Variable frequency compressor energy-saving refrigeration control system and method

Abstract

This application discloses an energy-saving refrigeration control system and method for a variable frequency compressor, relating to the field of refrigeration control. It addresses the problem of unobservable oil accumulation in the evaporator at deep low frequencies, which is confused with the actual heat load yet requires opposite frequency adjustments, making differentiation difficult. The system acquires measurable signals from the acquisition loop to construct a mechanism-constrained oil accumulation state observer. The difference between the precipitation accumulation rate and the oil return removal rate is used to inversely determine the evaporator oil accumulation state. The system decouples the dual-channel deviation of the equivalent heat transfer capacity relative to the absence of an oil accumulation reference, separating the actual heat load component from the heat transfer attenuation component caused by oil accumulation. Based on this, the correspondence between frequency and cooling capacity is corrected online, and the target frequency is calculated only based on the actual heat load to determine the cooling capacity. Oil is returned as needed based on the estimated oil accumulation. This application separates the two causes without an oil accumulation sensor, breaking the closed loop where low frequencies exacerbate oil accumulation, and attenuation is misinterpreted as a basis for frequency reduction, thus balancing oil return, temperature control, and energy saving.

Description

Technical Field

[0001] This application relates to the field of frequency conversion control technology for refrigeration equipment, specifically to an energy-saving refrigeration control system and method for a frequency conversion compressor. Background Technology

[0002] Variable frequency compressor refrigeration technology continuously adjusts the compressor's operating frequency, allowing the system's output cooling capacity to track the actual heat load of the controlled space in real time. This significantly reduces energy consumption compared to fixed frequency systems under partial load conditions and has been widely applied in air conditioning, refrigeration, freezing, and various temperature-controlled equipment. In these systems, the lighter the heat load, the lower the operating frequency provided by the control system, and the compressor's long-term operation in the low-frequency range has become the norm for energy-saving operation. With the rise of applications such as off-grid photovoltaic power supply, vaccine cold chain, biological sample preservation, and precision temperature control, the condition of compressors continuously operating stably in a very low-frequency range for extended periods is becoming increasingly common, placing higher demands on the operational quality of variable frequency control in the low-frequency range.

[0003] When a variable frequency refrigeration system operates at low frequencies, a natural law exists within the system that varies with frequency. The compressor displacement decreases as the operating frequency decreases, and the refrigerant flow velocity within the evaporator heat exchange tubes and suction pipe decreases accordingly. Meanwhile, the refrigeration lubricating oil returns to the compressor via circulation with the refrigerant, and the refrigerant's ability to carry lubricating oil is strongly dependent on the flow rate. When the refrigerant flow rate is lower than the critical flow rate required to carry lubricating oil, some lubricating oil will precipitate from the refrigerant, adhere to the inner wall of the evaporator heat exchange tubes, and remain there. The lower the evaporation temperature, the lower the refrigerant's solubility in the oil and the higher the viscosity of the lubricating oil, resulting in more severe precipitation and retention. The oil film remaining in the evaporator covers the heat exchange surface, forming additional oil film thermal resistance, reducing the heat transfer coefficient at that point, and causing the actual heat exchange capacity of the evaporator to gradually decrease over operating time.

[0004] To address the difficulty of lubricating oil return under low-frequency operation, existing technologies generally rely on conservative worst-case scenarios. These include setting a minimum operating frequency for the compressor, forcibly increasing the operating frequency at fixed intervals to flush out oil from the pipeline, or installing hardware such as oil bends, gas-liquid separators, or oil return bypasses to assist oil return. However, these methods do not reveal the true oil status inside the evaporator and cannot determine the exact amount of lubricating oil remaining at any given moment. Consequently, they can only trigger oil return at fixed intervals or thresholds. The consequence is that even with low oil levels, the frequency may be forced to increase, sacrificing energy efficiency and temperature stability; or, even with high oil levels, failure to remove the excess oil leads to a continuous decline in heat exchange capacity. The lack of a match between the oil return action and the actual oil level makes it difficult to implement the return process on demand.

[0005] Meanwhile, existing variable frequency control technologies, whether threshold-based control that determines shutdown, frequency reduction, or frequency limiting based on thresholds such as room temperature, coil temperature, and pressure, demand-based control that calculates the target frequency from cooling demand, or intelligent control that seeks optimal energy efficiency, all implicitly assume that the relationship between compressor operating frequency and system cooling capacity is relatively fixed and can be pre-calibrated, and that the evaporator's heat exchange capacity is considered a known characteristic that does not drift over time. However, under conditions of long-term oil accumulation during low-frequency operation, this relationship will subtly deviate from the calibrated value due to the slow decay of heat exchange capacity. The aforementioned control technologies lack the means to detect this drift and still calculate the target frequency according to a fixed relationship. The deviation between the adjustment result and the actual cooling demand increases continuously with operating time, and both energy saving and temperature control effects deteriorate accordingly.

[0006] Existing refrigeration soft measurement and fault diagnosis technologies, while capable of inferring performance parameters such as evaporator temperature and cooling capacity, or identifying several typical discrete faults, can utilize signals like current, vibration, temperature, and pressure. However, their goal is primarily to measure a definable performance parameter or determine whether a certain type of fault has occurred. These technologies can detect a decline in system heat exchange capacity or operating performance, but struggle to further distinguish whether this decline stems from a genuine change in the heat load of the controlled space or from heat exchange attenuation caused by oil accumulation inside the evaporator. In fact, these two causes exhibit highly similar or even identical characteristics in measurable signals such as temperature, pressure, and superheat, making it difficult to separate them through passive observation of steady-state deviations alone. Furthermore, they require diametrically opposed frequency adjustments. Misinterpreting heat exchange attenuation caused by oil accumulation as an increase or decrease in heat load and adjusting the frequency accordingly will only push the system deeper into a low-frequency operating range unfavorable for oil return. Current technologies have not yet proposed an effective means to differentiate between these problems where the causes are the same but the responses are opposite.

[0007] The aforementioned difficulties are significantly amplified in certain specialized application scenarios. Refrigeration equipment, such as off-grid photovoltaic vaccine refrigerators, ultra-low temperature biological sample freezers, and precision constant temperature cabinets, which require low heat leakage, high insulation, and stable temperature control over extended periods, are designed to keep the compressor running continuously at a very low frequency range for extended periods, rather than frequently switching between high and low frequencies like conventional air conditioners, in pursuit of energy conservation and temperature stability. It is precisely under these sustained, deep low-frequency operating conditions that the flow rate threshold required to carry lubricating oil cannot be consistently and persistently met. The oil accumulation in the evaporator transforms from an occasional, transient phenomenon in conventional systems that can be naturally cleared by high-frequency operation into a long-term state of continuous, monotonous growth that is difficult to spontaneously dissipate. In off-grid photovoltaic power supply scenarios, available electricity fluctuates with sunlight conditions, further compressing the window of opportunity for oil removal through high-frequency operation; in ultra-low temperature scenarios, the extremely low evaporation temperature further reduces the refrigerant's solubility for oil, leading to even more severe oil accumulation. In these scenarios, the oil accumulation in the evaporator, which is slow to accumulate and cannot be directly observed by the loop sensors, becomes a hidden factor that continuously changes the relationship between frequency and cooling capacity. This makes the aforementioned drift and its indistinguishable causes the primary obstacle that frequency control must face. Existing variable frequency control, oil return control, and refrigeration soft measurement technologies have all failed to provide a solution at this intersection. Summary of the Invention

[0008] To address the problem of hidden drift in variable frequency compressor refrigeration systems during deep low-frequency continuous operation, where the amount of oil slowly accumulating in the evaporator due to difficulties in low-frequency oil return is not directly observable by sensors in the circuit, yet continuously and non-linearly alters the relationship between the operating frequency and the actual cooling capacity, and the technical problem of existing control systems being unable to distinguish between the highly similar observable phenomena of actual heat load changes and oil-induced heat transfer attenuation in evaporation temperature, suction pressure, and superheat, despite requiring diametrically opposed frequency adjustment directions, this invention provides an energy-saving refrigeration control system and method for variable frequency compressors. The purpose of this invention is to invert the unobservable oil state of the evaporator from measurable signals, thereby separating the actual heat load from the oil-induced heat transfer attenuation, and subsequently correcting the correspondence of frequency control online and implementing oil return as needed. This breaks the self-reinforcing closed loop where low-frequency operation exacerbates oil accumulation, and oil attenuation is misinterpreted as a basis for frequency reduction, while ensuring evaporator oil return and maintaining stable temperature, thus achieving energy-saving refrigeration control for variable frequency compressors.

[0009] In one aspect, the present invention provides an energy-saving refrigeration control method for a variable frequency compressor, characterized in that it includes: Step 1: Collect and preprocess the measurable signals of the variable frequency compressor refrigeration system to obtain the timing sequence of physical quantities reflecting the system's operating status. The measurable signals of the loop include evaporation temperature, suction pressure, suction superheat, compressor operating current, operating frequency, and controlled space temperature. Step 2: Construct an oil storage state observer constrained by mechanism, and express the time change rate of oil storage in the evaporator by the difference between the precipitation accumulation rate and the return oil removal rate. The difference expression constitutes the evolution model of oil storage in the evaporator. The oil storage state that cannot be directly observed in the evaporator is inverted from the time series of the physical quantities and the estimated value of oil storage state is output. Step 3: Decouple the deviation of the equivalent heat transfer capacity of the evaporator relative to its oil-free baseline value through two channels, separating it into the actual heat load component and the heat transfer attenuation component caused by oil storage. Step 4: Based on the oil storage status and the actual heat load component, correct the correspondence between the operating frequency and the cooling capacity online, determine the required cooling capacity only according to the actual heat load component, and calculate the corrected correspondence of the required cooling capacity as the target operating frequency; Step 5: Output the target operating frequency to the frequency converter to drive the compressor.

[0010] Specifically, in step one, the measurable signals of the aforementioned loop are collected at a fixed sampling period and filtered and denoised to eliminate abnormal jump values, thereby obtaining a stable and usable physical quantity timing sequence. In addition to the measurable signals of the aforementioned loop, signals such as condensing temperature, exhaust pressure, exhaust temperature, compressor input power, and target temperature control value can also be collected to more completely characterize the system operating status. Preferably, ambient temperature, door opening signal, or photovoltaic input power are also collected as external disturbance quantities for subsequent load component estimation.

[0011] Furthermore, step one also includes: estimating the refrigerant mass flow rate based on the relationship between the operating frequency, the suction pressure, and the suction superheat, according to the compressor volumetric efficiency; and calculating the apparent refrigerant velocity in the evaporator and suction pipe by combining the geometric parameters of the evaporator heat exchange tubes and the refrigerant density; and estimating the current equivalent heat transfer capacity characterization quantity of the evaporator based on the evaporation temperature, the controlled space temperature, and the refrigerant mass flow rate. The apparent refrigerant velocity, as a key physical quantity driving the evolution of the oil storage state, directly determines the refrigerant's ability to carry lubricating oil; the equivalent heat transfer capacity characterization quantity reflects the actual heat flow transferred by the evaporator under a given temperature difference and flow rate, and is the core observation for subsequent inversion of the oil storage state and decoupled load.

[0012] For step two, an oil storage state observer constrained by a mechanism is constructed and run online. The oil storage state is characterized by the amount of oil stored in the evaporator or the corresponding oil film thermal resistance. Specifically, in step two, the precipitation accumulation rate is an increasing function of the degree of under-velocity (the apparent flow rate of the refrigerant is lower than the carrying critical flow rate) and the oil solubility and viscosity determined by the evaporation temperature. The oil return removal rate is an increasing function of the degree of over-velocity (the apparent flow rate of the refrigerant is higher than the carrying critical flow rate), and the carrying critical flow rate varies with the evaporation temperature and refrigerant properties. The residuals in the oil storage state observer that are difficult to model accurately are fitted using a physically constrained neural network. This physically constrained neural network takes the apparent flow rate of the refrigerant, the evaporation temperature, the suction superheat, and the running time as inputs, and outputs corrections to the precipitation accumulation rate and the oil film thermal resistance, and forces the oil storage amount to be non-negative, the accumulation rate to be monotonic with the degree of under-velocity, and the oil film thermal resistance to be monotonic with the oil storage amount. The above evolutionary model follows the natural law of refrigerant carrying lubricating oil, and expresses the time change rate of oil storage as the difference between the precipitation accumulation rate and the return oil removal rate. The application of physical constraints avoids the neural network from giving estimation results that violate natural laws in the data sparse region.

[0013] Furthermore, step two also includes: superimposing the oil film thermal resistance onto the evaporator heat transfer thermal resistance to obtain the equivalent heat transfer capacity affected by oil storage, and mapping it to the operating cost required to reach the target temperature at the current operating frequency; using the equivalent heat transfer capacity and the operating cost as observations, a state observer is used to recursively estimate the oil storage state, minimizing the deviation between the observation model output and the measured observations, and outputting the oil storage state estimate and its uncertainty. Thus, without a direct oil storage sensor, the unobservable oil storage state of the evaporator is continuously inverted from the measurable signals, and the output uncertainty is used for subsequent frequency correction and oil return trade-off assessment to estimate the reliability.

[0014] For step three, a dual-channel decoupling discrimination is performed on the deviation of the equivalent heat transfer capacity obtained in step one relative to its baseline value without oil storage and without additional attenuation. Specifically, step three includes: extracting the rate of change, monotonicity, reversibility, and correlation with external disturbances of the deviation at multiple time scales; processing the above features using a time-series feature extraction network, outputting the ratio of the actual heat load component to the oil storage-induced heat transfer attenuation component, and using the oil storage state estimate output in step two as a priori constraint for the oil storage-induced heat transfer attenuation component, so that the sum of the two components is consistent with the deviation; wherein the actual heat load component is rapidly reversible in response to frequency adjustment and is correlated with the external disturbance, and the oil storage-induced heat transfer attenuation component slowly and monotonically accumulates with the duration of deep low-frequency operation, has a slow response to frequency changes, and exhibits hysteresis characteristics. The decoupling utilizes the essential differences in dynamic characteristics between actual heat load changes and oil accumulation for attribution: the former changes rapidly and reversibly with changes in external disturbances such as ambient temperature, door opening, and photovoltaic input, and the response frequency adjusts, while the latter exhibits path-dependent characteristics that accumulate over time, thus distinguishing the two causes that appear to be confused in terms of temperature, pressure, and superheat.

[0015] Furthermore, step three also includes: superimposing a small frequency disturbance, whose amplitude and duty cycle are constrained by the stable temperature band of the controlled space, onto the operating frequency as an excitation; extracting the frequency domain symmetry and time domain hysteresis area from the response of the equivalent heat transfer capacity characterization quantity relative to this small frequency disturbance; the actual heat load component exhibits a rapid symmetrical response to positive and negative disturbances, and the oil-induced heat transfer attenuation component exhibits a hysteresis area; this hysteresis area is used as the observation of the oil-induced heat transfer attenuation component and fed back to the oil storage state observer. The amplitude and duty cycle of the small frequency disturbance are set aside without disrupting the stable temperature of the controlled space, thereby further improving the reliability of separating the two causes of the actual heat load component and the oil-induced heat transfer attenuation component without affecting the temperature control quality, and providing the oil storage state observer with an observation that directly reflects the oil-induced attenuation.

[0016] For step four, the correspondence between frequency and cooling capacity is corrected online based on the estimated oil state value from step two and the actual heat load component obtained from step three. Specifically, the nominal frequency-cooling capacity correspondence is corrected using the oil film thermal resistance corresponding to the estimated oil state value, resulting in a real-time correspondence reflecting the current heat exchange attenuation caused by oil storage. This allows the control system to provide a target operating frequency that takes into account the heat exchange attenuation under the same target cooling capacity. The required cooling capacity is determined only based on the actual heat load component obtained from the separation, avoiding incorrect frequency reduction due to the miscalculation of oil storage-induced attenuation into the load. The corrected correspondence of the required cooling capacity is converted into the target operating frequency and output to the inverter driver to drive the compressor. Furthermore, when step three determines that the decrease in heat exchange capacity is mainly due to oil storage-induced attenuation rather than a reduction in actual load, the control system suppresses the tendency to continue reducing the frequency, thereby breaking the self-reinforcing closed loop where low-frequency operation exacerbates oil storage, and oil storage attenuation is misinterpreted as a basis for frequency reduction.

[0017] Furthermore, the method also includes step six, performing a dynamic trade-off between oil return and energy saving based on oil storage perception: The energy-saving target, temperature control deviation target, and the future heat exchange attenuation cost caused by oil accumulation are incorporated into the objective function in the rolling time domain, with the evolution model of the evaporator oil storage volume as the state constraint; the optimal operating frequency sequence for the objective function is solved in the rolling time domain; when the estimated oil storage state value exceeds the dynamic threshold that varies with the evaporation temperature and operating conditions, the oil return frequency is increased, with the output amplitude and duration matching the estimated oil storage volume, so that the apparent refrigerant flow rate exceeds the critical flow rate required for the lubricating oil to be carried back to the compressor by the refrigerant, thus clearing the oil storage; after the oil return is completed, the oil storage state observer confirms that the estimated oil storage state value has fallen, ending the oil return and resuming energy-saving operation. Therefore, the oil return volume and increase magnitude are determined as needed based on the estimated oil storage volume rather than implemented at fixed time intervals, avoiding excessive or insufficient forced oil flushing under conservative presets. Preferably, the dynamic threshold and oil return timing are determined under the constraint of ensuring that the temperature of the controlled space does not exceed the allowable temperature control zone, so as to balance the oil return demand and temperature stability.

[0018] Furthermore, for off-grid photovoltaic (PV) power-powered continuous operation scenarios for vaccine refrigeration or cryogenic biological sample freezing, the available PV power window is incorporated as a constraint or incentive into the solution of the objective function in the rolling time domain, prioritizing the increase in oil return frequency during periods of ample PV power. For cryogenic scenarios, the oil storage state observer corrects the dependence of the critical flow rate and the precipitation accumulation rate on the evaporation temperature. This maintains the required temperature range for vaccines or biological samples while completing on-demand oil return with minimal necessary amplitude and frequency, and adapts the oil storage state estimation and oil return decision to extremely low evaporation temperature conditions where oil solubility significantly decreases and oil accumulation intensifies.

[0019] Furthermore, the method also includes: the oil condition observer performs online adaptive updates based on the deviation between the measured and predicted values ​​during operation to adapt to individual differences in refrigerant charge, evaporator structure, and lubricating oil grade. This allows for continuous correction of the oil condition estimation and frequency control during long-term operation, enabling the method to adapt to individual differences in different models and operating conditions without requiring individual calibration for each piece of equipment.

[0020] Another aspect of the present invention provides an energy-saving refrigeration control system for a variable frequency compressor, characterized in that it comprises: The signal acquisition and preprocessing module is used to acquire and preprocess the loop measurable signals of the variable frequency compressor refrigeration system to obtain the timing of physical quantities reflecting the system's operating status. The loop measurable signals include evaporation temperature, suction pressure, suction superheat, compressor operating current, operating frequency, and controlled space temperature. The oil storage state observation module is used to construct a mechanism-constrained oil storage state observer. The difference between the precipitation accumulation rate and the return oil removal rate is used to express the time change rate of the oil storage in the evaporator. The difference expression constitutes the evolution model of the oil storage in the evaporator. The oil storage state that cannot be directly observed in the evaporator is inverted from the time series of the physical quantities and the estimated value of the oil storage state is output. The load and oil storage decoupling discrimination module is used to decouple the deviation of the evaporator's equivalent heat transfer capacity characterization quantity relative to its oil-free reference value in two channels, separating it into the real heat load component and the oil storage-induced heat transfer attenuation component. The frequency control correction module is used to correct the correspondence between the operating frequency and the cooling capacity online based on the oil storage status and the actual heat load component. It determines the required cooling capacity only according to the actual heat load component, and calculates the corrected correspondence of the required cooling capacity as the target operating frequency and outputs it to the frequency converter to drive the compressor to run. The oil return and energy saving balance module is used to dynamically balance oil return and energy saving based on the oil storage state estimate.

[0021] The system's signal acquisition and preprocessing module, oil storage status observation module, load and oil storage decoupling judgment module, frequency control correction module, and oil return energy-saving trade-off module respectively execute the above steps. The signal acquisition and preprocessing module, oil storage status observation module, and load and oil storage decoupling judgment module execute steps one to three in sequence, the frequency control correction module executes steps four and five, and the oil return energy-saving trade-off module executes step six. The function of each module is consistent with the description of the corresponding method steps, and the corrected target operating frequency is output to the frequency converter to control the compressor operation.

[0022] The beneficial effects of this invention are as follows: By using a mechanism-constrained oil retention state observer to invert the unobservable oil retention state within the evaporator from the measurable signals of the loop, it can answer the question of how much lubricating oil is currently retained without installing an oil retention sensor inside the evaporator. This overcomes the limitations of existing low-frequency oil return technologies, which are all based on worst-case conservative presets and do not know the true oil retention state. Through dual-channel decoupling discrimination, it separates the actual heat load change from the oil retention-induced heat transfer attenuation, two causes that are highly mixed at the observable levels of evaporation temperature, suction pressure, and superheat, but are diametrically opposed in the direction of frequency adjustment. This prevents the control system from misinterpreting oil retention-induced attenuation as an increase in heat load and incorrectly reducing the frequency, thereby breaking the self-reinforcing closed loop where low-frequency operation exacerbates oil retention, and oil retention attenuation is misinterpreted as a basis for frequency reduction. On this basis, it corrects the correspondence between frequency and cooling capacity online and determines the frequency solely based on the actual heat load. The required cooling capacity ensures that the operating frequency tracks the actual load rather than the correspondence caused by hidden drift contamination, guaranteeing the accuracy of energy-saving operation. Furthermore, by incorporating energy-saving targets, temperature control deviation targets, and future heat exchange attenuation costs into rolling time-domain optimization, and implementing oil return frequency increases that match the amplitude and duration based on the estimated oil level, the forced oil flushing at fixed time intervals is replaced. This achieves energy-saving operation while ensuring evaporator oil return and maintaining stable temperature in the controlled space. For special scenarios such as off-grid photovoltaic vaccine refrigeration, ultra-low temperature biological sample freezing, and precision constant temperature operation, which involve long-term deep low-frequency continuous operation, the method and system can still balance oil return, temperature control, and energy saving in scenarios where conventional frequency conversion control is ineffective. This demonstrates significant practical value by incorporating the photovoltaic power window into optimization, correcting the observer based on the extremely low evaporation temperature, and adaptively updating online for individual differences. Attached Figure Description

[0023] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] Figure 1 This is an overall flowchart of the variable frequency compressor energy-saving refrigeration control method of the present invention.

[0025] Figure 2 This is a data flow diagram of the oil storage state observer and the physical constraint neural network of the present invention.

[0026] Figure 3 This is a structural block diagram of the variable frequency compressor energy-saving refrigeration control system of the present invention.

[0027] Figure 4 This is a cross-sectional schematic diagram of the accumulation and removal of lubricating oil film inside the heat exchange tubes of the evaporator.

[0028] Figure 5 This is a comparison chart showing the operating effects of the present invention and undecoupled control under various operating conditions. Detailed Implementation

[0029] To make the objectives and technical solutions of this invention clearer, the embodiments of this invention will be further described in detail below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of this invention and should not be construed as limiting the scope of protection of this invention.

[0030] Example 1 This embodiment provides an energy-saving refrigeration control method for a variable frequency compressor. It uses an off-grid photovoltaic vaccine cold storage box as an example, employing a variable frequency drive-driven rolling rotor compressor with an internal volume of approximately 6.2 cubic centimeters per revolution, using R290 as refrigerant, and featuring a finned tube heat exchanger with an inner diameter D of approximately 7 millimeters. The device operates continuously in a deep low-frequency range (approximately 18 to 30 Hz), with a target temperature control value of 5 degrees Celsius and an allowable temperature control range of ±3 degrees Celsius. The evaporation temperature remains consistently between -25 and -30 degrees Celsius. The overall process of this method is as follows: Figure 1 As shown, steps one through six are executed sequentially, with each step completed serially within the same control cycle, which is set to 1 second.

[0031] Step 1: Collect and preprocess the measurable signals of the variable frequency compressor refrigeration system during operation to obtain the time sequence of physical quantities reflecting the system's operating status.

[0032] In sub-step 1, the loop measurable signals are acquired at a fixed sampling period and then filtered and denoised. The loop measurable signals include evaporation temperature, condensation temperature, suction pressure, discharge pressure, suction superheat, discharge temperature, compressor operating current or input power, controlled space temperature, target temperature control value, and the compressor's current operating frequency. In this embodiment, the sampling period for slow variables such as evaporation temperature and controlled space temperature is 1 second, while the sampling period for fast variables such as suction pressure and compressor operating current is 100 milliseconds, i.e., a sampling frequency of 10 Hz. For each signal, a sliding median filter with a length of 5 points is first applied to remove pulse-type abnormal jump values. The judgment criterion is that if the deviation of a single point from the median of its immediate and subsequent neighborhoods exceeds 8% of the signal range, it is considered abnormal and replaced with the median. Then, a first-order low-pass filter with a cutoff frequency of approximately 0.05 Hz is applied to suppress measurement noise, resulting in a stable and usable physical quantity timing sequence. Preferably, ambient temperature, door opening signal, or photovoltaic input power are also collected as external disturbance quantities for subsequent load component estimation; the door opening signal is a switching quantity, and the photovoltaic input power sampling period is 1 second.

[0033] In sub-step 2, the refrigerant mass flow rate is estimated based on the compressor's current operating frequency, suction pressure, and suction superheat according to the compressor's volumetric efficiency relationship. The apparent flow rate of the refrigerant in the evaporator and suction pipe is then calculated by combining the evaporator heat exchanger tube geometry parameters with the refrigerant density. The refrigerant mass flow rate is estimated based on the operating frequency, the compressor's theoretical displacement per revolution, volumetric efficiency, and suction density according to the volumetric efficiency relationship. The volumetric efficiency is a function of the operating frequency and compression ratio; in this embodiment, it is taken as 0.72 to 0.80 in the 18 to 30 Hz range, increasing slightly with increasing frequency. The suction density is obtained from the suction pressure and suction superheat based on the physical properties of refrigerant R290, and is approximately 2.1 kg / m³ under the above operating conditions. Taking an operating frequency of 20 Hz, a displacement per revolution of 6.2 cubic centimeters, and a volumetric efficiency of 0.74, the estimated refrigerant mass flow rate is approximately 0.19 g / s. The refrigerant mass flow rate is then converted into the in-pipe velocity based on the flow cross-section, yielding the apparent refrigerant velocity. Taking a suction pipe inner diameter of 7 mm and a flow cross-sectional area of ​​approximately 38.5 square millimeters, the calculated apparent velocity in the suction pipe is approximately 2.4 m / s. In the evaporator heat exchange tubes, because the flow cross-section is approximately 1.4 times larger than that of the suction pipe, the apparent velocity drops to approximately 1.7 m / s. This apparent velocity, as a key physical quantity driving the evolution of the oil storage state, directly determines the refrigerant's ability to carry lubricating oil.

[0034] In sub-step 3, the current equivalent heat transfer capacity of the evaporator is estimated using the evaporation temperature, the controlled space temperature, and the refrigerant mass flow rate. This capacity is taken as the ratio of the actual heat flow transferred by the evaporator to its logarithmic mean temperature difference, i.e., the product of the equivalent heat transfer coefficient and the heat transfer area (denoted as the UA value), reflecting the actual heat flow transferred by the evaporator under a given temperature difference and flow rate. In this embodiment, the UA value is approximately 12 W / K in the oil-free baseline state, decreasing as oil accumulates. It is a core observation for subsequent inversion of the oil-storage state and decoupled load.

[0035] Step two involves constructing and running an online mechanism-constrained oil storage state observer to invert the unobservable oil storage state within the evaporator from the time series of physical quantities obtained in Step one. The oil storage state is characterized by the amount of oil stored in the evaporator or the corresponding oil film thermal resistance. The data flow between the oil storage state observer and the physical constraint neural network is as follows: Figure 2 As shown.

[0036] In sub-step 1, an evolution model of the oil storage state is established. This model follows the natural law of refrigerant carrying lubricating oil, expressing the time change rate of oil storage as the difference between the precipitation accumulation rate and the return oil removal rate. The precipitation accumulation rate is taken as the degree of under-velocity (the apparent flow rate of the refrigerant is lower than the critical carrying flow rate) and the increasing function of oil solubility and viscosity determined by the evaporation temperature. In this embodiment, the precipitation accumulation rate is taken as... Where v is the apparent flow rate of the refrigerant, vcrit To carry the critical flow rate, φ is the evaporation temperature correction factor, k acc The precipitation coefficient is given; the oil return removal rate is taken as a linearly increasing function of the degree of overspeed of the refrigerant apparent velocity exceeding the critical carrying velocity, with a proportionality constant of k. rem Coefficient k acc Take 4.0 × 10 -3 k rem Take 1.8 × 10 -2 (Units are grams per second (meters per second) to the corresponding power). The evaporation temperature correction factor φ reflects that the lower the evaporation temperature, the lower the oil solubility, the higher the viscosity, and the more intense the precipitation. It increases exponentially with the decrease of the evaporation temperature relative to the reference temperature, which is taken as -10 degrees Celsius. At an evaporation temperature of -30 degrees Celsius, φ is approximately 5.3, which significantly amplifies the precipitation.

[0037] The critical carryover velocity itself varies with the evaporation temperature and refrigerant properties. This embodiment uses an expression based on the gas-liquid two-phase carryover criterion. Where g is the acceleration due to gravity, D is the inner diameter of the heat exchange tube, and ρ l With ρ g These are the liquid and gaseous densities of the refrigerant at the evaporation temperature, respectively, C. crit Let C be a dimensionless coefficient. crit Given an XLR of 1.25, a gravitational acceleration of 9.81 m / s², a heat exchange tube inner diameter of 7 mm, and liquid phase density of approximately 555 kg / m³ and gas phase density of approximately 4.8 kg / m³ at an evaporation temperature of -30°C, the calculated critical carry-over velocity is approximately 3.5 m / s. It is evident that at 20 Hz and an apparent velocity of 1.7 m / s, the apparent velocity is significantly lower than the critical carry-over velocity of 3.5 m / s, with a velocity deficit of approximately 1.8 m / s. The precipitation accumulation rate dominates, resulting in continuous oil accumulation in the evaporator. The accumulation and removal morphology of the lubricating oil film within the evaporator heat exchange tubes is as follows: Figure 4 As shown: During deep low-frequency continuous operation, the apparent flow rate is lower than the critical flow rate. The precipitated lubricating oil adheres to the inner wall of the heat exchange tube to form an oil film. The oil film thermal resistance increases and the equivalent heat exchange capacity decreases. When the oil storage status observer determines that on-demand oil return is triggered and the apparent flow rate is increased to a level higher than the critical flow rate, the stagnant oil is re-carried and removed by the refrigerant, and the heat exchange capacity is restored.

[0038] In sub-step 2, an observation model is established to represent the oil storage state as an observable quantity. The oil film thermal resistance is superimposed on the evaporator heat transfer thermal resistance to obtain the equivalent heat transfer capacity affected by oil storage, which is then mapped to the operating cost required to reach the target temperature at the current frequency. The oil film thermal resistance is taken as a monotonic function of the oil storage amount. In this embodiment, the oil film thermal resistance is taken to be directly proportional to the oil storage amount and inversely proportional to the thermal conductivity of the lubricating oil and the wetting area. The thermal conductivity of the lubricating oil is approximately 0.13 W / m Kelvin. When the oil storage amount accumulates to approximately 3 grams, the oil film thermal resistance formed covering the heat transfer surface is approximately 0.008 Kelvin / W. The equivalent heat transfer capacity affected by oil storage is obtained by superimposing the oil film thermal resistance in series with the heat transfer thermal resistance without oil storage. Based on a baseline UA value of 12 watts per Kelvin and an oil film thermal resistance of 0.008 Kelvin per watt, the UA value drops to approximately 10.9 watts per Kelvin due to oil retention, resulting in a decrease in heat exchange capacity of approximately 9%. This translates to an increase in the operating frequency required to reach the target temperature or an increase in operating costs.

[0039] In sub-step 3, using the equivalent heat transfer capacity characterization quantity obtained in step 1 and the required operating cost as observations, a state observer is used to recursively estimate the oil storage state, minimizing the deviation between the observation model output and the measured observations. Due to the nonlinear dependence of the evolution model on the apparent flow rate, this embodiment uses an extended Kalman observer for recursion, with the oil storage quantity as the state quantity. In the prediction step, the oil storage quantity estimate from the previous moment is advanced by one control cycle along the difference between the precipitation accumulation rate and the return oil removal rate, according to the evolution model, to obtain the predicted oil storage quantity. Simultaneously, the estimation error covariance is propagated forward according to the Jacobian matrix of the evolution model and the process noise covariance. In the update step, the observation model first calculates the predicted observation value, then calculates the observation residual, i.e., the difference between the measured UA value and the predicted UA value, and corrects the predicted oil storage quantity according to the Kalman gain to obtain the estimated oil storage quantity. The Kalman gain is calculated from the prediction error covariance, the observation Jacobian, and the observation noise covariance in the standard extended Kalman form. In this embodiment, the standard deviation of process noise is taken as 0.02 g and the standard deviation of observation noise is taken as 0.3 watts per Kelvin. Thus, without a direct oil storage sensor, the estimated value of the oil storage state of the evaporator and its uncertainty (i.e., the estimation error covariance) are continuously output. The output uncertainty is used for subsequent frequency correction and oil return trade-off assessment to evaluate the reliability of the estimate.

[0040] Furthermore, the parts remaining in the evolutionary and observational models due to the difficulty in accurately modeling the refrigerant and lubricating oil properties are fitted using a physically constrained neural network. During the model inference phase, this neural network performs a forward calculation online once per control cycle: its input is an input vector composed of four physical quantities with definite technical meanings—the refrigerant apparent flow rate, evaporation temperature, suction superheat, and running time—passing through two fully connected hidden layers of 32 neurons each. The output is a correction for the precipitation accumulation rate and a correction for the oil film thermal resistance. To enforce the physical constraints, the output layer applies a softplus activation to the precipitation accumulation rate correction to ensure it is always non-negative, thus guaranteeing a non-negative oil content. The connection weights from the inputs to the outputs related to the underspeed and oil content in the neural network are constrained to be non-negative through an exponential transformation, ensuring that the precipitation accumulation rate is monotonic with the underspeed and the oil film thermal resistance is monotonic with the oil content. These corrections are respectively superimposed on the precipitation accumulation rate in sub-step 1 and the oil film thermal resistance in sub-step 2, participating in the observer recursion. The application of physical constraints prevents neural networks from giving estimation results that violate natural laws in sparse data regions. For example, in sparse sample regions with extremely low evaporation temperatures and long operating times, the amount of oil stored can still be guaranteed to accumulate monotonically.

[0041] During the model training phase, the training data for the physically constrained neural network comes from bench experiments and high-fidelity simulations with oil storage reference calibration. On the bench, true oil storage value sequences at different frequencies, evaporation temperatures, and operating times are obtained using weighing or oil circulation rate measurement methods. These sequences, along with synchronously recorded apparent flow rate, evaporation temperature, and superheat, form a sample. During preprocessing, each input physical quantity is standardized to zero mean and unit variance according to its range, and then segmented into 256-point sequence segments based on operating time. The training objective is to ensure that the predicted UA value output by the observation model is close to the measured UA value, and that the estimated oil storage value is close to the true oil storage value. A penalty term for violating non-negativity and monotonicity constraints is added as a soft constraint to assist in the convergence of hard constraints. The Adam optimizer is used for training, with a batch size of 64 and an initial learning rate of 2.3 × 10⁻⁶. -4 The training was performed using a segmented strategy, decreasing by 0.6 times every 40 rounds, with a maximum of 300 training rounds. The training was terminated early when the mean square error of the UA value prediction on the validation set stopped decreasing for 20 consecutive rounds. The evaluation metrics were the mean absolute error of the oil inventory estimation and the root mean square error of the UA value prediction. In this embodiment, the mean absolute error of the oil inventory estimation after training was approximately 0.25 grams.

[0042] Step 3: Perform dual-channel decoupling discrimination on the deviation of the equivalent heat transfer capacity obtained in Step 1 from its baseline value without oil storage and without additional attenuation, and separate the deviation into the actual heat load component and the heat transfer attenuation component caused by oil storage.

[0043] In sub-step 1, the dynamic characteristics of the deviation at multiple time scales are extracted, including its rate of change, monotonicity, reversibility, and correlation with external disturbances. In this embodiment, the moving average and slope of the deviation are calculated at three time windows: 1 minute, 10 minutes, and 1 hour, and the moving cross-correlation coefficients between the deviation and external disturbances such as ambient temperature, door opening signal, and photovoltaic input power are calculated.

[0044] In sub-step 2, the fundamental differences in dynamic characteristics between the actual heat load change and oil accumulation are used for attribution. The actual heat load component is characterized by rapid and reversible response to frequency adjustment and correlation with external disturbances such as ambient temperature, door opening, and photovoltaic input. Its characteristics include a high cross-correlation coefficient with external disturbances (greater than 0.6 in this embodiment), a rapid decline in deviation after frequency adjustment, and a reversible path. The oil accumulation-induced heat transfer attenuation component is characterized by slow and monotonically accumulating with depth and low-frequency operation duration, a slow response to frequency changes, and path-dependent hysteresis characteristics. Its characteristics include a consistently positive window slope over a long period, a low cross-correlation coefficient with external disturbances, a slow decline in deviation after frequency adjustment, and non-overlapping rise and fall paths.

[0045] In sub-step 3, a temporal feature extraction network is used to process the aforementioned dynamic features, outputting the ratio of the true heat load component to the oil-induced heat transfer attenuation component in the deviation. The oil storage state estimate output in step 2 is used as a priori constraint for the oil-induced heat transfer attenuation component, ensuring that the sum of the two components matches the measured deviation. During the model inference phase, the temporal feature extraction network takes the cross-correlation coefficient between the aforementioned multi-scale dynamic features and external disturbances as input, passes through a gated recurrent unit and a fully connected layer, and outputs the ratio of the true heat load component to the deviation (values ​​from 0 to 1, normalized by sigmoid). This ratio is multiplied by the measured deviation to obtain the true heat load component; the remainder is the oil-induced heat transfer attenuation component. To ensure that the sum of the two components is consistently consistent with the measured deviation, the network only outputs the distribution ratio, not their absolute values. The heat transfer attenuation corresponding to the oil film thermal resistance output in step 2 is used as the prior center for the oil-induced heat transfer attenuation component, and a penalty for deviation between the oil-induced heat transfer attenuation component and this prior is added to the loss. The training of the time-series feature extraction network is performed jointly with or independently of the neural network in step two. The training data consists of operational sequences labeled with load change events and oil accumulation events. The Adam optimizer is used, with an initial learning rate of 1.7 × 10⁻⁶. -4 Batch size 32.

[0046] Preferably, in sub-step 3, a small frequency disturbance with limited amplitude is actively superimposed onto the operating frequency as an excitation, and the response of the equivalent heat transfer capacity to this disturbance is observed. In this embodiment, a small square wave-shaped frequency disturbance with an amplitude of approximately ±0.5 Hz and a period of approximately 120 seconds is superimposed. Its amplitude and duty cycle are limited by temperature stability constraints, under the premise of not disrupting the temperature stability of the controlled space (temperature fluctuation does not exceed ±0.2 degrees Celsius). The actual heat load component responds quickly and symmetrically to the disturbance, that is, the UA value response is symmetrical about the direction of the disturbance when the frequency is increased by 0.5 Hz and decreased by 0.5 Hz, and the rise and fall paths are basically the same. The component of heat transfer attenuation due to oil storage exhibits a slow hysteresis response, and its UA value forms a non-closed loop as the frequency increases and decreases. The hysteresis area enclosed by this loop on the frequency-UA value plane is extracted, and the hysteresis area is used as the observation feedback of the heat transfer attenuation component due to oil storage to the oil storage state observer in step 2. The larger the hysteresis area, the more significant the attenuation due to oil storage. In this embodiment, the oil content is approximately 3 grams, and the hysteresis area is approximately 0.4 Hz multiwatts per Kelvin. This further improves the reliability of the separation of the two causes and provides an observational measure that directly reflects the oil storage-induced decay for the oil storage state observer.

[0047] Step four: Based on the estimated oil storage state value from step two and the actual heat load component obtained from step three, correct the correspondence between the operating frequency and the cooling capacity online and calculate the target operating frequency.

[0048] In sub-step 1, the nominal frequency-cooling capacity correspondence is corrected using the oil film thermal resistance corresponding to the estimated oil storage state, resulting in a real-time correspondence reflecting the heat transfer attenuation caused by the current oil storage. The nominal correspondence is the monotonic relationship of cooling capacity with frequency when there is no oil storage (in this embodiment, 20 Hz corresponds to approximately 68 watts of cooling capacity, and 30 Hz corresponds to approximately 105 watts). During correction, the ratio of the UA value affected by oil storage to the baseline UA value is used as the heat transfer attenuation factor. The nominal cooling capacity is reduced by this attenuation factor to obtain the real-time cooling capacity. When the oil film thermal resistance is 0.008 Kelvin per watt and the attenuation factor is approximately 0.91, the actual cooling capacity at 20 Hz decreases from 68 watts to approximately 62 watts, causing the control system to output a target frequency that takes into account the heat transfer attenuation at the same target cooling capacity. That is, to achieve the same cooling capacity, the frequency needs to be adjusted up by approximately 1 to 2 Hz.

[0049] In sub-step 2, the required cooling capacity is determined solely based on the actual heat load component obtained from step 3, avoiding the erroneous frequency reduction due to oil-induced heat transfer attenuation being mistakenly included in the load calculation. In other words, the required cooling capacity is given only by the load demand corresponding to the actual heat load component; the oil-induced heat transfer attenuation component is not included in the calculation of the required cooling capacity.

[0050] In sub-step 3, the corrected correspondence of the required cooling capacity is converted into the target operating frequency. Specifically, when step 3 determines that the decrease in heat exchange capacity is mainly due to oil accumulation-induced attenuation rather than a decrease in actual load, the control system suppresses the tendency to continue reducing the frequency: In this embodiment, when the proportion of oil accumulation-induced heat exchange attenuation to the deviation exceeds 0.6, further frequency reduction is prohibited on the current basis, and only maintaining or increasing the frequency is allowed, thereby breaking the self-reinforcing closed loop where low-frequency operation exacerbates oil accumulation, and oil accumulation attenuation is misinterpreted as the basis for frequency reduction. If this distinction is not made, the control system will misjudge the decrease in UA value as a decrease in load and reduce the frequency to below 18 Hz, causing the apparent flow rate to further fall below the carrying critical flow rate and oil accumulation to accelerate.

[0051] Step five: The target operating frequency calculated in step four is output to the variable frequency drive to drive the compressor. In this embodiment, the controller sends the target operating frequency to the power conversion circuit of the variable frequency drive via a communication interface, which adjusts the actual operating frequency of the compressor.

[0052] Step six involves dynamically balancing oil return and energy saving based on oil storage perception, and adaptively deciding whether and how to implement oil return based on the estimated oil storage status.

[0053] In sub-step 1, the energy-saving target, the temperature control deviation target, and the future heat exchange attenuation cost caused by oil accumulation are all incorporated into a rolling time-domain objective function, and the oil storage state evolution model from step 2 is used as the state constraint for this time-domain optimization. In this embodiment, the rolling time domain length is 30 minutes, the offline step length is 1 minute, for a total of 30 steps, and the objective function is the sum of the weighted terms of energy consumption, the squared weighted terms of temperature control deviation, and the weighted terms of the future attenuation cost caused by the final oil storage amount. The weights of the three terms in this embodiment are 1.0, 8.0, and 0.5, respectively. The oil storage evolution model is embedded in the optimization as an equality state constraint, and the controlled space temperature must be within a temperature control zone of ±3 degrees Celsius as an inequality constraint.

[0054] In sub-step 2, the optimal operating frequency sequence for the objective function is solved within the rolling time domain. When the estimated oil storage state exceeds a dynamic threshold that varies with evaporation temperature and operating conditions, an increase in the oil return frequency with amplitude and duration matching the estimated oil storage is output. In this embodiment, the dynamic threshold is taken as a function that decreases exponentially with the decrease in evaporation temperature relative to the reference temperature. The baseline threshold is 3.5 grams, the reference temperature is -10 degrees Celsius, and the threshold is approximately 2.3 grams when the evaporation temperature is -30 degrees Celsius. The lower the evaporation temperature, the earlier the oil return is triggered. When the estimated oil level exceeds the dynamic threshold, the calculated increase in oil return frequency and duration are matched with the estimated oil level. In this embodiment, when the oil level is approximately 3 grams, the frequency is increased from 20 Hz to approximately 45 Hz and maintained for approximately 90 seconds. This increases the apparent flow velocity within the evaporator heat exchange tubes from 1.7 m / s to approximately 4.2 m / s, exceeding the critical carry-over velocity of 3.5 m / s by approximately 0.7 m / s. At this point, the oil return and removal rate becomes dominant, and the oil is re-carried back to the compressor by the refrigerant. The oil return amount and increase are determined as needed based on the estimated oil level, rather than implemented at fixed time intervals. The larger the oil level, the higher the increase and the longer the duration.

[0055] In sub-step 3, after the oil return is completed, the observer confirms that the estimated oil level has dropped, thereby ending the oil return process and resuming energy-saving operation. In this embodiment, when the estimated oil level drops to below approximately 0.5 grams (i.e., to about 20% of the threshold), the oil return is deemed sufficient, the oil return frequency increase is stopped, and the frequency is returned to the target frequency calculated in step 4, resuming energy-saving operation. Preferably, the dynamic threshold and the oil return timing are determined under the constraint of ensuring that the temperature of the controlled space does not exceed the allowable temperature control zone. That is, the temperature drop in the controlled space caused by the short-term high-frequency operation during the oil return period must always be maintained within 5 degrees Celsius plus or minus 3 degrees Celsius of the allowable temperature control zone to balance the oil return requirements and temperature stability.

[0056] Furthermore, for special application scenarios involving long-term, deep, low-frequency continuous operation, this embodiment makes the following adaptations based on the oil return and energy-saving trade-off in step six above.

[0057] Firstly, for off-grid photovoltaic vaccine refrigeration, ultra-low temperature biological sample freezing, or precise constant temperature continuous operation scenarios, the long-term monotonic accumulation characteristics of the oil storage state evolution model are incorporated into the rolling time domain optimization. This ensures that the oil return frequency is increased with the minimum necessary magnitude and frequency while meeting the temperature control zone constraints. Because these scenarios involve prolonged periods in the low-frequency range and monotonic oil accumulation, this embodiment incorporates the heat transfer attenuation cost of the next few hours into the terminal term of the objective function. This makes the optimization tend to return oil at a smaller magnitude earlier, before the oil storage becomes severe, thus avoiding the need for a large-scale, high-frequency oil return that would disrupt the temperature control after the oil storage becomes too high.

[0058] Secondly, for off-grid photovoltaic power supply scenarios, the available photovoltaic power window is incorporated as a constraint or incentive into the time-domain optimization in step six, so that the oil return frequency increase is prioritized during periods of abundant photovoltaic power. In this embodiment, the available photovoltaic power at each moment is used as the upper limit constraint for the power of the oil return frequency increase, and the electricity consumption during the photovoltaic surplus period is given a low weight in the energy consumption term of the objective function (in this embodiment, the energy consumption weight during the photovoltaic surplus period is reduced to 0.3, which is the usual weight). This ensures that the 45 Hz oil return frequency increase is prioritized during the noon photovoltaic power surplus period, so as to complete the on-demand oil return while maintaining the temperature control zone of 5 degrees Celsius ± 3 degrees Celsius required for vaccines.

[0059] Thirdly, in ultra-low temperature scenarios where the extremely low evaporation temperature leads to a significant decrease in oil solubility and aggravated oil accumulation, the observer and neural network correct the strong dependence of the critical flow rate and precipitation accumulation rate on the evaporation temperature. In this embodiment, when applied to ultra-low temperature biological sample freezing scenarios with evaporation temperatures as low as -70 degrees Celsius, the liquid and gas phase densities in the critical flow rate expression are revalued according to the properties at -70 degrees Celsius, and the evaporation temperature correction factor φ is increased accordingly, so that the precipitation accumulation rate estimation and threshold triggering are adapted to the extremely low evaporation temperature conditions. Furthermore, the observer and time-series feature extraction network perform online adaptive updates based on the deviation between the measured observations and the predicted quantities during operation: In this embodiment, a recursive least squares method is used to adjust the evolution model coefficients k. acc k rem The neural network output layer bias is adjusted online with a slow time scale and a forgetting factor of 0.995 to adapt to individual differences in refrigerant charge, evaporator structure and lubricant grade. This allows for continuous correction of oil status estimation and frequency control during long-term operation without the need for individual calibration for each device.

[0060] In summary, this embodiment, through steps one to six, inverts the unobservable oil storage state of the evaporator from the measurable signals of the loop, separates the actual heat load component from the heat transfer attenuation component caused by oil storage, corrects the correspondence between frequency and cooling capacity online, and implements oil return as needed. Under the premise of ensuring oil return to the evaporator and maintaining a stable temperature in the controlled space, it achieves energy-saving refrigeration control of the variable frequency compressor.

[0061] Example 2 In a preferred embodiment of the present invention, taking an off-grid photovoltaic-powered vaccine refrigerator as an example, the mechanism-constrained oil storage state observer and its driven frequency control correction process are described in detail. The vaccine refrigerator uses a single-stage variable frequency compressor, equipped with a plate or finned tube evaporator, and the temperature control range is set to 2 degrees Celsius to 8 degrees Celsius. In order to meet the requirements of continuous day and night heat preservation and photovoltaic power supply fluctuations, the compressor operates in a deep low-frequency range far below the rated frequency for a long time, making oil return difficult and oil storage slowly accumulating. This is the operating condition that this embodiment focuses on addressing.

[0062] Specifically, the signal acquisition and preprocessing module acquires measurable signals from the loop at a fixed sampling period, preferably between 1 and 10 seconds, to capture the slow-changing process at deep low frequencies without introducing excessive noise. The acquired signals include evaporation temperature, suction pressure, suction superheat, compressor operating current, operating frequency, and controlled space temperature. Additionally, condensing temperature, compressor input power, ambient temperature, door opening signal, and photovoltaic input power are acquired as external disturbances. The raw signals are first filtered with median to remove anomalous jump values, and then denoised with a first-order low-pass or moving average filter with a time constant of 30 to 120 seconds to obtain a stable physical quantity timing sequence. Furthermore, the refrigerant mass flow rate is estimated based on the relationship between operating frequency, suction pressure, and suction superheat, according to the compressor's volumetric efficiency. Specifically, the volumetric delivery rate on the suction side is determined by the compressor's theoretical displacement, operating frequency, and volumetric efficiency. This volumetric efficiency is then converted to the refrigerant mass flow rate by combining the suction specific volume determined by suction pressure and suction superheat. The volumetric efficiency is taken as an empirical function of pressure ratio and operating frequency. The refrigerant mass flow rate is then combined with the refrigerant density and converted according to the flow cross-section of the evaporator heat exchange tubes to obtain the apparent refrigerant velocity in the evaporator and suction pipe. The current equivalent heat transfer capacity of the evaporator is estimated by the evaporation temperature, the controlled space temperature, and the refrigerant mass flow rate. Specifically, it is taken as the ratio of the current heat transfer flux to the heat transfer temperature difference, i.e., the apparent value of the product of the equivalent heat transfer coefficient and the heat transfer area.

[0063] The oil storage state observation module uses the oil storage volume in the evaporator or the corresponding oil film thermal resistance as state variables to construct a mechanism-constrained oil storage state evolution model. This evolution model expresses the rate of change of oil storage volume over time as the difference between the precipitation accumulation rate and the oil return removal rate. Specifically, the precipitation accumulation rate is taken as the product of the under-speed degree and the oil property factor: the under-speed degree is defined as the positive part of the critical flow rate minus the current apparent refrigerant flow rate, then normalized to the critical flow rate, characterizing the degree of insufficient refrigerant carrying capacity for lubricating oil; the oil property factor is an increasing function of oil solubility and oil viscosity determined by the evaporation temperature. The lower the evaporation temperature, the lower the oil solubility in the refrigerant, the more free oil precipitates, the higher the oil viscosity, and the easier it is to remain. Therefore, the oil property factor increases as the evaporation temperature decreases. The oil return removal rate is taken as an increasing function of the over-speed degree: the over-speed degree is defined as the positive part of the current apparent refrigerant flow rate minus the critical flow rate, then normalized, characterizing the refrigerant's ability to flush and carry lubricating oil. The critical flow rate varies with evaporation temperature and refrigerant properties, preferably characterized by a monotonically decreasing relationship with evaporation temperature as the main variable, meaning that the lower the evaporation temperature, the higher the required critical flow rate.

[0064] For the parts of the refrigerant-lubricating oil two-phase flow and oil film retention process that are difficult to model accurately in the above evolution model, a physically constrained neural network is used to fit the residual correction amount. Preferably, the physically constrained neural network is a fully connected network with 3 to 5 layers, with 32 to 128 neurons in each layer. The activation function is a smooth and monotonic Softplus or a modified exponential linear unit (ELU) to facilitate the application of monotonicity constraints. The network inputs are the refrigerant apparent flow rate, evaporation temperature, suction superheat, and running time, and the outputs are the correction amount for the precipitation accumulation rate and the correction amount for the oil film thermal resistance. To ensure the network output does not violate the natural law of refrigerant carrying oil, three physical constraints are enforced during training and inference: First, a non-negative constraint on oil quantity, achieved by applying a positive part to the oil quantity state or using non-negative parameterization; second, a monotonic constraint on the accumulation rate with respect to the degree of underspeed, achieved by applying a non-negative penalty to the partial derivative of the network with respect to the underspeed input; and third, a monotonic constraint on oil film thermal resistance with respect to oil quantity, achieved by applying a non-negative penalty to the partial derivative of the oil film thermal resistance mapping with respect to oil quantity. The total loss during training is preferably a weighted sum of the data fitting term and the physical constraint penalty term. The data fitting term is the mean square error between the observed model output's equivalent heat transfer capacity and the measured equivalent heat transfer capacity. The physical constraint penalty term is the weighted sum of the aforementioned monotonicity and non-negativity violations. The weights are selected based on a trade-off between fitting accuracy and constraint satisfaction on the validation set. The weight of the physical constraint penalty term is preferably 0.1 to 10 times the weight of the data fitting term and is gradually increased during training.

[0065] Furthermore, the oil storage state observation module superimposes the oil film thermal resistance onto the heat transfer thermal resistance of the evaporator when there is no oil storage, obtaining the equivalent heat transfer capacity affected by oil storage, and mapping it to the operating cost required to reach the target temperature at the current operating frequency. Using the equivalent heat transfer capacity and operating cost as observations, the state observer in the form of extended Kalman filtering or unscented Kalman filtering recursively estimates the oil storage state. In each sampling period, the oil storage state and its covariance are first predicted by the evolution model and the neural network correction quantity, and then the state estimation is obtained by correcting the difference between the predicted observation and the measured observation through Kalman gain, so as to minimize the deviation between the observation model output and the measured observation, and output the oil storage state estimate and its uncertainty. The uncertainty is given by the state estimation error covariance, which is used for subsequent frequency correction and oil return trade-off assessment to evaluate the reliability of the estimate.

[0066] The frequency control correction module corrects the nominal frequency-cooling capacity relationship based on the oil film thermal resistance corresponding to the estimated oil storage state. Specifically, the nominal relationship is given by the frequency-cooling capacity calibration curve under the no-oil-storage benchmark. After substituting the oil film thermal resistance into the heat transfer calculation, the reduction coefficient of cooling capacity due to current oil storage-induced heat transfer attenuation is obtained. Based on this, a real-time relationship reflecting the current oil storage-induced heat transfer attenuation is obtained, enabling the control system to output a target operating frequency higher than the nominal value, taking into account heat transfer attenuation, under the same target cooling capacity. At the same time, the required cooling capacity is determined only based on the actual heat load component obtained by the load and oil storage decoupling discrimination module, avoiding the miscalculation of oil storage-induced attenuation into the load and incorrect frequency reduction. The corrected real-time relationship of the required cooling capacity is converted into the target operating frequency and output to the frequency converter. When the decoupling discrimination module determines that the decrease in heat transfer capacity is mainly due to oil storage-induced attenuation rather than a decrease in actual load, the frequency control correction module suppresses the tendency to continue reducing the frequency, thereby breaking the self-reinforcing closed loop where low-frequency operation exacerbates oil storage, and oil storage attenuation is misread as the basis for frequency reduction.

[0067] In off-grid photovoltaic power supply scenarios, the oil return energy-saving trade-off module incorporates the available photovoltaic power window as a constraint or incentive into the objective function solution in the rolling time domain. Specifically, the objective function integrates three terms in the rolling time domain: energy-saving objective, temperature control deviation objective, and the future heat exchange attenuation cost caused by oil accumulation. It uses the oil accumulation evolution model as the state constraint and the temperature control zone of the controlled space not exceeding 2 to 8 degrees Celsius as the inequality constraint. When solving for the optimal operating frequency sequence of the objective function, the photovoltaic power prediction window is introduced as an incentive term, so that the increase in oil return frequency with amplitude and duration matching the estimated oil accumulation is prioritized during periods of sufficient photovoltaic power. Preferably, when the estimated oil level exceeds a dynamic threshold that varies with evaporation temperature and operating conditions, and when it is predicted that photovoltaic power will enter a sufficient window, the oil return frequency is increased. This causes the apparent refrigerant flow rate to exceed the critical flow rate required for the lubricating oil to be carried back to the compressor by the refrigerant, thus clearing the oil. If the oil level is close to the threshold but the photovoltaic power is not yet sufficient, the oil return frequency is appropriately advanced or delayed within the temperature control range to avoid high-frequency oil return during nighttime periods when energy storage is limited. After the oil return is completed, the oil level observer confirms that the estimated oil level has fallen below the threshold, ending the oil return and resuming energy-saving operation. Therefore, the oil return amount and increase are determined as needed based on the estimated oil level rather than at fixed time intervals, avoiding excessive or insufficient forced oil flushing under conservative presets, while matching the oil return energy consumption with the photovoltaic power generation rhythm.

[0068] By employing the above-described preferred embodiment, without installing any oil retention sensors within the evaporator, the oil retention state, which is not directly observable in the evaporator, can be retrieved online from the measurable signals in the loop. This answers the question of how much lubricating oil is currently retained, a question that existing low-frequency oil return technologies cannot answer. Compared to timed oil return strategies that conservatively preset based on worst-case scenarios and forcibly increase the frequency of flushing over a fixed cumulative duration, this method triggers oil return on demand based on the estimated oil retention amount. This reduces unnecessary high-frequency flushing times and corresponding energy consumption. Especially in scenarios with off-grid photovoltaic power supply and limited energy storage capacity, the timing of oil return matches the photovoltaic power window, reducing additional energy storage requirements. Since frequency control tracks the separated real heat load component rather than the correspondence of hidden contamination caused by oil retention attenuation, the operating frequency is no longer mistakenly increased due to heat transfer attenuation caused by oil retention being interpreted as increased load, nor is it mistakenly decreased due to similar appearances being interpreted as decreased load. This makes energy-saving operation in the deep low-frequency range more accurate and temperature control more stable. The physical constraint neural network's fitting of the residuals is limited to a feasible region that satisfies the following conditions: non-negative oil content, monotonicity with underspeed, and monotonicity of oil film thermal resistance with oil content. Even under extremely low-frequency and extremely low-temperature conditions with sparse training data, the network output will not provide estimation results that violate the natural laws of refrigerant carrying oil, thus enhancing the robustness of the estimation. The uncertainty of the observer output further provides a confidence basis for oil return and frequency reduction decisions. When the estimation confidence is low, a relatively conservative approach can be taken to avoid erroneous frequency adjustments caused by miscalculations.

[0069] The aforementioned beneficial effects are achieved through a technical principle: although the amount of oil in the evaporator cannot be directly measured by any sensor within the circuit, its accumulation and removal strictly follow the refrigerant's carry-over law for lubricating oil. When the apparent flow rate of the refrigerant is lower than the critical carry-over flow rate that varies with the evaporation temperature, the lubricating oil precipitates and remains; when it is higher than the critical flow rate, it is flushed back. Therefore, the rate of change of the amount of oil over time can be expressed as the difference between the precipitation accumulation rate, determined by the degree of under-speed and over-speed, and the oil return removal rate. This mechanism establishes a causal relationship between the unobservable state and measurable physical quantities such as apparent flow rate and evaporation temperature, allowing the oil state to be mathematically inverted from measurable observables by a state observer. This is the fundamental reason why the amount of oil can be known without the need for additional sensors. The reason for using a physically constrained neural network to fit the two-phase retention residuals, which are difficult to characterize precisely, instead of allowing an unconstrained network to fit freely, is that the state of oil storage has no direct label and the samples are sparse under extreme conditions. Without constraints, the network may fit results that violate physics, such as negative oil storage or non-monotonicity with the degree of underspeed, in the sparse data region. Applying monotonicity and non-negativity constraints is equivalent to injecting prior physical knowledge into the solution space, restricting the network output to the physically feasible region. This ensures that the estimation does not deviate from natural laws while maintaining the flexibility of the fitting. This is the principle behind the enhanced robustness. The reason why the actual heat load change and the heat transfer decay caused by oil storage can be separated is that although they are highly similar in steady-state phenomena such as evaporation temperature, suction pressure, and superheat, they have essential differences in dynamic characteristics: the actual heat load response is rapidly reversible with frequency adjustment and is related to external disturbances such as ambient temperature, door opening, and photovoltaic input, while the oil storage-induced decay accumulates slowly and monotonically with the duration of deep low-frequency operation, has a slow response to frequency changes, and exhibits path-dependent hysteresis characteristics. It is by utilizing this dynamic difference for attribution that the control system can distinguish between two causes with opposite directions, avoid erroneous frequency reduction, and thus break the self-reinforcing closed loop. When the external disturbance in the input signal is nearly constant for a long time and the system has not experienced a frequency change sufficient to excite different responses from the two causes for a long time, the separability of the two components in dynamic characteristics will decrease. At this time, the decoupling accuracy may be affected, which can be compensated for by active excitation in subsequent implementation methods.

[0070] Example 3 In another embodiment of the present invention, an ultra-low temperature biological sample cryopreservation facility is used as the application scenario, and an alternative solution that differs from the aforementioned passive characteristic attribution is adopted in the decoupling technology path between load and oil storage. The target temperature of the cryopreservation facility is -60 degrees Celsius to -80 degrees Celsius, with extremely low evaporation temperature. The solubility of lubricating oil in the refrigerant decreases significantly with decreasing temperature, while the oil viscosity increases. The tendency for free oil to precipitate and remain is significantly aggravated. Moreover, the heat load in this scenario is nearly constant over a long period, with few external disturbances. The aforementioned passive characteristic attribution, which relies on the correlation of external disturbances and natural frequency fluctuations, is insufficiently separable under such conditions. Therefore, this embodiment uses active micro-perturbation excitation and hysteresis area, which directly reflects the causal criterion of heat transfer attenuation caused by oil storage, as the core decoupling method.

[0071] As an alternative, the load and oil storage decoupling discrimination module no longer primarily relies on multi-scale passive feature extraction of steady-state deviations. Instead, it actively superimposes a sequence of minute frequency disturbances, whose amplitude and duty cycle are constrained by a closed-loop temperature stability band in the controlled space, onto the operating frequency as an excitation. Specifically, the minute frequency disturbances are preferably symmetrical square or triangular wave jitters, with amplitudes ranging from 1% to 5% of the rated frequency and periods on the order of several minutes. During each disturbance period, the amplitude is limited in real-time by a closed-loop temperature stability band to ensure that the temperature of the controlled space never exceeds the allowable temperature control zone of the freezer. When the temperature approaches the boundary of the temperature control zone, the disturbance amplitude is automatically reduced or the excitation is paused. The response of the equivalent heat transfer capacity to the small frequency disturbance is analyzed by extracting two indicators: frequency domain symmetry and time domain hysteresis area. The actual heat load component exhibits a rapid and symmetrical response to both positive and negative disturbances, with its response trajectory on the equivalent heat transfer capacity-frequency plane almost overlapping and the enclosed hysteresis area approaching zero. However, the component of heat transfer attenuation due to oil accumulation is asymmetrical because the flow velocity thresholds corresponding to oil precipitation and removal are inherently asymmetrical. When the frequency increases, the apparent flow velocity exceeds the critical flow velocity, triggering oil return and removal; when the frequency decreases, it exceeds the critical flow velocity, triggering precipitation accumulation. The asymmetry of the thresholds and rates of these two components causes the response trajectory of the equivalent heat transfer capacity to the frequency increase and decrease segments to not overlap, forming a quantifiable hysteresis loop on this plane. The area of ​​the hysteresis loop is used as a direct observation of the oil storage-induced heat transfer attenuation component. Its magnitude increases with the increase of oil storage volume. Based on this, the oil storage-induced heat transfer attenuation component in the equivalent heat transfer capacity deviation is separated, and the remaining part is attributed to the real heat load component. The hysteresis area is fed back to the oil storage state observer as an observation that directly reflects the oil storage-induced attenuation and participates in the state correction.

[0072] In the specific extraction of the hysteresis area, it is preferable to use multiple disturbance cycles as windows, pair the equivalent heat transfer capacity characterization quantity with the concurrent operating frequency into ordered data pairs, fit the response curve of the equivalent heat transfer capacity to the frequency according to the rising and falling frequency segments, and then use the area of ​​the closed region enclosed by the two curves as the hysteresis area. In order to suppress measurement noise, the hysteresis area obtained from several consecutive disturbance cycles can be weighted and averaged, and the average window length can be adjusted according to the uncertainty of the oil storage state given by the observer. When the uncertainty is high, the window length can be extended to obtain a more stable index. To further improve the separability during periods of sparse disturbance, a lightweight temporal feature extraction network can be used to process the aforementioned frequency increase / decrease response pairs and their hysteresis indices. This network outputs the ratio of the actual heat load component to the component of heat transfer attenuation caused by oil storage. The estimated value of the oil storage state output by the observer is used as a priori constraint on the component of heat transfer attenuation caused by oil storage, ensuring that the sum of the two components deviates from the measured equivalent heat transfer capacity. This temporal feature extraction network preferably adopts a one-dimensional convolutional or gated recurrent structure. A scale of one to two layers and a number of channels of 16 to 64 are sufficient to meet the requirements of online computation. It does not need to use a network of the same scale as the passive attribution method.

[0073] Furthermore, considering the extremely low evaporation temperature under ultra-low temperature conditions, this embodiment modifies the temperature dependence of the critical flow rate and precipitation accumulation rate in the oil storage state observer. Specifically, the critical flow rate is expressed as a monotonically decreasing relationship with evaporation temperature as the main variable. For every certain decrease in evaporation temperature, the critical flow rate increases accordingly to reflect the physical fact that higher flow rates are needed to carry high-viscosity lubricating oil at low temperatures. Simultaneously, the growth rate of oil property factors in the precipitation accumulation rate as evaporation temperature decreases is increased, enabling the observer to predict faster oil accumulation at extremely low evaporation temperatures, consistent with the actual decrease in oil solubility and intensified precipitation at low temperatures. Correspondingly, the dynamic threshold of the oil return energy-saving trade-off module is appropriately lowered under ultra-low temperature conditions, and the amplitude of the increase in oil return frequency is appropriately increased, so as to complete on-demand oil return with the smallest possible amplitude and frequency while maintaining the required ultra-low temperature control zone for the sample.

[0074] When using a rolling time-domain tradeoff, this implementation can also schedule the oil return frequency increase during periods of higher system margin. However, its triggering no longer depends on external load fluctuations, but is jointly determined by the hysteresis area measured by the active perturbation and the oil storage state estimate output by the observer: oil return is triggered when the hysteresis area and the oil storage estimate both exceed the dynamic threshold. After oil return is completed, oil return ends only when the hysteresis area falls back and the observer confirms that the oil storage estimate has decreased. As an optional alternative, the causal criterion of active perturbation excitation and hysteresis area can also be used in conjunction with the passive characteristic attribution in the aforementioned implementation. When external perturbations are sufficient, passive attribution is the main factor, and when perturbations are scarce, active excitation is the main factor, with the two complementing each other. It should be noted that although the active perturbation is constrained by the temperature stability band, when the heat capacity of the controlled space is small and the temperature control band is extremely narrow, the amplitude of the injectable perturbation is limited, and the signal-to-noise ratio of the hysteresis area may decrease. In this case, it is advisable to appropriately extend the perturbation period or increase the multi-period averaging to extract a stable hysteresis index. When the oil content is extremely small and insufficient to form a distinguishable hysteresis loop, the ability of this method to distinguish trace oil content also decreases accordingly. It can be combined with the mechanistic prediction of the observer to cover the low oil content range. Therefore, in the scenario of continuous operation at ultra-low temperature with constant heat load and few external perturbations, this embodiment can still reliably separate the two causes of attenuation, namely, the actual heat load and oil content, by actively injecting controlled excitation and using hysteresis asymmetry as the causal criterion. This constitutes a substantial difference in technical approach from the aforementioned embodiment that mainly uses passive characteristic attribution.

[0075] Example 4 This embodiment provides an energy-saving refrigeration control system for a variable frequency compressor, used to implement the control method described in Embodiment 1. Its overall structure is as follows: Figure 3 As shown, the system includes a signal acquisition and preprocessing module, an oil storage status observation module, a load and oil storage decoupling discrimination module, a frequency control correction module, and a return oil energy-saving trade-off module. Each of these modules corresponds to the steps in Example 1. Specifically, the signal acquisition and preprocessing module, the oil storage status observation module, and the load and oil storage decoupling discrimination module execute steps one to three in sequence, the frequency control correction module executes steps four and five, and the return oil energy-saving trade-off module executes step six. The function of each module is consistent with the description of the corresponding method steps, and the corrected target operating frequency is output to the frequency converter to control the compressor operation.

[0076] The signal acquisition and preprocessing module is used to acquire and preprocess the measurable loop signals of the variable frequency compressor refrigeration system during operation to obtain the timing sequence of physical quantities reflecting the system's operating status. The measurable loop signals include evaporation temperature, suction pressure, suction superheat, compressor operating current, operating frequency, and controlled space temperature. This module is further configured to: acquire the measurable loop signals at a fixed sampling period and perform filtering and noise reduction; firstly, apply a sliding median filter to remove pulse-type abnormal jump values, and then apply a first-order low-pass filter to suppress measurement noise; estimate the refrigerant mass flow rate based on the compressor's volumetric efficiency relationship using the operating frequency, suction pressure, and suction superheat, and calculate the apparent refrigerant velocity in the evaporator and suction pipe by combining the evaporator heat exchanger tube geometric parameters and refrigerant density; and estimate the current equivalent heat transfer capacity of the evaporator using the evaporation temperature, controlled space temperature, and refrigerant mass flow rate. In terms of hardware deployment, the module consists of a sensor array composed of temperature sensors, pressure transmitters and current transformers arranged at various measuring points in the refrigeration circuit. It is connected to the controller via analog-to-digital conversion and data acquisition interface; slow variables and fast variables are collected separately using different sampling periods.

[0077] The oil storage state observation module is used to construct and operate an online oil storage state observer constrained by a mechanism. It expresses the time-varying rate of change of the oil storage volume in the evaporator as the difference between the precipitation accumulation rate and the oil return removal rate. From the time series of these physical quantities, it inversely retrieves the oil storage state of the evaporator, which is not directly observable. The oil storage state is characterized by the oil storage volume in the evaporator or its corresponding oil film thermal resistance. This module is further configured as follows: the precipitation accumulation rate is taken as an increasing function of the degree of under-velocity (the apparent refrigerant flow rate being lower than the critical carrying velocity) and the oil solubility and viscosity determined by the evaporation temperature; the oil return removal rate is taken as an increasing function of the degree of over-velocity (the apparent refrigerant flow rate being higher than the critical carrying velocity), where the critical carrying velocity varies with the evaporation temperature and refrigerant properties; the oil film thermal resistance is superimposed on the evaporator heat transfer thermal resistance to obtain the equivalent heat transfer capacity affected by oil storage and mapped as an operating cost. Using the equivalent heat transfer capacity and operating cost as observations, the state observer recursively estimates the oil storage state, outputting the estimated oil storage state value and its uncertainty. This module also includes a physical constraint neural network unit, which is used to fit the residuals in the evolution model and the observation model that are difficult to model accurately. It takes the refrigerant apparent flow rate, evaporation temperature, suction superheat and running time as inputs, and outputs the correction amount for precipitation accumulation rate and oil film thermal resistance. It is configured to force the physical constraints of non-negative oil content, monotonic accumulation rate with underspeed degree, and monotonic oil film thermal resistance with oil content.

[0078] The load and oil storage decoupling discrimination module is used to decouple the deviation of the evaporator's equivalent heat transfer capacity characterization quantity relative to its oil-free baseline value through a dual-channel process, separating it into a true heat load component and an oil storage-induced heat transfer attenuation component. This module is further configured to: extract the rate of change, monotonicity, reversibility, and correlation with external disturbances of the deviation at multiple time scales; process the above features using a time-series feature extraction network, output the ratio of the true heat load component to the oil storage-induced heat transfer attenuation component, and use the oil storage state estimate output by the oil storage state observation module as a priori constraint for the oil storage-induced heat transfer attenuation component, ensuring that the sum of the two components is consistent with the deviation. Preferably, this module is also configured to superimpose a small frequency disturbance with amplitude and duty cycle constrained by the controlled space temperature steady band onto the operating frequency as an excitation, extract the frequency domain symmetry and time domain hysteresis area of ​​the equivalent heat transfer capacity response to the disturbance, and use the hysteresis area as the observed quantity of the oil storage-induced heat transfer attenuation component to feed back to the oil storage state observation module.

[0079] The frequency control correction module is used to correct the correspondence between operating frequency and cooling capacity online based on the oil storage state and the actual heat load component. It determines the required cooling capacity only based on the actual heat load component, calculates the corrected correspondence of the required cooling capacity as the target operating frequency, and outputs it to the inverter driver to drive the compressor. This module is further configured to: correct the nominal frequency-cooling capacity correspondence using the oil film thermal resistance corresponding to the estimated oil storage state, obtaining a real-time correspondence reflecting the current oil storage-induced heat exchange attenuation; when the decoupling discrimination module determines that the decrease in heat exchange capacity is mainly due to oil storage-induced attenuation rather than a reduction in actual load, it suppresses the tendency to continue reducing the frequency, thereby breaking the self-reinforcing closed loop where low-frequency operation exacerbates oil storage, and oil storage attenuation is misinterpreted as a basis for frequency reduction.

[0080] The oil return energy-saving trade-off module is used to dynamically trade off oil return and energy saving based on the estimated oil storage status. This module is further configured to: incorporate the energy-saving target, temperature control deviation target, and the future heat exchange attenuation cost caused by oil accumulation into a rolling time-domain objective function, using the evolution model of evaporator oil storage as a state constraint; solve for the optimal operating frequency sequence of the objective function within the rolling time domain; when the estimated oil storage status exceeds a dynamic threshold that varies with evaporation temperature and operating conditions, output an increase in oil return frequency with amplitude and duration matching the estimated oil storage; after oil return is completed, the oil storage status observation module confirms that the estimated oil storage status has fallen back, ending oil return and resuming energy-saving operation.

[0081] The data flow relationship between the above modules is as follows: the signal acquisition and preprocessing module outputs physical quantity timing, refrigerant apparent flow rate and equivalent heat transfer capacity characterization quantity as common inputs to the oil storage state observation module and the load and oil storage decoupling discrimination module; the oil storage state estimate output by the oil storage state observation module is used as a priori constraint for the decoupling discrimination module and sent to the frequency control correction module and the oil return energy-saving trade-off module; the actual heat load component output by the decoupling discrimination module is sent to the frequency control correction module, and the extracted hysteresis area is fed back to the oil storage state observation module; the target operating frequency obtained by the frequency control correction module and the oil return energy-saving trade-off module is finally output to the frequency converter driver.

[0082] In terms of hardware deployment, the functional modules of this system can be embedded as software program modules in the controller built into the frequency converter drive, or deployed in an independent control unit or edge computing device that communicates with the frequency converter drive. The controller includes a processor and a memory. The memory stores program instructions that can be executed by the processor, which then executes these instructions to implement the functions of the aforementioned modules. Signals from the sensor array enter the controller via the acquisition interface. The target operating frequency calculated by the controller is sent to the power conversion circuit of the frequency converter drive via the communication interface, which adjusts the compressor's operating frequency. The model parameters of the neural network unit can be stored in memory after offline training and are adaptively updated online during long-term operation based on the deviation between measured observations and predicted values ​​to adapt to individual differences in refrigerant charge, evaporator structure, and lubricating oil grade. The above deployment methods are general descriptions and do not depend on specific brands or models of hardware.

[0083] Example 5 This embodiment uses a continuous temperature-controlled refrigeration scenario of a precision constant temperature cabinet as an example to illustrate the implementation process and technical effects of the present invention. This precision constant temperature cabinet is used for long-term constant temperature preservation of temperature-sensitive materials. The target temperature control value of the controlled space is set at 5 degrees Celsius, with an allowable temperature control range of ±3 degrees Celsius. It employs a rolling rotor compressor driven by a variable frequency drive, using R290 as the refrigerant. Due to the consistently low and relatively stable heat load within the cabinet, the compressor operates continuously in the deep low-frequency range of 18 to 30 Hz for extended periods, resulting in an evaporation temperature that remains consistently between -25 and -30 degrees Celsius. The specific difficulty in this scenario lies in the fact that the change in heat load is extremely weak and slow, which is highly similar to the heat transfer decay caused by oil accumulation in terms of observable phenomena such as evaporation temperature, suction pressure, and superheat. Conventional control methods based on apparent quantities are difficult to distinguish between the two, often misjudging the decrease in heat transfer capacity caused by oil accumulation as a decrease in load and continuously reducing the frequency, which in turn exacerbates the accumulation of oil and causes long-term drift in temperature control accuracy. On the other hand, the conventional method of forced high-frequency oil return based on fixed time intervals will periodically disturb the temperature stability inside the cabinet, making it difficult to meet the strict requirements of precise temperature control.

[0084] The specific implementation process of the method of this invention is as follows. First, the signal acquisition and preprocessing module acquires slow and fast variables at sampling periods of 1 second and 100 milliseconds respectively, and filters and removes noise. The apparent flow velocity of the refrigerant in the heat exchange tube of the evaporator is estimated to be about 1.7 m / s, which is significantly lower than the critical carry-over velocity of about 3.5 m / s at this evaporation temperature. Second, the oil storage state observation module recursively estimates the oil storage amount using an extended Kalman observer, and the physical constraint neural network corrects the residuals that are difficult to model online. After running for about 6 hours, the observer outputs an estimated oil storage amount of about 3 grams in the evaporator, corresponding to an oil film thermal resistance of about 0.008 Kelvin per watt. The equivalent heat transfer capacity is reduced from the baseline of about 12 watts per Kelvin to about 10.9 watts per Kelvin. Subsequently, the load and oil storage decoupling discrimination module extracts the multi-scale dynamic characteristics of the deviation. Combined with the hysteresis area of ​​approximately 0.4 Hz multiplied by watts per Kelvin obtained from superimposed small frequency disturbances with an amplitude of approximately ±0.5 Hz and a period of approximately 120 seconds, it is determined that the proportion of the oil storage-induced heat transfer attenuation component in the deviation exceeds 0.6, thus suppressing frequency reduction. Based on this, the frequency control correction module corrects the nominal cooling capacity of 20 Hz from 68 watts to approximately 62 watts and appropriately increases the target frequency. When the estimated oil storage exceeds the dynamic threshold of approximately 2.3 grams at the evaporation temperature, the oil return energy-saving trade-off module, under rolling time-domain optimization, outputs a frequency increase that raises the frequency from 20 Hz to approximately 45 Hz and maintains it for approximately 90 seconds, causing the apparent flow velocity to rise to approximately 4.2 m / s, exceeding the critical carry-over velocity. The stored oil is then carried back to the compressor by the refrigerant (the physical process is as follows). Figure 4 (As shown); During the oil return process, the temperature inside the cabinet is maintained within the allowable temperature control range. The oil return process ends and energy-saving operation resumes when the estimated oil level drops below approximately 0.5 grams.

[0085] To verify the effectiveness of the technology, a comparative evaluation was conducted on the same constant temperature cabinet for 30 consecutive days. The evaluation indicators were the temperature pass rate within the controlled temperature zone, the accuracy rate of correctly attributing the implicit deviation caused by heat exchange attenuation due to oil storage, and the energy consumption per unit of cooling capacity. The results showed that when using the method of this invention, the temperature pass rate of the controlled space was approximately 98.6%, and the accuracy rate of cause identification was approximately 93.2%, which is a significant improvement compared to the conventional method that does not distinguish between load and oil storage and only adjusts the frequency based on the apparent heat exchange capacity (temperature pass rate approximately 90.4%, no cause identification capability, and energy consumption per unit of cooling capacity is approximately 11% higher). Further ablation comparisons revealed the following: When the oil storage state observation module was removed (i.e., oil storage state was not inverted, and frequency adjustment was directly based on apparent deviation), the accuracy of cause identification dropped to approximately 78.5%, and the temperature compliance rate dropped to approximately 93.1%, indicating that oil storage state inversion significantly contributed to correct attribution. When the observation module was retained but the dual-channel decoupling discrimination module was removed (i.e., the actual heat load component and the oil storage-induced heat transfer attenuation component were not separated), the temperature compliance rate dropped to approximately 94.8%, and the unit cooling capacity energy consumption increased by approximately 6%, indicating that decoupling discrimination significantly contributed to avoiding erroneous frequency reduction and ensuring energy-saving accuracy. When the on-demand oil return of the oil return energy-saving trade-off module was removed and replaced with forced oil return at fixed time intervals, although the oil storage was cleared, the temperature fluctuation amplitude increased and the compliance rate within the temperature control zone dropped to approximately 95.5% due to the mismatch between the oil return timing and the load inside the cabinet. In multiple batches of assessments under different months and different ambient temperature conditions, the above indicators showed reasonable fluctuations of approximately ±1.5 percentage points, consistent with the characteristics of real continuous operation. Furthermore, under four typical operating conditions—light load and low frequency, medium load, heavy load, and frequent door opening disturbances—the daily power consumption and temperature fluctuation amplitude of the method of this invention were compared with those of the undecoupled control. The results are as follows: Figure 5 As shown, the daily power consumption of this invention is lower than that of undecoupled control under all operating conditions (e.g., from approximately 12.6 kWh to approximately 10.4 kWh under light load and low frequency conditions), and the temperature fluctuation of the controlled space is also significantly reduced (e.g., from approximately 2.4 Kelvin to approximately 0.9 Kelvin under conditions with frequent door opening disturbances). This indicates that the improvements in energy saving and stable temperature control are valid under all operating conditions. The above data shows that this invention, through the synergy of oil storage state inversion, dual-channel decoupling, and on-demand oil return, simultaneously achieves a balance between temperature control accuracy, cause identification, and energy-saving operation in scenarios where conventional methods are difficult to implement, such as precision constant temperature continuous refrigeration. This verifies the feasibility of the technical solution and the technical contributions of each innovation.

[0086] It should be noted that the functions corresponding to the various method steps of the present invention can also be carried by a computer-readable storage medium: the computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the variable frequency compressor energy-saving refrigeration control method described in Embodiment 1. Accordingly, the present invention can also be implemented as an electronic device, including a processor and a memory, wherein the memory stores a computer program that can be executed by the processor, and the processor, when executing the computer program, implements the steps of the above method.

[0087] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A variable frequency compressor energy-saving refrigeration control method, characterized in that, include: Step 1: Collect and preprocess the measurable signals of the variable frequency compressor refrigeration system to obtain the timing sequence of physical quantities reflecting the system's operating status. The measurable signals of the loop include evaporation temperature, suction pressure, suction superheat, compressor operating current, operating frequency, and controlled space temperature. Step 2: Construct an oil storage state observer constrained by mechanism, and express the time change rate of oil storage in the evaporator by the difference between the precipitation accumulation rate and the return oil removal rate. The difference expression constitutes the evolution model of oil storage in the evaporator. The oil storage state that cannot be directly observed in the evaporator is inverted from the time series of the physical quantities and the estimated value of oil storage state is output. Step 3: Decouple the deviation of the equivalent heat transfer capacity of the evaporator relative to its oil-free baseline value through two channels, separating it into the actual heat load component and the heat transfer attenuation component caused by oil storage. Step 4: Based on the oil storage status and the actual heat load component, correct the correspondence between the operating frequency and the cooling capacity online, determine the required cooling capacity only according to the actual heat load component, and calculate the corrected correspondence of the required cooling capacity as the target operating frequency; Step 5: Output the target operating frequency to the frequency converter to drive the compressor.

2. The variable frequency compressor energy-saving refrigeration control method according to claim 1, characterized in that, Step one also includes: estimating the refrigerant mass flow rate based on the relationship between the operating frequency, the suction pressure, and the suction superheat and the compressor volumetric efficiency, and calculating the apparent flow rate of the refrigerant in the evaporator and suction pipe by combining the geometric parameters of the evaporator heat exchange tube with the refrigerant density; estimating the current equivalent heat exchange capacity characterization quantity of the evaporator based on the evaporation temperature, the controlled space temperature, and the refrigerant mass flow rate.

3. The variable frequency compressor energy-saving refrigeration control method according to claim 2, characterized in that, In step two, the precipitation accumulation rate is an increasing function of the degree of under-velocity (the apparent flow rate of the refrigerant is lower than the critical flow rate) and the oil solubility and viscosity determined by the evaporation temperature. The oil return removal rate is an increasing function of the degree of over-velocity (the apparent flow rate of the refrigerant is higher than the critical flow rate). The critical flow rate varies with the evaporation temperature and refrigerant properties. The residuals in the oil storage state observer that are difficult to model accurately are fitted using a physical constraint neural network. This physical constraint neural network takes the apparent flow rate of the refrigerant, the evaporation temperature, the suction superheat, and the running time as inputs, and outputs corrections to the precipitation accumulation rate and oil film thermal resistance. It also forces the oil storage volume to be non-negative, the accumulation rate to be monotonic with the degree of under-velocity, and the oil film thermal resistance to be monotonic with the oil storage volume.

4. The variable frequency compressor energy-saving refrigeration control method according to claim 3, characterized in that, Step two further includes: superimposing the oil film thermal resistance onto the evaporator heat exchange thermal resistance to obtain the equivalent heat exchange capacity affected by oil storage, and mapping it to the operating cost required to reach the target temperature at the current operating frequency; using the equivalent heat exchange capacity and the operating cost as observations, using a state observer to recursively estimate the oil storage state, minimizing the deviation between the observation model output and the measured observations, and outputting the oil storage state estimate and its uncertainty.

5. The variable frequency compressor energy-saving refrigeration control method according to claim 1, characterized in that, Step three includes: extracting the rate of change, monotonicity, reversibility, and correlation with external disturbances of the deviation at multiple time scales; processing the above features using a time-series feature extraction network, outputting the ratio of the actual heat load component to the oil storage-induced heat transfer attenuation component, and using the oil storage state estimate output in step two as a priori constraint for the oil storage-induced heat transfer attenuation component, so that the sum of the two components is consistent with the deviation; wherein the actual heat load component is rapidly reversible in response to frequency adjustment and is correlated with the external disturbance, and the oil storage-induced heat transfer attenuation component slowly and monotonically accumulates with the duration of deep low-frequency operation, has a slow response to frequency changes, and exhibits hysteresis characteristics.

6. The variable frequency compressor energy-saving refrigeration control method according to claim 5, characterized in that, Step three also includes: superimposing a small frequency disturbance with amplitude and duty cycle constrained by the temperature stability band of the controlled space onto the operating frequency as an excitation; extracting the frequency domain symmetry and time domain hysteresis area from the response of the equivalent heat transfer capacity characterization quantity to the small frequency disturbance; the real heat load component exhibits a fast symmetrical response to positive and negative disturbances; the oil storage-induced heat transfer attenuation component exhibits a hysteresis area; and using this hysteresis area as the observation of the oil storage-induced heat transfer attenuation component is fed back to the oil storage state observer.

7. The variable frequency compressor energy-saving refrigeration control method according to claim 1, characterized in that, The process also includes step six, which involves a dynamic trade-off between oil storage perception and oil return and energy saving: The energy saving target, temperature control deviation target, and the future heat exchange attenuation cost caused by oil accumulation are incorporated into the objective function in the rolling time domain, with the evolution model of the evaporator oil storage as the state constraint. The optimal operating frequency sequence for the objective function is solved in the rolling time domain. When the estimated oil storage state exceeds a dynamic threshold that varies with the evaporation temperature and operating conditions, the oil return frequency, with an output amplitude and duration matching the estimated oil storage, is increased. This causes the apparent refrigerant flow rate to exceed the critical flow rate required for the lubricating oil to be carried back to the compressor by the refrigerant, thus clearing the oil. After the oil return is completed, the oil storage state observer confirms that the estimated oil storage state has fallen back, ending the oil return process and resuming energy-saving operation.

8. The variable frequency compressor energy-saving refrigeration control method according to claim 7, characterized in that, For off-grid photovoltaic-powered vaccine refrigeration or cryogenic biological sample freezing continuous operation scenarios, the available photovoltaic power window is incorporated as a constraint or incentive into the solution of the objective function in the rolling time domain, so that the increase in oil return frequency is prioritized during periods of sufficient photovoltaic power; for cryogenic scenarios, the oil storage state observer corrects the dependence of the carrying critical flow rate and the precipitation accumulation rate based on the evaporation temperature.

9. The variable frequency compressor energy-saving refrigeration control method according to claim 1, characterized in that, Also includes: The oil condition observer is updated online adaptively based on the deviation between the measured and predicted values ​​during operation, in order to adapt to individual differences in refrigerant charge, evaporator structure, and lubricating oil grade.

10. A variable frequency compressor energy-saving refrigeration control system, characterized in that, include: The signal acquisition and preprocessing module is used to acquire and preprocess the loop measurable signals of the variable frequency compressor refrigeration system to obtain the timing of physical quantities reflecting the system's operating status. The loop measurable signals include evaporation temperature, suction pressure, suction superheat, compressor operating current, operating frequency, and controlled space temperature. The oil storage state observation module is used to construct a mechanism-constrained oil storage state observer. The difference between the precipitation accumulation rate and the return oil removal rate is used to express the time change rate of the oil storage in the evaporator. The difference expression constitutes the evolution model of the oil storage in the evaporator. The oil storage state that cannot be directly observed in the evaporator is inverted from the time series of the physical quantities and the estimated value of the oil storage state is output. The load and oil storage decoupling discrimination module is used to decouple the deviation of the evaporator's equivalent heat transfer capacity characterization quantity relative to its oil-free reference value in two channels, separating it into the real heat load component and the oil storage-induced heat transfer attenuation component. The frequency control correction module is used to correct the correspondence between the operating frequency and the cooling capacity online based on the oil storage status and the actual heat load component. It determines the required cooling capacity only according to the actual heat load component, calculates the corrected correspondence of the required cooling capacity as the target operating frequency, and outputs it to the frequency converter to drive the compressor to run.

11. The variable frequency compressor energy-saving refrigeration control system according to claim 10, characterized in that, Also includes: The oil return energy-saving trade-off module is used to perform a dynamic trade-off between oil return and energy saving based on the estimated oil storage state. The dynamic trade-off between oil return and energy saving includes: incorporating the energy saving target, the temperature control deviation target, and the future heat exchange attenuation cost caused by oil accumulation into the objective function in the rolling time domain, and using the evolution model of the evaporator oil storage as the state constraint, solving for the optimal operating frequency sequence of the objective function in the rolling time domain. When the estimated oil storage state exceeds the dynamic threshold that changes with the evaporation temperature and operating conditions, the module outputs an increase in the oil return frequency with an amplitude and duration that matches the estimated oil storage.

Images