Refrigerator defrosting control method based on coupling of BP neural network and fuzzy control

Abstract

The application discloses a refrigerator defrosting control method based on coupling of BP neural network and fuzzy control, and is characterized in that: data-driven modeling is used to realize deep perception of dynamic characteristics of a refrigeration system; a BP neural network is used to learn actual data signals of a compressor and an evaporator, and to indirectly estimate hidden parameters such as refrigerant charging quantity and frost layer shape distribution; a fuzzy controller is used to dynamically adjust a defrosting strategy according to a neural network prediction result, and to realize collaborative optimization of refrigeration cycle parameters and a frosting process. The neural network is used to learn environmental parameters on line, and to optimize a fuzzy rule base in real time; the neural network is used to directly map bottom-layer physical characteristics such as a compressor operation mode and evaporator structure parameters, and to output a dynamic compensation signal; and the model is continuously iterated by accumulating historical data, and is adapted to regional / seasonal differences.

Description

Technical Field

[0001] This invention relates to intelligent control technology for refrigerators, and in particular to a refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control. Background Technology

[0002] When household refrigerators perform food preservation and refrigeration / freezing functions, evaporator frost formation is always a key factor affecting performance. Due to the high humidity inside the refrigerator and the evaporator surface temperature being below the dew point, water vapor condenses on the evaporator surface and forms a frost layer with poor thermal conductivity over time. This not only reduces heat exchange efficiency, affecting food preservation, but also causes wear and tear on evaporator components, increased energy consumption, and shortened equipment lifespan due to long-term frost formation.

[0003] Existing intelligent defrosting technology utilizes fuzzy control to achieve adaptive adjustment of complex, nonlinear, and multivariable systems. It can dynamically optimize defrosting strategies based on parameters such as temperature and humidity, offering greater precision compared to traditional timed control. For example, patent publication number CN115371338B describes a refrigerator defrosting control method. First, it calculates and corrects the initial time step for gradually reducing power based on parameters such as compressor running time, initial defrosting power, and ambient temperature. Then, it gradually reduces the defrosting power, controlling the time step by setting the temperature rise amplitude of the defrosting sensor. Finally, based on the running time and temperature rise rate within the previous time step, it continuously adjusts the power and temperature rise amplitude within subsequent time steps, continuously reducing the electric heater power for defrosting until the set conditions are met and the defrosting mode is exited. Another example is patent publication number CN105004127B, which proposes a refrigerator defrosting control method that uses temperature and humidity sensors to detect the temperature and humidity around the evaporator. These sensors transmit the detected temperature and humidity signals to a control board, which compares the received signals with preset values ​​to control the heating element and compressor's on / off state.

[0004] The limitation of these fuzzy controls in refrigerator defrosting control lies in their ability to passively respond to surface parameters of evaporator frost formation, such as frost thickness and temperature changes, while lacking direct control over the underlying dynamic characteristics of the refrigeration system.

[0005] These influencing factors include: refrigeration cycle parameters (refrigerant pressure, flow rate, and throttling device status), compressor operating characteristics (speed, efficiency, and start-stop frequency), evaporator physical structure (fin spacing, material thermal conductivity), and system ventilation conditions (fan speed, duct resistance). Traditional fuzzy control relies on a preset rule base, making it difficult to capture the heat and mass transfer coupling relationship between refrigerant phase change and frosting processes, and also unable to compensate in real time for compressor performance degradation or fin spacing changes caused by evaporator frosting. Therefore, it is necessary to introduce control methods that can overcome the limitations of empirical rules in single fuzzy control and significantly improve defrosting accuracy and system energy efficiency under complex operating conditions. Summary of the Invention

[0006] The purpose of this invention is to solve the above problems and provide a refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, which has the characteristics of dynamic adaptability, low-level regulation capability, multi-objective optimization, and data-driven characteristics.

[0007] The above-mentioned technical problems of the present invention are mainly solved by the following technical solution: a refrigerator defrosting control method based on BP neural network and fuzzy control coupling, characterized by: achieving deep perception of the dynamic characteristics of the refrigeration system through data-driven modeling: the BP neural network learns the actual data signals of the compressor and evaporator to indirectly estimate implicit parameters such as refrigerant charge and frost morphology distribution; the fuzzy controller dynamically adjusts the defrosting strategy according to the neural network prediction results to achieve synergistic optimization of refrigeration cycle parameters and the frosting process, including the following:

[0008] S1 - Data acquisition and preprocessing; S2 - Constructing a BP neural network prediction model for refrigerator defrosting control; S3 - BP neural network model training; S4 - BP neural network model performance evaluation; S5 - Constructing a coupled BP neural network and fuzzy control model; S6 - Adjusting the weights of the fuzzy control rule base; S7 - Dynamic adjustment of refrigerator defrosting; S8 - Constructing a comprehensive defrosting evaluation function.

[0009] In the aforementioned refrigerator defrosting control method based on BP neural network and fuzzy control coupling, as a preferred embodiment, S1-data acquisition and preprocessing: acquire refrigerator operation data, clean and normalize the data, and divide it into training set, validation set and test set;

[0010] Normalization scales data to a specified range: [0,1] or [-1,1]; the formula is:

[0011]

[0012] In the formula: x′ is the normalized data, x is the input feature; min(x) is the minimum value of the feature; max(x) is the maximum value of the feature;

[0013] The model is trained using the training set through iterative backpropagation and parameter update processes; the model's hyperparameters are tuned using the validation set; and the model's performance is evaluated using the test set.

[0014] In the aforementioned refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, the refrigerator operating data includes: evaporator temperature, ambient temperature and humidity, door opening and closing frequency, historical defrosting records, refrigeration cycle parameters, compressor operating characteristics, evaporator physical structure parameters, and system ventilation condition parameters.

[0015] In the aforementioned refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, as a preferred option, S-2, constructing a BP neural network prediction model for refrigerator defrosting control: the BP neural network consists of an input layer, a hidden layer and an output layer, that is, constructing a BP neural network model with evaporator temperature, environmental parameters and user behavior as inputs, and predicted frost thickness and defrosting time as outputs.

[0016] In the aforementioned refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, the BP neural network model in S-3 is trained using the ReLU activation function and the Adam optimizer. In the mathematical expression of the ReLU activation function, for an input value x, its output y can be represented by the following piecewise function:

[0017] y = ReLU(x) = max(0, x)

[0018] In the aforementioned refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, as a preferred embodiment, the performance evaluation of the S-4 and BP neural network models involves: adjusting the model hyperparameters using a grid search algorithm and evaluating the model using the mean squared error (MSE) model; where the mean squared error is the loss function used to measure the difference between the predicted output of the neural network and the true label, and is represented by σ.

[0019]

[0020] In the formula: n is the number of input layer nodes, x i This is the predicted output for the i-th sample. This is the average.

[0021] In the aforementioned refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, S-5, when constructing the coupling model of BP neural network and fuzzy control, a fuzzy control rule base is established, the neural network prediction results are converted into fuzzy linguistic variables, and the defrosting heating power and duration are determined through fuzzy reasoning.

[0022] In the aforementioned refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, as a preferred embodiment, S6-fuzzy control rule base weight adjustment, based on the actual defrosting needs and experience of the refrigerator, uses the output of the BP neural network prediction model as the input of the fuzzy control rule model, and dynamically adjusts the fuzzy rule weights through online learning to form a closed-loop optimization system; the coupled model refers to a collaborative decision-making system formed by the neural network prediction module and the fuzzy control module through data interaction.

[0023] In the aforementioned refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, as a preferred option, during the dynamic adjustment of refrigerator defrosting in S7, the cumulative defrosting data is statistically analyzed and compared with the state at adjacent time points, which is the key to the refrigerator defrosting system shifting from passive response to active optimization.

[0024] In the aforementioned refrigerator defrosting control method based on BP neural network and fuzzy control coupling, as a preferred embodiment, S8-constructs a comprehensive defrosting evaluation function, including constructing a comprehensive evaluation function including defrosting efficiency, energy consumption, and temperature fluctuation, inputting correction instructions to calculate the evaluation index at time t, and outputting a control signal if the index at time t is better than that at time t-1; otherwise, the model is retrained.

[0025] Construct a comprehensive evaluation function J:

[0026] J=αE+βΔT+γD

[0027] In the formula, E: defrosting energy consumption (Wh), calculated in real time by a current sensor; ΔT: temperature fluctuation range inside the refrigerator (°C) of the defrosting device; D: residual frost thickness (mm), obtained by weighing or laser ranging; α, β, γ: weighting coefficients;

[0028] Optimization objective: Minimize J, i.e.:

[0029] J * =min(αE+βΔT+γD)

[0030] This technical solution utilizes a coupling technique of BP neural network (backpropagation neural network) and fuzzy control to achieve deep perception of the dynamic characteristics of the refrigeration system through data-driven modeling: the BP neural network can learn signals such as compressor current and evaporator inlet and outlet temperature difference to indirectly estimate implicit parameters such as refrigerant charge and frost morphology distribution; the fuzzy controller dynamically adjusts the defrosting strategy based on the neural network prediction results, achieving coordinated optimization of refrigeration cycle parameters and the frosting process. This composite control architecture breaks through the empirical rule limitations of single fuzzy control, significantly improving defrosting accuracy and system energy efficiency under complex operating conditions.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] (1) Dynamic adaptability: The fuzzy rule base is optimized in real time by learning environmental parameters such as temperature and humidity, door opening frequency and frost rate through neural network online learning.

[0033] (2) Low-level control capability: The neural network directly maps the low-level physical characteristics such as the compressor operation mode and evaporator structural parameters, and outputs dynamic compensation signals, such as adjusting the opening degree of the throttling device.

[0034] (3) Multi-objective optimization: Simultaneously optimize multiple indicators such as defrosting efficiency, energy consumption reduction, and component life.

[0035] (4) Data-driven characteristics: The model is continuously iterated by accumulating historical data to adapt to regional / seasonal differences. Attached Figure Description

[0036] Figure 1 This is a general flowchart of the present invention.

[0037] Figure 2 This is a flowchart of the construction process of a BP neural network prediction model according to the present invention. Detailed Implementation

[0038] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0039] This embodiment presents a refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control. It achieves deep perception of the dynamic characteristics of the refrigeration system through data-driven modeling: the BP neural network learns the actual data signals of the compressor and evaporator to indirectly estimate implicit parameters such as refrigerant charge and frost morphology distribution; the fuzzy controller dynamically adjusts the defrosting strategy according to the neural network prediction results to achieve synergistic optimization of refrigeration cycle parameters and the frosting process.

[0040] Specifically, by acquiring parameters such as evaporator temperature, ambient temperature and humidity, door opening and closing frequency, historical defrosting records, refrigeration cycle parameters, compressor operating characteristics, evaporator physical structure parameters, and system ventilation conditions, a BP neural network prediction model for refrigerator defrosting control is constructed. The data is cleaned, preprocessed, and normalized, and the dataset is divided for training. Then, a grid search algorithm is used to adjust the model hyperparameters, and the mean square error (MSE) model is used to evaluate the performance of the BP neural network model. The predicted frost thickness and defrosting time are used as inputs to fuzzy control. The defrosting strategy—defrosting start time, defrosting heating power, and defrosting interval—is dynamically and proactively adjusted through fuzzy rules.

[0041] Includes the following, such as Figure 1 As shown:

[0042] S1. Data Acquisition and Preprocessing

[0043] Acquire refrigerator operating data, including evaporator temperature, ambient temperature and humidity, door opening and closing frequency, historical defrosting records, refrigeration cycle parameters (refrigerant pressure, flow rate, and throttling device status), compressor operating characteristics (speed, efficiency, and start / stop frequency), evaporator physical structure parameters (fin spacing, material thermal conductivity), and system ventilation condition parameters (fan speed, duct resistance), etc. Clean and normalize the data, and divide it into training set, validation set, and test set.

[0044] Normalization scales data to a specified range: [0,1] or [-1,1]; the formula is:

[0045]

[0046] In the formula: x′ is the normalized data, x is the input feature; min(x) is the minimum value of the feature; max(x) is the maximum value of the feature.

[0047] Dataset partitioning: Train the model using the training set, minimizing the loss function through iterative backpropagation and parameter update processes. Tune the model's hyperparameters using the validation set. Evaluate the model's performance using the test set.

[0048] S2. Construct a BP neural network prediction model for refrigerator defrosting control.

[0049] A BP neural network consists of an input layer, hidden layers, and an output layer. It constructs a BP neural network model that takes evaporator temperature, environmental parameters, and user behavior as inputs, and outputs predicted frost thickness and defrosting time. The steps for constructing a BP neural network are as follows:

[0050] (1) Network initialization: Determine the system input (X,Y), the number of nodes n, l, m in the input layer, hidden layer, and output layer, and initialize the connection weights ω between neurons in each layer. ij and ω jk Given the learning rate and neuron activation function, initialize the thresholds a and b of the hidden and output layers.

[0051] (2) Hidden layer output calculation:

[0052]

[0053] In the formula: X: input variable; ω ij : Connection weight; a: Threshold; H: Hidden layer output.

[0054] f: Hidden layer activation function, its expression is as follows:

[0055]

[0056] In the formula, e is the natural constant.

[0057] (3) Output layer calculation:

[0058]

[0059] Where: ω jk b: connection weight; 0: threshold; O: output of output layer.

[0060] S3. Train the collected data using a BP neural network model.

[0061] Training is performed using the ReLU activation function and the Adam optimizer. In the mathematical expression of the ReLU activation function, for an input value x, its output y can be represented by the following piecewise function:

[0062] y = ReLU(x) = max(0, x)

[0063] Adam optimization combines the ideas of momentum and adaptive learning rate. It simultaneously calculates the exponentially weighted average of the first moment (mean) and second moment (variance) of the gradient and uses this information to adjust the learning rate. It can adaptively adjust the learning rate for different parameters and performs well when dealing with sparse gradients. See also Figure 2 The diagram shows the process of building a prediction model using a BP neural network.

[0064] S4: Adjust model hyperparameters using a grid search algorithm and evaluate the performance of the BP neural network model using the mean squared error (MSE) model.

[0065] Mean squared error (MSE) is a loss function used to measure the difference between the predicted output of a neural network and the true label. MSE is generally represented by σ.

[0066]

[0067] In the formula: n: number of input layer nodes; x i The predicted output for the i-th sample; Average.

[0068] S5. Establish a fuzzy control rule base and form a coupled model of BP neural network and fuzzy control.

[0069] A fuzzy control rule base is established, and the neural network prediction results are transformed into fuzzy linguistic variables, such as "high frost layer" and "low humidity". The defrosting heating power and duration are determined through fuzzy reasoning.

[0070] Fuzzy control is an intelligent control method based on fuzzy set theory, which transforms expert knowledge and experience into fuzzy rules. Fuzzy control rules contain the following logic:

[0071] If the predicted frost thickness is greater than the threshold and the ambient humidity is high, then the defrosting time will be prolonged.

[0072] IF gate switching frequency is high THEN shortens defrosting interval.

[0073] Defuzzification is performed using the center of gravity method, and specific control parameters are output.

[0074] S6. Adjustment of weights in the fuzzy control rule base

[0075] Based on actual needs and experience with refrigerator defrosting, the output of the BP neural network prediction model is used as the input to the fuzzy control rule model. The fuzzy rule weights are dynamically adjusted through online learning to form a closed-loop optimization system. The coupled model refers to a collaborative decision-making system formed by data interaction between the neural network prediction module and the fuzzy control module.

[0076] S7. Refrigerator defrosting dynamic adjustment strategy based on cumulative data

[0077] Statistical analysis of cumulative defrost data and comparison of states at adjacent times is key to shifting the refrigerator defrost system from "passive response" to "active optimization".

[0078] By real-time statistics of the cumulative defrosting time and frequency of the refrigerator, and comparing defrosting parameters (such as heating power and duration) with operating status (such as compressor start-stop frequency) at adjacent time points (t and t-1), the system automatically triggers strategy adjustments when the cumulative value exceeds a preset threshold (such as more than 3 defrostings or more than 2 hours of defrosting in a single day) or when an abnormal state is detected (such as sensor failure or abnormal frost thickness). Adjustment measures include extending the defrosting interval, increasing heating power, and sending maintenance reminders, forming a closed-loop control mechanism to optimize defrosting efficiency, reduce energy consumption, and ensure safe equipment operation.

[0079] S8. Construct a comprehensive defrosting evaluation function.

[0080] The comprehensive defrosting evaluation function includes a comprehensive evaluation function that considers defrosting efficiency, energy consumption, and temperature fluctuation. It calculates the evaluation index at time t based on the input correction command. If the index at time t is better than that at time t-1, a control signal is output; otherwise, the model is retrained. The comprehensive evaluation function J is constructed as follows:

[0081] J=αE+βΔT+γD

[0082] Where E: defrosting energy consumption (Wh), which is calculated in real time using a current sensor;

[0083] ΔT: Temperature fluctuation range inside the refrigerator (°C) due to the defrosting device;

[0084] D: Residual frost thickness (mm), obtained by weighing or laser ranging;

[0085] α, β, γ: weighting coefficients.

[0086] Optimization objective: Minimize J, i.e.:

[0087] J * =min(αE+βΔT+γD)

[0088] The above embodiments are illustrative of the present invention and not intended to limit the invention. The described embodiments are merely some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art inspired by the present invention without creative effort are within the scope of protection of the present invention.

Claims

1. A refrigerator defrosting control method based on the coupling of BP neural network and fuzzy control, characterized in that: Data-driven modeling enables deep perception of the dynamic characteristics of the refrigeration system: a BP neural network learns actual data signals from the compressor and evaporator to indirectly estimate implicit parameters such as refrigerant charge and frost morphology distribution; a fuzzy controller dynamically adjusts the defrosting strategy based on the neural network prediction results, achieving coordinated optimization of refrigeration cycle parameters and the frosting process, including the following: S1 - Data Acquisition and Preprocessing: Acquire refrigerator operation data, clean and normalize the data, and divide it into training set, validation set and test set; Refrigerator operating data includes evaporator temperature, ambient temperature and humidity, door opening and closing frequency, historical defrosting records, refrigeration cycle parameters, compressor operating characteristics, evaporator physical structure parameters, and system ventilation condition parameters; Normalization scales data to a specified range: [0,1] or [-1,1]; the formula is... ； In the formula: x′ represents the normalized data, and x represents the input feature; min(x) is the minimum value of this feature; max(x) is the maximum value of this feature; The model is trained using the training set through iterative backpropagation and parameter update processes; the model's hyperparameters are tuned using the validation set; and the model's performance is evaluated using the test set. S2 - Constructing a BP neural network prediction model for refrigerator defrosting control: A BP neural network consists of an input layer, a hidden layer, and an output layer. It constructs a BP neural network model that takes evaporator temperature, environmental parameters, and user behavior as inputs and frost thickness prediction and defrosting time as outputs. S3-BP neural network model training: Training is performed using the ReLU activation function and the Adam optimizer; the mathematical expression of the ReLU activation function is the following piecewise function. ； In the formula, x is the input value and y is the output value; Performance evaluation of the S4-BP neural network model: The model hyperparameters were tuned using a grid search algorithm, and the mean squared error (MSE) model was used for evaluation. The MSE is the loss function, used to measure the difference between the neural network's predicted output and the true label. The MSE is represented by σ, and the formula is: ； In the formula: n is the number of input layer nodes, For the first The predicted output for each sample, The average; S5 - Constructing a coupled model of BP neural network and fuzzy control: Establish a fuzzy control rule base, transform the neural network prediction results into fuzzy linguistic variables, and determine the defrosting heating power and duration through fuzzy reasoning; S6-Fuzzy Control Rule Base Weight Adjustment: Based on the actual needs and experience of refrigerator defrosting, the output of the BP neural network prediction model is used as the input of the fuzzy control rule model. The weights of the fuzzy rules are dynamically adjusted through online learning to form a closed-loop optimization system. The coupled model refers to the collaborative decision-making system formed by the neural network prediction module and the fuzzy control module through data interaction. S7 - Refrigerator defrosting dynamic adjustment; Among them, the key to the refrigerator defrosting system's shift from passive response to active optimization is to collect and compare the cumulative defrosting data and the status at adjacent time points. S8 - Construct a comprehensive defrosting evaluation function: The model includes a comprehensive evaluation function that includes defrosting efficiency, energy consumption, and temperature fluctuation. The model is calculated at time t by inputting correction instructions. If the evaluation index at time t is better than that at time t-1, the model is output as a control signal. Otherwise, the model is retrained. Construct a comprehensive evaluation function The formula is ； Optimization goal: Minimize The formula is ； In the formula, E: defrosting energy consumption (Wh), which is calculated in real time by a current sensor; D: Temperature fluctuation range inside the refrigerator (°C) due to defrosting device; D: Residual frost thickness (mm), obtained by weighing or laser ranging. Weighting coefficients.

Images