Multi-model adaptive control method based on robust optimization
By adopting a multi-model adaptive control method based on robust optimization, the problems of temperature stability and energy consumption optimization of smart refrigerators under complex operating conditions are solved, and efficient temperature control and energy consumption management are achieved.
Patent Information
- Application Number
- CN202511340781.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-19
- Publication Date
- 2025-11-21
AI Technical Summary
Traditional adaptive control methods cannot effectively address the issues of temperature stability and energy consumption optimization in intelligent refrigerators under complex operating conditions, especially under sudden changes in dynamic operating conditions and system uncertainty disturbances, which leads to a decline in controller performance.
A robust optimization-based multi-model adaptive control method is adopted. By constructing several sub-models and prediction error indices, and combining them with robust optimization technology, the real-time monitoring and dynamic estimation of refrigerator operating parameters are realized, and the execution commands are dynamically adjusted.
It achieves temperature stability and energy consumption optimization for refrigerators under complex operating conditions, improves control accuracy and speed, ensures temperature fluctuations within ±0.3℃, and reduces energy consumption by more than 18%.
Smart Images
Figure CN120993747A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of intelligent household appliance control, and in particular to a multi-model adaptive control method based on robust optimization. BACKGROUND
[0002] The intelligent refrigerator is a type of refrigerator that can be intelligently controlled, and is mainly applied to the fields of food storage and smart home. The intelligent refrigerator maintains the best storage state of food materials through a digital temperature control system, supports food preservation monitoring, shelf life reminders, and diet matching suggestions.
[0003] In the field of intelligent home device control, the design of traditional adaptive control methods usually relies on system models with fixed or slowly time-varying parameters, and the core assumption is that the operating environment does not change or changes slowly. However, as a commonly used household appliance, the intelligent refrigerator faces extremely complex working conditions in actual operation: 1. There are dynamic working condition mutations, such as high-frequency opening of the door by the user to take and put food materials (4-8 times in 1 hour), a large amount of food materials being stored at one time (the weight of food materials in the refrigerator compartment increases from 3 kg to 8 kg), and drastic fluctuations in environmental temperature and humidity (the temperature outside the box increases from 25°C to 35°C); 2. There are system uncertainty disturbances, such as sensor drift (the infrared weighing sensor shows that the weight does not change continuously for 12 hours, but the door body is frequently opened and closed), actuator parameter fluctuation (compressor speed deviation ±100 r / min), and external strong interference (photovoltaic power fluctuation ±50 W). When the above complex situations occur, the identifier of the traditional adaptive control is difficult to quickly track the mutation of the system parameters, and is prone to the problem of "model mismatch" - for example, the target temperature of the refrigerator compartment is 2°C, but the actual temperature is continuously 3°C and the compressor is already running at full load, which ultimately leads to a decline in the performance of the controller and the inability to meet the core requirements of the refrigerator for temperature stability and energy optimization. SUMMARY
[0004] Therefore, the embodiments of the present application provide a multi-model adaptive control method based on robust optimization to solve the problem that the performance requirements of the intelligent refrigerator for temperature stability and energy optimization cannot be met by using the traditional adaptive control method in the prior art.
[0005] The embodiments of the present application provide a multi-model adaptive control method based on robust optimization, which comprises:
[0006] Obtaining a plurality of working condition sub-model prediction errors according to refrigerator operating parameters;
[0007] Determining a prediction error index corresponding to each sub-model according to the plurality of sub-model prediction errors, respectively;
[0008] Obtaining a controller weight of each sub-model according to a preset normalization coefficient and each prediction error index;
[0009] The outputs of all sub-model controllers and their corresponding weights are fused together to obtain the execution instructions.
[0010] Optionally, the construction of several sub-models includes:
[0011] The prediction function for the low-load stable operating condition sub-model is set based on the refrigerator's refrigeration temperature, compressor speed, and ambient temperature disturbances:
[0012]
[0013] Control law for the sub-model under low-load stable operating conditions:
[0014] Compressor speed f1 = 2000 - 50 × (y 目标温度 -y 实际温度 When y 目标温度 -y 实际温度 At ≤0.3℃, the compressor speed is maintained at 2000 r / min; the fan air volume f2 = 3 m³ / min. 3 / h; damper opening f3 = 40;
[0015] The low-load stable operating condition sub-model is applicable to the following scenarios: nighttime, no door open, and food weight within the first preset weight.
[0016] The prediction function for the load fluctuation condition sub-model based on the refrigerator's refrigeration temperature, compressor speed, and ambient temperature disturbance settings is as follows:
[0017]
[0018] Control law for the sub-model of medium load fluctuation condition:
[0019] Compressor speed f1 = 2200 - 80 × (y 目标温度 -y 实际温度 ); Fan air volume f2 = 4m³ 3 / h; damper opening f3 = 50;
[0020] The medium load fluctuation sub-model is applicable to the following scenarios: daytime, low door opening frequency, and food weight within the second preset weight; the lower limit of the second preset weight is the upper limit of the first preset weight.
[0021] The prediction function for the high-load fluctuation condition sub-model is set based on the refrigerator's refrigeration temperature, compressor speed, and ambient temperature disturbances:
[0022]
[0023] Control law for sub-model under high load fluctuation conditions:
[0024] Compressor speed f1 = 2800 - 100 × (y 目标温度 -y实际温度 ), maximum rotation speed f 1max = 3500 r / min; fan air volume f2 = 6 m 3 / h, immediately increased to 7 m 3 / h after the door is opened; damper opening f3 = 70;
[0025] The high-load fluctuation working condition sub-model is applicable to the following scenarios: cooking period, high door opening frequency, and food material weight greater than or equal to the upper limit of the second preset weight.
[0026] wherein, represents the current predicted refrigeration temperature of the refrigerator, i = 1, 2, 3; y(t-1) represents the refrigeration temperature collected at the previous moment of the refrigerator, R 当前 represents the current compressor rotation speed, unit: r / min, R max , R min respectively represent the maximum rotation speed and the minimum rotation speed of the compressor, T max , T min respectively represent the minimum refrigeration effect temperature and the maximum refrigeration effect temperature corresponding to the minimum rotation speed and the maximum rotation speed of the compressor; d(t) represents the environmental temperature disturbance.
[0027] Optionally, the construction of the plurality of sub-models further comprises:
[0028] Based on the compressor power consumption, the fan power consumption, and the photovoltaic power, a photovoltaic power supply working condition sub-model prediction function is set:
[0029]
[0030] Photovoltaic power supply working condition sub-model control law:
[0031] Compressor rotation speed f1 = 2500-30xP, P is the photovoltaic power, unit: kW, maximum rotation speed f 1max = 3500 r / min; fan air volume f2 = 5 m 3 / h; when the photovoltaic power is greater than or equal to 0.2 kW, the defrosting period = original period-0.5xP;
[0032] wherein, represents the current predicted power consumption of the refrigerator, u1(t) represents the current compressor power consumption of the refrigerator, u2(t) represents the current fan power consumption of the refrigerator, P(t) represents the current photovoltaic output power, unit: kW, and the unit of the defrosting period is hour; the photovoltaic power supply working condition sub-model is applicable to the following scenarios: user installation of photovoltaic and photovoltaic output power greater than or equal to a preset power.
[0033] Optionally, the construction of the plurality of sub-models further comprises:
[0034] The defrosting transition condition sub-model prediction function is set based on the refrigerator refrigeration temperature, defrosting heater power and damper closing degree:
[0035]
[0036] The defrosting transition condition sub-model control law is:
[0037] The defrosting heater power f4 is 800 W for the first 10 minutes and 500 W for the last 10 minutes; the damper opening degree f3 = 0; after the defrosting is completed, the compressor speed is immediately increased to 3200 r / min and is returned to stable within 10 minutes;
[0038] Wherein, u3(t) represents the defrosting heater power, with the unit of W, and u4(t) represents the damper closing degree, with the value range of 0-100%; the applicable scene of the defrosting transition condition sub-model is that after the defrosting is started, within the preset defrosting time range, the evaporator frosting thickness is greater than or equal to the preset thickness.
[0039] Optionally, the prediction error index corresponding to each sub-model is determined according to the prediction error of each sub-model, and the calculation method comprises:
[0040] The parameters of different categories are normalized to eliminate the dimension effect;
[0041] In the low load stable condition sub-model, the medium load fluctuation condition sub-model and the high load fluctuation condition sub-model, the difference between the current refrigeration temperature and the current predicted refrigeration temperature of the refrigerator is calculated to obtain a temperature error;
[0042] The first temperature error square integral within the first preset time is calculated to obtain the first prediction error index, the second prediction error index and the third prediction error index, and the calculation method is as follows:
[0043]
[0044] In the photovoltaic power supply condition sub-model, the difference between the current energy consumption of the refrigerator and the current predicted energy consumption of the refrigerator is calculated to obtain an energy consumption error;
[0045] The energy consumption error square integral within the second preset time is calculated to obtain the fourth prediction error index, and the calculation method is as follows:
[0046]
[0047] In the defrosting transition condition sub-model, the difference between the current refrigeration temperature of the refrigerator and the current predicted refrigeration temperature of the refrigerator is calculated to obtain a temperature error;
[0048] The temperature error square integral within the third preset time is calculated to obtain the fifth prediction error index, and the calculation method is as follows:
[0049]
[0050] Optionally, the weight of each sub-model controller is obtained according to the preset normalization coefficient and each prediction error index, and the calculation method comprises:
[0051]
[0052] wherein λ is the preset normalization coefficient.
[0053] Optionally, the parameter indexes of different categories are normalized to eliminate the dimensional influence in the step, and the normalization method comprises:
[0054]
[0055] wherein ε j (t) represents the parameter index, is the regression vector corresponding to the parameter index ε j (t). Optionally, the prediction error of the working condition sub-model is obtained according to the running parameters of the refrigerator, and further comprises:
[0056] The dead zone processing is performed on the prediction error of the working condition sub-model, and the calculation method is as follows:
[0057]
[0058] wherein Δ is the dead zone threshold.
[0059] Optionally, after the fusion output of all sub-model controller outputs and corresponding sub-model controller weights is obtained, the execution instruction is further obtained, and the method further comprises:
[0060] The running parameters of the execution instruction are modified by forced parameter convergence and / or limiting the physical boundary of the parameter.
[0061] Optionally, the method further comprises:
[0062] The deviation between the actual refrigeration temperature of the refrigerator and the predicted refrigeration temperature of the refrigerator is obtained at a first preset frequency, and if the deviation value exceeds the preset deviation value for three times in succession, the model parameter is modified;
[0063] In the scenario of no door opening, if the infrared weighing sensor displays weight fluctuation at a preset time point, the automatic calibration is triggered to the average weight in the previous one hour;
[0064] The temperature and humidity sensor is calibrated at a second preset frequency.
[0065] The beneficial effects of the present application are:
[0066] The embodiment provides a multi-model adaptive control method based on robust optimization, takes a perception-decision-execution closed loop control as core logic, and realizes refrigerator adaptive control by combining an MMAC system and a robust optimization technique through three-level processes of online monitoring (perception), dynamic estimation (decision) and real-time adjustment (execution), uses a multi-model to approximate dynamic performance of a system, and designs a multi-model adaptive controller based on the multi-model, so that good control precision, speed and stability are achieved for a complex system. BRIEF DESCRIPTION OF DRAWINGS
[0067] The features and advantages of the present application will be more clearly understood through the following detailed description taken in conjunction with the accompanying drawings, which are given by way of illustration and are not to be considered limiting of the present application, in which:
[0068] Fig. 1 A flow chart of a multi-model adaptive control method based on robust optimization in the embodiment of the present application is shown;
[0069] Fig. 2 A whole block diagram of a multi-model adaptive control system based on robust optimization in the embodiment of the present application is shown;
[0070] Fig. 3 A principle diagram of a multi-model adaptive control system based on robust optimization in the embodiment of the present application is shown. DETAILED DESCRIPTION
[0071] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.
[0072] As shown in the figure, Figs. 1 to 3 The embodiment of the present application provides a multi-model adaptive control method based on robust optimization, which comprises the following steps:
[0073] In step S10, a plurality of working condition sub-model prediction errors are acquired according to refrigerator operation parameters.
[0074] In the embodiment, the refrigerator operation parameters comprise user frequency of opening the refrigerator door, food weight, temperature and humidity, and actuator state parameters. The actuator comprises a compressor, a fan, a damper opening, a defrosting heater and a feedback actuator.
[0075] In the specific embodiment, the acquisition of the operation parameters comprises:
[0076] Door switch sensor: Real-time acquisition of door opening and closing state, such as door opening time 0.5s, 3 times of opening door within 1 hour, to judge the frequency of food taking and placing.
[0077] Infrared weighing sensor: installed in each storage layer of the refrigerator, monitoring the weight change of food, such as the weight of food in the refrigerator compartment decreasing from 5kg to 3.2kg, to infer the food load.
[0078] Temperature and humidity sensor: respectively collecting the temperature and humidity of the refrigerator compartment (target 2-8℃), the freezer compartment (target -18℃±2℃), and the environment outside the box (such as 30℃ / 60%RH outside the box, and the current 5℃ / 55%RH in the refrigerator compartment).
[0079] Photovoltaic power sensor (if the user accesses photovoltaic): Real-time acquisition of photovoltaic power supply power, such as current power supply 200W, power fluctuation ±50W, to judge the priority of grid interaction.
[0080] Actuator state sensor: monitoring the running speed of the compressor (such as 3000r / min), the air volume of the fan (such as 5m 3 / h), the damper opening (such as 60%), and the defrost heater temperature (such as 50℃), to feedback the actuator working state.
[0081] Step S20: Determine the prediction error index corresponding to each sub-model according to the prediction error of each sub-model.
[0082] In this embodiment, multiple sub-models are preset to apply to different use scenarios. When determining which specific sub-model the refrigerator currently corresponds to, multiple types of operating parameters need to be comprehensively considered for auxiliary judgment, and then the prediction error is converted into a prediction error index.
[0083] Step S30: Obtain the weight of each sub-model controller according to the preset normalization coefficient and each prediction error index.
[0084] In this embodiment, the normalization coefficient is set to ensure stable weight calculation and avoid sudden increase or decrease of the weight of a certain model.
[0085] Step S40: Fuse the outputs of all sub-model controllers and the corresponding sub-model controller weights to obtain the execution instruction.
[0086] In this embodiment, each sub-model controller corresponds to a set of control law. The sub-model controller weights obtained in step S30 are used to weight and sum all sub-model controller corresponding control laws to obtain the final instruction of the actuator.
[0087] The embodiment provides a multi-model adaptive control method based on robust optimization, takes a perception-decision-execution closed loop control as core logic, realizes refrigerator adaptive control by combining an MMAC system and a robust optimization technology through three-level processes of online monitoring (perception), dynamic estimation (decision) and real-time adjustment (execution).
[0088] Based on multi-dimensional data of online monitoring, two core tasks of load state determination and model / sensor abnormality compensation are completed through a "food material-environment dynamic load model calculation unit", so as to provide a decision basis for subsequent control adjustment. For example, according to coupled data of "food material weight + door opening frequency + environment temperature and humidity", a load model is updated in real time, and a refrigeration intensity requirement is output.
[0089] Based on the above model design idea, construction of a plurality of sub-models provided by the embodiment includes:
[0090] A low-load stable working condition sub-model M1 prediction function is set based on refrigerator refrigeration temperature, compressor speed and environment temperature disturbance:
[0091]
[0092] The low-load stable working condition sub-model is applicable to the following scenarios: night (22:00-6:00), no door opening and food material weight less than 3 kg.
[0093] A low-load stable working condition sub-model control law is:
[0094] Compressor speed f1 = 2000-50x(y 目标温度 -y 实际温度 ), when y 目标温度 -y 实际温度 ≤0.3℃, the compressor speed is maintained at 2000r / min; fan air volume f2 = 3m 3 / h; damper opening degree f3 = 40.
[0095] A medium-load fluctuation working condition sub-model M2 prediction function is set based on refrigerator refrigeration temperature, compressor speed and environment temperature disturbance:
[0096]
[0097] The medium-load fluctuation working condition sub-model is applicable to the following scenarios: day (8:00-18:00), low door opening frequency (1-3 times / hour) and medium food material weight (3-8 kg).
[0098] A medium-load fluctuation working condition sub-model control law is:
[0099] Compressor speed f1 = 2200-80x(y 目标温度 -y 实际温度); fan air volume f2 = 4 m 3 / h; damper opening f3 = 50.
[0100] High-load fluctuation working condition sub-model M3 prediction function is set based on refrigerator cold storage temperature, compressor speed and environmental temperature disturbance:
[0101]
[0102] High-load fluctuation working condition sub-model applicable scenarios are: cooking period (11:00-13:00 / 17:00-19:00), high door opening frequency (4-8 times / hour) and large amount of food (more than 8 kg).
[0103] High-load fluctuation working condition sub-model control law:
[0104] Compressor speed f1 = 2800-100×(y 目标温度 -y 实际温度 ), maximum speed f 1max = 3500 r / min; fan air volume f2 = 6 m 3 / h, immediately increased to 7 m 3 / h after opening; damper opening f3 = 70.
[0105] Wherein, represents the current predicted cold storage temperature of the refrigerator, i = 1, 2, 3; y(t-1) represents the cold storage temperature collected at the previous moment of the refrigerator, R 当前 represents the current compressor speed, unit: r / min, R max , R min respectively represent the maximum speed and minimum speed of the compressor, T max , T min respectively represent the minimum refrigeration effect temperature and the maximum refrigeration effect temperature corresponding to the minimum speed and the maximum speed of the compressor; d(t) represents the environmental temperature disturbance.
[0106] As an optional implementation, the construction of several sub-models also includes:
[0107] Photovoltaic power supply working condition sub-model M4 prediction function is set based on compressor power consumption, fan power consumption and photovoltaic power:
[0108]
[0109] Photovoltaic power supply working condition sub-model control law:
[0110] Compressor speed f1 = 2500-30×P, P is photovoltaic power, unit: kW, maximum speed f 1max = 3500 r / min; fan air volume f2 = 5 m 3When photovoltaic power is greater than or equal to 0.2 kW, defrosting period = original period - 0.5 x P;
[0111] wherein, represents the current predicted power consumption of the refrigerator, u1(t) represents the current compressor power consumption of the refrigerator, u2(t) represents the current fan power consumption of the refrigerator, P(t) represents the current photovoltaic output power, and the units are all kW, and the unit of the defrosting period is hour; the applicable scene of the photovoltaic power supply working condition submodel is that the user installs photovoltaic and the photovoltaic output power is greater than or equal to 100 W.
[0112] As an optional implementation, the construction of the several submodels further includes:
[0113] The defrosting transition working condition submodel M5 prediction function is set based on the refrigeration temperature of the refrigerator, the defrosting heater power and the damper closing degree:
[0114]
[0115] The defrosting transition working condition submodel control law is:
[0116] The defrosting heater power f4 is 800 W for the first 10 minutes and 500 W for the last 10 minutes; the damper opening degree f3 = 0; after the defrosting is completed, the compressor speed is immediately increased to 3200 r / min, and the speed is returned to stable within 10 minutes;
[0117] wherein, u3(t) represents the defrosting heater power, the unit is W, and u4(t) represents the damper closing degree, the value range is 0-100%; the applicable scene of the defrosting transition working condition submodel is that after the defrosting is started, within the preset defrosting time range, the thickness of the evaporator frost is greater than or equal to the preset thickness.
[0118] In the embodiment, for each submodel, a special subcontroller is designed offline, the “control target + actuator strategy + control law” is clear, and it is ensured that each controller meets the target of “temperature fluctuation ≤±0.3 ℃, energy consumption reduction ≥18%” in the corresponding working condition.
[0119] As an optional implementation, the prediction error index corresponding to each submodel is determined according to the prediction error of the several submodels, and the calculation method includes:
[0120] The different types of parameter indexes are normalized to eliminate the dimension effect;
[0121] In the low-load stable working condition submodel, the medium-load fluctuation working condition submodel and the high-load fluctuation working condition submodel, the difference between the current refrigeration temperature and the current predicted refrigeration temperature of the refrigerator is calculated to obtain the temperature error;
[0122] The first temperature error square integral in the first preset time is calculated to obtain the first prediction error index, the second prediction error index and the third prediction error index, and the calculation method is as follows:
[0123]
[0124] Wherein, i = 1, 2, 3; Δt1 is 10 seconds.
[0125] In the photovoltaic power supply working condition submodel, the difference between the current energy consumption of the refrigerator and the current predicted energy consumption of the refrigerator is calculated to obtain the energy consumption error;
[0126] The energy consumption error square integral in the second preset time is calculated to obtain the fourth prediction error index, and the calculation method is as follows:
[0127] Δt2 is 10 seconds.
[0128] In the defrosting transition working condition submodel, the difference between the current refrigeration temperature of the refrigerator and the current predicted refrigeration temperature of the refrigerator is calculated to obtain the temperature error;
[0129] The temperature error square integral in the third preset time is calculated to obtain the fifth prediction error index, and the calculation method is as follows:
[0130] Δt3 is 5 seconds.
[0131] As an optional embodiment, the weights of the submodel controllers are obtained according to the preset normalization coefficient and the respective prediction error indexes, and the calculation method comprises:
[0132]
[0133] Wherein, λ is the preset normalization coefficient. In specific embodiments, λ = 0.1.
[0134] As an optional embodiment, the parameters of different categories are normalized to eliminate the dimensional influence in the step, and the normalization method comprises:
[0135]
[0136] Wherein, ε j (t) represents the parameter index, is the regression vector corresponding to the parameter index ε j (t). For example, the regression vector of model M2 is
[0137] In online operation, the actual output of the refrigerator, such as the refrigeration temperature y(t) and the energy consumption E(t), is collected every 500 ms, and the predicted output of each submodel The matching weight ωi (t), for example: if it is daytime, no door opening, 5 kg of food, and an environment of 25℃, J2(t) of M2 = 0.02, J1(t) of M1 = 0.1, then ω2(t) ≈ 0.85, ω1(t) ≈ 0.1, and other model weights ≈ 0.05, indicating that the current working condition is most matched with M2.
[0138] According to the output f i (t) of each sub-controller and the weight ω i (t), the final control amount is calculated according to the formula and is landed to the actuator:
[0139] For example: the current working condition is during cooking (12:00), 3 door openings per hour, 7 kg of food, and an environment of 30℃, the model weights are ω2 = 0.3, ω3 = 0.6, ω1 = 0.1, and ω4 = ω5 = 0, then:
[0140] The final command of the compressor speed: f1 = 0.3 × 2200 + 0.6 × 2800 + 0.1 × 2000 = 2620 r / min;
[0141] The final command of the fan air volume: f2 = 0.3 × 4 + 0.6 × 6 + 0.1 × 3 = 5.1 m 3 / h (rounded to 5 m^3 / h);
[0142] The final command of the damper opening: f3 = 0.3 × 50 + 0.6 × 70 + 0.1 × 40 = 61 (rounded to 60%);
[0143] Through millisecond-level operation (≤50 ms), the command is issued to the actuator to realize collaborative regulation.
[0144] In order to avoid the problems of false switching and response lag, the stability and convergence speed of the switching strategy are optimized:
[0145] The switching lag threshold is set: if the current model weight ω i (t) ≥ 0.6 and lasts for 5 seconds, switching to the controller corresponding to the model is allowed, so as to avoid false switching caused by temporary increase of the weight.
[0146] For high-load working conditions (such as M3), the weight calculation period is shortened from 500 ms to 300 ms, so that when the temperature rises sharply after the door is opened, the model can quickly converge to M3, and the compressor, fan, and temperature can respond in time, and the temperature recovery time is ≤5 minutes.
[0147] As an optional implementation, the prediction error of the plurality of working condition sub-models is obtained according to the refrigerator operating parameters, and further comprising:
[0148] The dead zone processing is performed on the prediction error of the plurality of working condition sub-models, and the calculation method is as follows:
[0149]
[0150] wherein, Delta is a dead zone threshold.
[0151] As an optional implementation, after the fusion output of all sub-model controller outputs and corresponding sub-model controller weights to obtain the execution instruction, further comprising:
[0152] The running parameters of the execution instruction are modified by forcing parameter convergence and / or limiting parameter physical boundaries.
[0153] In this embodiment, the parameter vector theta (t) is set, such as the food material heat dissipation coefficient in the adaptive model, the compressor efficiency coefficient:
[0154] The sigma correction coefficient sigma is set to 0.01, and the parameter update law is: Wherein, Gamma = 0.5 is an adaptive gain, which ensures that the parameter adjustment speed is moderate and will not be too slow to cause temperature out of control.
[0155] The convex set Omega is defined: based on the physical meaning of the refrigerator parameters, such as the compressor speed theta1: 1500 r / min <= theta1 <= 3500 r / min, the food material heat dissipation coefficient theta2: 0.1 W / (m·℃) <= theta2 <= 0.5 W / (m·℃);
[0156] Projection rule: if the parameter is updated theta (t) is in int (Omega), such as theta1 = 2500 r / min, then the parameter is retained; if Such as theta1 = 3600 r / min, which exceeds the maximum speed, the parameter is projected to the boundary and corrected to 3500 r / min.
[0157] The running parameters of the execution instruction are modified by forcing parameter convergence, even if the system is not continuously excited under special working conditions, such as the opening frequency of the door after the party is over, the parameter such as the heat dissipation coefficient can be forced to converge exponentially to a reasonable range, avoiding parameter drift.
[0158] The running parameters of the execution instruction are modified by limiting the physical boundaries of the parameters, ensuring that the parameters are always within the physically feasible range, avoiding unreasonable situations such as compressor speed exceeding the upper limit and heat dissipation coefficient being negative.
[0159] As an optional implementation, further comprising:
[0160] The deviation of the actual refrigerator refrigeration temperature and the refrigerator predicted refrigeration temperature is obtained every 10 minutes, and if the deviation value exceeds 0.5℃ for 3 consecutive times, the model parameter is modified, for example, the food material heat dissipation coefficient is adjusted from 0.2 to 0.25.
[0161] Every Sunday at 2:00 a.m., based on the logic of no change in the weight of the food materials during the non-opening time, if the infrared weighing sensor shows that the weight fluctuation is greater than or equal to 0.1 kg, it is automatically calibrated to the average weight of the previous 1 hour.
[0162] Once a month, based on the logic of uniform internal temperature after the refrigerator is stopped (maintenance mode), if the temperature difference of the temperature and humidity sensors at different positions is greater than or equal to 0.3°C, the middle position sensor is used for calibration.
[0163] Although the embodiments of the present application are described in conjunction with the drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.
Claims
1. A multi-model adaptive control method based on robust optimization, characterized in that, include: Based on the refrigerator's operating parameters, several operating condition sub-model prediction errors are obtained; Based on the prediction errors of the aforementioned sub-models, the prediction error index corresponding to each sub-model is determined; The weights of each sub-model controller are obtained based on the preset normalization coefficients and the prediction error indices. The outputs of all sub-model controllers and their corresponding weights are fused together to obtain the execution instructions.
2. The multi-model adaptive control method based on robust optimization according to claim 1, characterized in that, The construction of several of the sub-models includes: The prediction function for the low-load stable operating condition sub-model is set based on the refrigerator's refrigeration temperature, compressor speed, and ambient temperature disturbances: Control law for the sub-model under low-load stable operating conditions: Compressor speed f1 = 2000 - 50 × (y 目标温度 -y 实际温度 When y 目标温度 -y 实际温度 At ≤0.3℃, the compressor speed is maintained at 2000 r / min; the fan air volume f2 = 3 m³ / min. 3 / h; damper opening f3 = 40; The low-load stable operating condition sub-model is applicable to the following scenarios: nighttime, no door open, and food weight within the first preset weight. The prediction function for the load fluctuation condition sub-model based on the refrigerator's refrigeration temperature, compressor speed, and ambient temperature disturbance settings is as follows: Control law for the sub-model of medium load fluctuation condition: Compressor speed f1 = 2200 - 80 × (y 目标温度 -y 实际温度 ); Fan air volume f2 = 4m³ 3 / h; damper opening f3 = 50; The medium-load fluctuation sub-model is applicable to the following scenarios: daytime, low door opening frequency, and food weight within the second preset weight; the lower limit of the second preset weight is the upper limit of the first preset weight. The prediction function for the high-load fluctuation condition sub-model is set based on the refrigerator's refrigeration temperature, compressor speed, and ambient temperature disturbances: Control law for sub-model under high load fluctuation conditions: Compressor speed f1 = 2800 - 100 × (y 目标温度 -y 实际温度 Maximum speed f 1max =3500r / min; Fan air volume f2 = 6m³ 3 / h, immediately rises to 7m after the door is opened 3 / h; Damper opening f3 = 70; The high-load fluctuation sub-model is applicable to the following scenarios: cooking time, high door opening frequency, and food weight greater than or equal to the upper limit of the second preset weight. in, This represents the current predicted refrigeration temperature of the refrigerator, where i = 1, 2, 3; y(t-1) represents the refrigeration temperature collected by the refrigerator at the moment before. R 当前 This indicates the current compressor speed, in r / min. max R min T represents the compressor's maximum and minimum speeds, respectively. max T min d(t) represents the minimum and maximum cooling effect temperatures corresponding to the compressor's minimum and maximum speeds, respectively; d(t) represents the ambient temperature disturbance.
3. The multi-model adaptive control method based on robust optimization according to claim 2, characterized in that, The construction of several of the sub-models also includes: The prediction function for the photovoltaic power supply sub-model is set based on compressor power consumption, fan power consumption, and photovoltaic power: Control law for photovoltaic power supply sub-model: Compressor speed f1 = 2500 - 30 × P, where P is the photovoltaic power in kW, and f is the maximum speed. 1max =3500r / min; Fan air volume f2 = 5m³ 3 / h; When the photovoltaic power is ≥0.2kW, the defrosting cycle = original cycle - 0.5×P; in, The current predicted power consumption of the refrigerator is represented by u1(t), the current power consumption of the refrigerator compressor is represented by u2(t), the current power consumption of the refrigerator fan is represented by P(t), and the current photovoltaic output power is represented by kW. The defrosting cycle is in hours. The photovoltaic power supply sub-model is applicable to the following scenarios: users install photovoltaics and the photovoltaic output power is greater than or equal to the preset power.
4. The multi-model adaptive control method based on robust optimization according to claim 3, characterized in that, The construction of several of the sub-models also includes: The prediction function for the defrosting transition condition sub-model is set based on the refrigerator's cooling temperature, defrosting heater power, and damper closure degree: Control law for the defrosting transition condition sub-model: Defrosting heater power f4: 800W for the first 10 minutes, 500W for the next 10 minutes; damper opening f3 = 0; after defrosting, the compressor speed immediately rises to 3200r / min and stabilizes within 10 minutes. Where u3(t) represents the power of the defrost heater in W, u4(t) represents the damper closure degree, with a value range of 0 to 100%; the applicable scenario for the defrost transition condition sub-model is: after defrosting is started, within the preset defrost time range, the frost thickness on the evaporator is greater than or equal to the preset thickness.
5. The multi-model adaptive control method based on robust optimization according to claim 4, characterized in that, The prediction error index for each sub-model is determined based on the prediction errors of the aforementioned sub-models, and the calculation method includes: Normalize the parameters of different categories to eliminate the influence of dimensions; In the low-load stable operating condition sub-model, the medium-load fluctuating operating condition sub-model, and the high-load fluctuating operating condition sub-model, the difference between the current refrigeration temperature of the refrigerator and the current predicted refrigeration temperature of the refrigerator is calculated to obtain the temperature error; Calculate the square integral of the first temperature error over the first preset time period to obtain the first prediction error index, the second prediction error index, and the third prediction error index. The calculation method is as follows: In the photovoltaic power supply sub-model, the difference between the current energy consumption of the refrigerator and the current predicted energy consumption of the refrigerator is calculated to obtain the energy consumption error; The fourth prediction error index is obtained by calculating the square integral of the energy consumption error within the second preset time period. The calculation method is as follows: In the defrost transition sub-model, the difference between the current refrigerator temperature and the current predicted refrigerator temperature is calculated to obtain the temperature error; The fifth prediction error index is obtained by calculating the square integral of the temperature error over a third preset time period. The calculation method is as follows:
6. The multi-model adaptive control method based on robust optimization according to claim 5, characterized in that, The weights of each sub-model controller are obtained based on preset normalization coefficients and various prediction error indices. The calculation method includes: Wherein, λ is the preset normalization coefficient.
7. The multi-model adaptive control method based on robust optimization according to claim 5, characterized in that, In the step of normalizing different categories of parameter indicators to eliminate the influence of dimensions, the normalization methods include: Where, ε j (t) represents the parameter index, To be related to the parameter index ε j The regression vector corresponding to (t).
8. The multi-model adaptive control method based on robust optimization according to claim 7, characterized in that, Based on the refrigerator's operating parameters, several operating condition sub-model prediction errors are obtained, including: Dead zone processing is applied to the prediction errors of several of the aforementioned sub-models for different operating conditions. The calculation method is as follows: Where Δ is the dead zone threshold.
9. The multi-model adaptive control method based on robust optimization according to claim 1, characterized in that, After fusing the outputs of all sub-model controllers and their corresponding weights to obtain the execution instruction, the method further includes: The execution parameters of the instruction are modified by forcing parameter convergence and / or limiting the physical boundaries of the parameters.
10. The multi-model adaptive control method based on robust optimization according to claim 1, characterized in that, Also includes: The deviation between the actual refrigeration temperature and the predicted refrigeration temperature of the refrigerator is obtained at a first preset frequency. If the deviation value exceeds the preset deviation value for three consecutive times, the model parameters are corrected. In scenarios where the door is not open, if the infrared weighing sensor shows weight fluctuations at a preset time point, it will trigger automatic calibration to the average weight of the previous hour. The temperature and humidity sensor is calibrated at a second preset frequency.