candidate actions. Construct a multi-objective function that includes at least noise and sound quality objectives, temperature and cooling safety objectives, and energy consumption objectives, and solve the optimal control vector under the constraints of equipment, cooling, and speed. Step S4 jointly optimizes the compressor frequency, evaporator fan speed, and condenser fan speed within a unified prediction window, such as... Figure 5 As shown, it specifically includes: S4.1: Construct noise and sound quality targets based on the future sound pressure level and sound quality indices predicted in S2:

[0057] in, To predict the A-weighted equivalent sound pressure level within the window, For low-frequency sound pressure levels, For pitch prominence, To comprehensively address the risks of intervention, As a normalization operator, different dimensional indices are normalized to [0,1] based on the device's allowable range or the quantiles of the training data; this objective avoids the problem of retaining obvious humming while only reducing broadband sound pressure by simultaneously considering broadband sound pressure, low-frequency sound pressure, pitch prominence, and interference risk.

[0058] S4.2: Assume there are M temperature measuring points inside the cabinet, and the target temperature is... Establish temperature and cooling safety goals: in, To predict the relative temperature of M temperature measurement points within the prediction window to the target temperature The mean absolute deviation For the temperature standard deviation of the zone, To predict the highest temperature, The upper limit of the allowable temperature, Refrigeration safety risks include excessively high exhaust temperature, insufficient superheat, and excessive suction and exhaust pressure or pressure ratio.

[0059] S4.3: Based on the predicted compressor power Evaporator fan power and condenser fan power Calculate the overall energy consumption for the prediction window: .

[0060] S4.4: Normalize the control vector according to the allowable range of each parameter. Then, construct a motion smoothness target:

[0061] in, The normalized control vector is represented by W, which is a positive semi-definite weight matrix. It can be a diagonal matrix of the importance weights of each normalized control quantity or constructed according to the inverse square of the allowable range of each control parameter. At the same time, additional penalties are set for frequent compressor start-stop, rapid fan speed reciprocation, and crossing of high-risk speed bands to reduce sudden changes in sound perception, actuator wear, and control oscillation.

[0062] S4.5: Will , , and After normalizing according to the allowable range of the project, construct the comprehensive evaluation function:

[0063] in, , , , , The weighting is dynamic. It increases when the door is open or the cabinet temperature approaches its upper limit. Increase the temperature when the cabinet temperature is stable and during peak business hours or quiet nighttime. Increase when cabinet temperature, sound pressure, and safety indicators are all within the allowable range. Temperature urgency can also be calculated based on the highest temperature inside the cabinet, door open status, restocking events, ambient temperature, business status, or presence of personnel. Harmony and environmental sensitivity The dynamic weights are adjusted based on the urgency of temperature and sensitivity to the acoustic environment. To prevent any target from being completely ignored, the lower limit of each weight can be set to 0.05~0.10, and the specific values ​​are determined through validation set and scenario experiments.

[0064] S4.6: Establish constraints: The control vector satisfies the compressor frequency upper and lower limits, the evaporator fan and condenser fan speed upper and lower limits, the maximum rate of change per unit time, the compressor minimum start time and minimum stop time, the fan minimum stable speed and the driver current limit. The prediction results meet the constraints of the highest temperature inside the cabinet, exhaust temperature, suction and exhaust pressure, pressure ratio and superheat. The overall interference risk of candidate velocity combinations must not exceed a preset safety threshold; During defrosting, sensor malfunction, or compressor protection mode, some control variables are locked according to the preset mode, and the optimizer is prohibited from changing related actions.

[0065] S4.7: Solving for the optimal control vector: Local candidate actions are generated based on the allowable range of change of the actual control vector at the previous moment, and candidate control vectors are obtained by combining them with supplementary candidate velocities in the dynamic risk map. The candidate control vector and the current historical multi-source state sequence are input into the multiphysics prediction model to obtain the predicted response vector, including the prediction results of noise, temperature, safety and energy consumption. Filter out candidate control vectors that violate constraints, have prediction uncertainties exceeding the threshold, or have excessively high interference risk thresholds. Then, classify the remaining candidate control vectors according to... , and Perform a non-dominated sort to form a Pareto candidate set; Based on dynamic weights, the optimal control vector is obtained by selecting the compromise solution closest to the ideal point of the current scene from the Pareto candidate set and performing a small amount of local refinement in its neighborhood.

[0066] S4.8: Optimal control vector Projected onto the current device's feasible control set This yields the executable security control vector:

[0067] In the formula, and These are the candidate control vector and the optimal control vector, respectively, normalized according to the allowable range. It is determined by the compressor start / stop logic, minimum fan air volume, speed change rate, temperature and pressure safety margins, and communication execution capability.

[0068] If the optimal control vector cannot be safely reached within the current control cycle, the direction from the previous control vector to the optimal control vector is truncated by the allowable step size. If there is no candidate control vector (candidate action) that satisfies all constraints, the safety baseline action is invoked based on the current heat load and cabinet type. The controller records the reasons for the optimal action, truncated action, or rollback action for subsequent model and constraint parameter updates.

[0069] Step S5: Deploy the multiphysics prediction model, dynamic risk map, and multi-objective collaborative optimization model to the edge control platform, periodically output the optimal compressor frequency, optimal evaporator fan speed, and optimal condenser fan speed, and send them down for execution.

[0070] Specifically, such as Figure 6 As shown in S5.1: The multiphysics prediction model, dynamic risk map, and multi-objective collaborative optimization model are deployed to the edge control platform of the refrigeration equipment. The edge control platform establishes communication interfaces with the compressor inverter, evaporator fan driver, condenser fan driver, temperature control board, pressure sensor, power meter, microphone, and vibration acquisition unit. Acoustic and vibration characteristics can be calculated at the acquisition end or the edge platform. Model inference and batch evaluation of candidate actions are performed on the edge platform. The platform saves parameters such as equipment configuration, number of blades, allowable speed range, cabinet resonance zone, temperature limit, and safety baseline actions.

[0071] S5.2: Collect the latest state data and update the historical time window of length L according to the preset control cycle. The preset control cycle is determined according to the thermal inertia of the refrigeration system and the response time of the actuator, and can be 2s~10s. The acoustic and vibration characteristics are updated using a sliding window of 0.5s~2s. Input the current historical state sequence and candidate actions into the multiphysics prediction model, and predict the noise, temperature, refrigeration safety and energy consumption response for the next H cycles. Write the measured and predicted errors into the online cache.

[0072] S5.3: Execute the executable safety control vector obtained in step S4, send the compressor target frequency to the inverter, and send the evaporator and condenser fan target speeds to the corresponding drivers, and read the actual arrival frequency, speed, and current. If the deviation between the command value and the actual value exceeds the allowable range, or if compressor protection, fan stall, abnormal door status, or sensor distortion occurs, stop the current optimization action and switch to the safety baseline strategy. After the action is executed, obtain the actual A-weighted and C-weighted sound pressure level, pitch prominence, component vibration, zone temperature, refrigeration safety indicators, and power / energy consumption, form the feedback for the next control cycle, and write it into the system. .

[0073] S5.4: Identify the operating mode based on the current door status, ambient temperature, highest temperature inside the cabinet, and running time. The operating modes include regular business mode, door opening mode, replenishment recovery mode, nighttime stable mode, defrosting mode, and fault protection mode. In door opening mode and replenishment recovery mode, temperature weighting is increased and the compressor and fan are allowed to accelerate rapidly within the low-risk speed range, while limiting sound pressure peaks and speed abrupt changes. In the standard operating mode, choose a speed combination with low pitch, low beat vibration, and sufficient temperature margin; In the nighttime stable mode, low-frequency sound pressure and overall energy consumption are reduced; In defrost mode, regular temperature control optimization is paused and the system operates according to the defrost logic. After defrosting is completed, a recovery action control is used to prevent the compressor and fan from crossing the resonance zone at the same time, which would cause short-term noise impact.

[0074] S5.5: Obtain the response vectors after executing the optimal compressor frequency, optimal evaporator fan speed, and optimal condenser fan speed, and write the multi-source state sequence, actual control vector, and actual response vector into the online incremental dataset. The feature fine-tuner in step S2 is triggered according to a preset time period or number of samples, and the order amplitude, coherence, resonance center and bandwidth in step S3 are recursively corrected using recent running data.

[0075] According to a specific implementation, candidate updates can be triggered after accumulating 500-5000 valid control samples or after continuous operation for 7-30 days. The updated model needs to be evaluated in parallel with the current model in shadow mode. Switching occurs when the prediction error, constraint satisfaction rate, and control stability are all better than the current model; otherwise, the original model is maintained. If the uncertainty of the prediction model output exceeds a threshold for 3-5 consecutive control cycles, or If an abnormal drift occurs, the online updates will be frozen and the system will revert to the most recently validated prediction model and risk map.

[0076] Based on the same technical concept, the present invention also provides a multi-objective collaborative noise reduction control system for a refrigeration equipment compressor and fan, used to implement the above-mentioned multi-objective collaborative noise reduction control method. The system includes: The data acquisition module is used to collect multi-source operating data during the operation of the refrigeration equipment. The multi-source operating data includes at least the temperature of each compartment inside the cabinet, the refrigeration pressure, the operating parameters of the compressor and the fan, and acoustic and vibration data. The offline dataset construction module is used to generate state vectors, actual control vectors, and future response vectors based on multi-source operating data, forming offline historical multi-source time series datasets under different operating condition combinations. The multiphysics prediction model module is used to take historical multi-source state sequences and candidate control vectors as inputs, output a predicted response vector and a prediction uncertainty. The predicted response vector includes at least acoustic response index, temperature response index and energy consumption response index. The dynamic risk map construction module is used to calculate the frequency overlap risk, low-difference frequency beat vibration risk, structural resonance excitation risk and low-frequency tone risk based on the compressor rotation frequency, fan blade passing frequency and cabinet structure resonance band, and to establish a dynamic risk map of compressor frequency-evaporator fan speed-condenser fan speed. The multi-objective collaborative optimization module is used to use the prediction results and prediction uncertainty of the multiphysics prediction model as the evaluation basis, the dynamic risk map as the candidate action prior filtering condition, and to construct a multi-objective function that includes at least noise and sound quality objectives, temperature and cooling safety objectives, and energy consumption objectives, and solves the optimal control vector under the constraints of equipment, cooling and speed. The closed-loop control execution module is used to send the optimal control vector to the compressor inverter, evaporator fan driver and condenser fan driver for execution, collect actual operation data to form an online incremental dataset, and update the multiphysics prediction model online based on the online incremental dataset.

[0077] In practical applications, the model and parameters can be recalibrated according to the specific cabinet type, compressor model, number of fan blades and cabinet resonance characteristics. Moreover, this invention can adapt to different structures such as vertical display cabinets, glass door cabinets, and workbench cabinets, as well as actual operating conditions such as dust accumulation in the air duct, loose installation and component performance degradation.

[0078] To verify the effectiveness of the noise reduction control method of this invention, a combined simulation and semi-physical test scenario was constructed for a three-door vertical commercial refrigerated display cabinet. The equipment has an effective volume of approximately 1200L, a target temperature range of 2℃ to 8℃, a compressor with 30Hz to 90Hz variable frequency control, an evaporator fan speed range of 800r / min to 1600r / min, and a condenser fan speed range of 700r / min to 1500r / min. A total of 12 temperature measuring points were arranged in the upper, middle, and lower layers of the cabinet. Vibration measuring points were arranged on the compressor mounting plate, the evaporator fan shroud, and the condenser fan bracket. The main microphone was positioned 1m directly in front of the cabinet, 1.5m above the ground. The experimental environment temperature varied within the range of 25℃ to 32℃, and combined disturbances were introduced, including a 20s door opening, the concentrated placement of materials exceeding the cabinet temperature, and increased air resistance on the condenser side.

[0079] Comparing existing methods with the control method of this invention, existing methods include: a fixed-parameter temperature control method, where the compressor starts and stops according to the temperature hysteresis and the fan operates at a fixed speed; an independent PI coordination method, where the compressor, evaporator fan, and condenser fan are independently adjusted according to temperature or pressure; a compressor single-objective noise reduction MPC method, which mainly optimizes the compressor sound pressure level while the fan operates according to conventional rules; and a step-by-step sequential optimization method, which first satisfies the temperature requirement and then reduces noise within the remaining margin. Each method is repeated 10 times using the same 10 sets of random seeds, initial cabinet temperature, environmental load, and disturbance time.

[0080] The sound pressure level and the highest temperature change inside the cabinet under the conditions of door opening, replenishment heat load, and door reopening are as follows: Figure 7 As shown, the fixed-parameter temperature control method results in a large peak in the A-weighted sound pressure level when the compressor and fan operate at high speed simultaneously after the heat load increases. Furthermore, the speed combination fluctuates continuously as it approaches a high-risk order region. While the independent PI method can shorten the temperature recovery time to some extent, the lack of frequency coordination between the compressor and fan control loops still leads to sound pressure peaks and secondary fluctuations. The method of this invention first increases the temperature weight after a disturbance occurs, synergistically improving cooling capacity within a low-interference-risk speed band. As the cabinet temperature recovers, the noise reduction and energy-saving weights are gradually increased. Therefore, the sound pressure peak is lower, recovery is faster, and the highest temperature inside the cabinet returns to a stable range earlier.

[0081] The 1 / 3 octave band sound pressure level of the fixed parameter temperature control and the method of the present invention under the same cooling load is as follows: Figure 8As shown, a fixed-parameter temperature control exhibits a low-order peak in the compressor frequency around 50Hz, a peak in the evaporator fan blade passing frequency around 160Hz, and a peak in cabinet structure resonance amplification around 250Hz. The method of this invention adjusts the combination of compressor and fan speeds to avoid the main frequencies from approaching each other and the cabinet resonance band, while maintaining airflow and cooling capacity while reducing the amplitude of the corresponding frequency bands. The 50Hz, 160Hz, and 250Hz frequency bands are reduced by approximately 7.4dB, 8.1dB, and 7.9dB, respectively, indicating that this method not only reduces broadband sound pressure levels but also actively avoids the low-frequency hum and narrow-band tone common in commercial refrigerated cabinets.

[0082] Figure 9 To meet the requirement that the highest temperature inside the cabinet does not exceed 8°C and the temperature standard deviation does not exceed 1.30°C, the samples from traditional or stepwise optimization strategies are mainly distributed in areas with high noise and high energy consumption. While compressor noise reduction can lower some sound pressure, poor matching between the fan and compressor may increase condensing pressure or temperature recovery time. This invention, through predictive models, interference risk maps, and Pareto selection, shifts the candidate points towards a lower noise and lower energy consumption region, forming a better Pareto front. This demonstrates that, under the condition of satisfying temperature constraints, multiple compromise solutions can still be obtained for scenario-specific selection.

[0083] The combined results of the five methods under the same equipment, environmental conditions and disturbance sequences are shown in Table 1. The data in the table are the mean ± standard deviation of 10 simulations. The equivalent A-weighted sound pressure level, characteristic tone prominence, cabinet temperature standard deviation, daily comprehensive energy consumption and temperature recovery time are all better the smaller they are, and the temperature qualification rate is better the larger they are.

[0084] Table 1. Comparison of the overall performance of different control methods for commercial refrigeration equipment

[0085] As shown in Table 1, the equivalent A-weighted sound pressure level of the method of the present invention is 44.7 dB(A), which is 7.1 dB(A) lower than that of the fixed parameter temperature control method and 3.1 dB(A) lower than that of the step-by-step sequential optimization method; the peak sound pressure level is reduced to 49.6 dB(A), and the characteristic tone prominence is reduced from 6.2 dB to 2.0 dB, indicating that the active avoidance of low-frequency hum and narrow-band tone through order interference and cabinet resonance can be significantly improved.

[0086] The standard deviation of the cabinet temperature using the method of this invention is 0.72℃, which is about 42.9% lower than that of the fixed parameter temperature control method; the maximum temperature deviation is 1.28℃, and the temperature qualification rate reaches 98.4%, indicating that the collaborative noise reduction does not come at the expense of refrigeration safety. The temperature recovery time after door opening and restocking is 13.1 minutes, which is about 29.6% shorter than that of the fixed parameter temperature control method, and also better than the compressor single-objective noise reduction MPC method.

[0087] Regarding energy consumption, the daily comprehensive energy consumption of the method of this invention is 9.18 kWh, which is about 11.9% lower than that of the fixed parameter temperature control method. This indicates that by optimizing the matching between the compressor and the evaporator / condenser fan, high-pressure ratio operation of the compressor due to insufficient airflow can be avoided, as well as ineffective high airflow. The above results show that, under the cabinet type, environment, and disturbance conditions set in this simulation, this invention can synergistically improve noise, sound quality, temperature uniformity, recovery speed, and comprehensive energy consumption within temperature and refrigeration safety constraints.

[0088] In general, various exemplary embodiments of the present invention can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of the present invention are illustrated or described as block diagrams, flowcharts, or represented using certain other images, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or certain combinations thereof.

[0089] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0090] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A multi-objective coordinated noise reduction control method for compressors and fans in refrigeration equipment, characterized in that, Includes the following steps: Step S1: Collect multi-source operating data of the refrigeration equipment, generate state vectors and actual control vectors based on the multi-source operating data, and form an offline historical multi-source time-series dataset under different operating condition combinations; the multi-source operating data includes at least the cabinet zone temperature, refrigeration pressure, compressor and fan operating parameters, acoustic and vibration data; Step S2: Train a multiphysics prediction model based on the offline historical multi-source time series dataset. The multiphysics prediction model takes the historical multi-source state sequence and candidate control vector as input, obtains the prediction response vector and outputs the prediction uncertainty. The prediction response vector includes at least acoustic response index, temperature response index and energy consumption response index. Step S3: Based on the compressor rotation frequency, fan blade passing frequency and cabinet structure resonance band, calculate the frequency overlap risk, low difference frequency beat vibration risk, structural resonance excitation risk and low frequency tone risk, and establish a dynamic risk map of compressor frequency-evaporator fan speed-condenser fan speed. Step S4: Construct a multi-objective collaborative optimization model, using the prediction results and prediction uncertainty of the multi-physics prediction model as evaluation criteria, and the dynamic risk map as a priori filtering condition for candidate actions. Construct a multi-objective function that includes at least noise and sound quality objectives, temperature and cooling safety objectives, and energy consumption objectives, and solve for the optimal control vector under the constraints of equipment, cooling, and speed. Step S5: Deploy the multiphysics prediction model, dynamic risk map, and multi-objective collaborative optimization model to the edge control platform, periodically output the optimal compressor frequency, optimal evaporator fan speed, and optimal condenser fan speed, and send them down for execution.

2. The multi-target coordinated noise reduction control method for compressors and fans in refrigeration equipment according to claim 1, characterized in that, The multiphysics prediction model in step S2 includes: The thermal-cooling state coding module is used to extract the spatial distribution characteristics of temperature and the thermal dynamic characteristics of the refrigeration system based on the historical multi-source state sequence, and output the thermal-cooling state code. The acoustic vibration order coding module is used to extract sound pressure, sound quality and rotational order features based on the historical multi-source state sequence, and output acoustic vibration state code. A control action encoding module is used to encode the candidate control vector and output the control action code; A cross-modal coupling fusion module is used to fuse the thermal-cooling state code, acoustic-vibration state code, and control action code to output a unified coupling feature; The multi-task prediction module is used to output the prediction response vector and the prediction uncertainty based on the unified coupling feature.

3. The multi-target coordinated noise reduction control method for compressors and fans in refrigeration equipment according to claim 1, characterized in that, The establishment of the offline historical multi-source time series dataset includes: The cabinet internal zone temperature includes data from multiple temperature measuring points arranged at different heights and horizontal positions within the cabinet; the refrigeration pressure includes the compressor suction pipe pressure and the exhaust pipe pressure; the compressor and fan operating parameters include at least the actual operating frequency of the compressor, the actual speed of the evaporator fan, the actual speed of the condenser fan, the compressor current, and the total power of the unit; the acoustic and vibration data include at least the overall sound pressure level, pitch prominence, and component vibration characteristics. Based on the multi-source operation data, state vectors and actual control vectors are established. The state vectors and actual control vectors at L time points are used to form historical time series inputs, and the response vectors of the next H control cycles are used as the supervision targets. By changing the ambient temperature, cabinet load, door opening frequency, compressor frequency, evaporator fan speed, and condenser fan speed, the state vector, actual control vector, and predicted response vector are made to cover multiple operating conditions, forming an offline historical multi-source time series dataset.

4. The multi-target coordinated noise reduction control method for compressors and fans in refrigeration equipment according to claim 2, characterized in that, The loss function of the multiphysics prediction model is: in, Predicting loss for sound pressure level and sound quality. To predict losses based on zoned temperature and cooling safety indicators, For power and energy consumption prediction losses, To predict the uncertainty calibration loss, Loss due to physical consistency constraints; , , , and For the corresponding weights.

5. The multi-target coordinated noise reduction control method for compressors and fans in refrigeration equipment according to claim 1, characterized in that, The establishment of the dynamic risk map includes: Calculate the k-th order rotational frequency based on the compressor rotational speed, calculate the blade passage frequency of the evaporator or condenser based on the number of blades and rotational speed, extract the electromagnetic sideband by combining the number of motor pole pairs, drive switching frequency or measured order spectrum, and calculate the harmonic frequency and speed sideband. The compressor order, fan blade frequency, and side band that meet the preset threshold are used to form the characteristic frequency set of the current candidate speed combination. After the equipment is first installed or repaired, or within a preset period, a restricted frequency sweep is performed to identify the resonant center frequency and bandwidth of the cabinet structure, forming a set of resonant bands and corresponding transfer coefficients. Based on the frequencies in the set of characteristic frequencies, calculate the frequency overlap risk, low-difference frequency beat vibration risk, and structural resonance risk, and calculate the comprehensive interference risk; By traversing the feasible sets of compressor frequency, evaporator fan speed, and condenser fan speed, the comprehensive interference risk of each candidate speed combination is calculated to form the dynamic risk map.

6. The multi-target coordinated noise reduction control method for compressors and fans in refrigeration equipment according to claim 1, characterized in that, The multi-objective function solution in step S4 includes: Based on the allowable range of change of the actual control vector at the previous moment, a local candidate action is generated, and combined with the supplementary candidate velocity in the dynamic risk map to obtain a candidate control vector; The candidate control vector and the current historical multi-source state sequence are input into the multiphysics prediction model to obtain the predicted response vector; The candidate control vectors are filtered according to constraints, prediction uncertainty thresholds and interference risk thresholds, and the remaining candidate control vectors are sorted non-dominated according to a multi-objective function to form a Pareto candidate set. The optimal control vector is obtained by selecting a compromise solution from the Pareto candidate set based on dynamic weights.

7. The multi-objective coordinated noise reduction control method for compressors and fans in refrigeration equipment according to claim 6, characterized in that, The multi-objective function includes: Noise and sound quality targets: ;in, To predict the A-weighted equivalent sound pressure level within the window, For low-frequency sound pressure levels, For pitch prominence, To comprehensively address the risks of intervention, For normalization operators; Temperature and refrigeration safety objectives: ;in, To predict the relative temperature of M temperature measurement points within the prediction window to the target temperature The mean absolute deviation For the temperature standard deviation of the zone, To predict the highest temperature, The upper limit of the allowable temperature, For refrigeration safety risks; Overall energy consumption for the forecast window: ; For the predicted compressor power, For evaporator fan power and This refers to the power of the condenser fan; Motion smoothness target: ; Here, W is the normalized control vector, and W is the positive semi-definite weight matrix. A comprehensive evaluation function is established based on each objective: , , , , , For dynamic weights.

8. The multi-objective coordinated noise reduction control method for compressors and fans in refrigeration equipment according to claim 1, characterized in that, The method further includes: Obtain the response vectors after executing the optimal compressor frequency, optimal evaporator fan speed, and optimal condenser fan speed, and write the multi-source state sequence, actual control vector, and actual response vector into the online incremental dataset; trigger the feature fine-tuner to update according to a preset period or number of samples.

9. The multi-objective coordinated noise reduction control method for compressors and fans in refrigeration equipment according to claim 1, characterized in that, Step S5 also includes scenario-based control: The operating mode is identified based on the current door status, ambient temperature, highest temperature inside the cabinet, and running time. The operating modes include regular business mode, door opening mode, replenishment recovery mode, nighttime stable mode, defrosting mode, and fault protection mode. In the door opening mode and replenishment recovery mode, the temperature weight is increased and the compressor and fan are allowed to accelerate rapidly within the low-risk speed range, while limiting the sound pressure peak and speed mutation. In the aforementioned standard operating mode, select a speed combination with low volume, low beat vibration, and sufficient temperature margin; In the nighttime stable mode, low-frequency sound pressure and overall energy consumption are reduced; In the defrosting mode, regular temperature control optimization is paused and the system operates according to the defrosting logic. After defrosting is completed, the set of candidate control actions is restored.

10. A multi-target collaborative noise reduction control system for compressors and fans in refrigeration equipment, characterized in that, For implementing the multi-objective cooperative noise reduction control method as described in claim 1, the system comprises: The data acquisition module is used to collect multi-source operating data during the operation of the refrigeration equipment. The multi-source operating data includes at least the temperature of each compartment inside the cabinet, the refrigeration pressure, the operating parameters of the compressor and the fan, and acoustic and vibration data. The offline dataset construction module is used to generate state vectors, actual control vectors, and future response vectors based on the multi-source operating data, forming offline historical multi-source time-series datasets under different operating condition combinations. The multiphysics prediction model module is used to take historical multi-source state sequences and candidate control vectors as inputs, predict the response vectors for multiple future control cycles and output the prediction uncertainty. The response vectors include at least acoustic response indicators, temperature response indicators and energy consumption response indicators. The dynamic risk map construction module is used to calculate the frequency overlap risk, low-difference frequency beat vibration risk, structural resonance excitation risk and low-frequency tone risk based on the compressor rotation frequency, fan blade passing frequency and cabinet structure resonance band, and to establish a dynamic risk map of compressor frequency-evaporator fan speed-condenser fan speed. The multi-objective collaborative optimization module is used to take the prediction results and prediction uncertainty of the multi-physics prediction model as the evaluation basis, take the dynamic risk map as the candidate action prior filtering condition, construct a multi-objective function that includes at least noise and sound quality objectives, temperature and cooling safety objectives and energy consumption objectives, and solve the optimal control vector under the constraints of equipment, cooling and speed. The closed-loop control execution module is used to send the optimal control vector to the compressor inverter, evaporator fan driver and condenser fan driver for execution, collect actual operation data to form an online incremental dataset, and update the multiphysics prediction model online based on the online incremental dataset.