Search

CN-121996905-A - Improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion

CN121996905ACN 121996905 ACN121996905 ACN 121996905ACN-121996905-A

Abstract

The invention belongs to the technical field of artificial intelligence and discloses an improved CNN air-cooled refrigerator energy consumption online estimation algorithm combining dynamic feature expansion, which comprises the following steps of 1, collecting data, preprocessing and normalizing, 2, feature extraction, splicing the normalized data at the current and historical moments to form expansion input features, calculating the change amplitude of an operation state, determining the length of a time window, 3, constructing a lightweight improved CNN, carrying out one-dimensional convolution extraction on time sequence data to obtain local time sequence features, carrying out convolution and pooling, carrying out regression output energy consumption estimation through a full connection layer, 4, quantization processing, 5, constructing training samples, and calculating loss after training 6, Solidifying the model parameters after training convergence and deploying the model parameters to a refrigerator controller, circularly executing and outputting in real time . The algorithm improves the prediction accuracy of the energy consumption of the refrigerator and meets the real-time operation requirement of the low-power consumption embedded system.

Inventors

  • ZHANG WEI
  • LI HAORAN

Assignees

  • 余姚市机器人研究中心
  • 浙江大学

Dates

Publication Date
20260508
Application Date
20251217

Claims (10)

  1. 1. An improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic characteristic expansion is characterized by comprising the following steps: step 1, acquiring running state data of a refrigerator to form a running state vector, preprocessing the running state vector to form continuous and stable time sequence data, and normalizing the time sequence data; Step 2, performing feature extraction on the standardized time series data by adopting a convolutional neural network, splicing the standardized data at the current moment and the historical moment by adopting a sliding time window to form an extended input feature, calculating the change amplitude of the running state, and determining the length of the time window; step 3, constructing a lightweight improved CNN, carrying out one-dimensional convolution on time sequence data in a time dimension to extract local time sequence characteristics, carrying out convolution and pooling to form high-level characteristics, and carrying out regression output energy consumption estimation through a full-connection layer; Step 4, carrying out quantization processing on the model parameters and the intermediate features, and converting floating point data into integer data with lower bit width; Step 5, constructing a training sample by using historical operation data, calculating the difference loss between the energy consumption estimated predicted value and the energy consumption actual value by using a mean square error as a loss function, and performing parameter optimization by using Adam to meet a preset condition, and judging that the energy consumption estimated predicted value and the energy consumption actual value are converged; Step 6, solidifying and deploying the model parameters after training convergence to a refrigerator controller, circularly executing the model parameters during online operation, and outputting the model parameters in real time As a result of the online energy consumption estimation.
  2. 2. The improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion according to claim 1, wherein in step 1, the operation state data comprises compressor operation frequency Rotation speed of evaporating fan Rotational speed of condensing fan Temperature of air in cabinet Ambient temperature ; The expression of the running state vector is: Wherein, the Is shown in the first A state vector formed by the sampling moments.
  3. 3. The improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion according to claim 1, wherein in the step 1, the preprocessing operation comprises data synchronization, outlier processing, missing value compensation and smoothing operation.
  4. 4. The improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion according to claim 1, wherein in the step 1, the normalization is to convert refrigerator operation state parameters with different physical meanings and different dimensions into time series input data with uniform dimensions, and the normalized processing calculation expression is as follows: Wherein, the Representing the sensor data after normalization, The mean value of the data is represented, Representing standard deviation of data, then at The standardized running state vector expression obtained at each sampling moment is as follows: Wherein, the Representing the normalized compressor operating parameters, Represents the operation parameters of the standardized evaporating fan, Represents the operation parameters of the condensing fan after standardization, The standardized cabinet temperature parameters are represented, Representing the normalized ring temperature parameter.
  5. 5. The improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion according to claim 1, wherein in the step 2, the calculation process of the feature extraction is as follows: feature extraction is carried out on any input time sequence, and the output feature calculation expression after convolution is as follows: Wherein, the In order to input a signal to the device, For the convolution kernel weights, Is the length of the convolution kernel; the calculation expression of the extended input features is as follows: Wherein, the Representing the input characteristics after expansion, Representing the normalized running state vector at the current time, The time step is represented by a time step, Is shown in the first Sampling time; the calculation expression of the operation state change amplitude is as follows: Wherein, the A standardized running state vector at the current moment; the calculation expression of the time window length is as follows: Wherein, the , The window length is respectively lower and upper, In order to adjust the coefficient of the power supply, The representation is limited to the result that is included, Representing a rounding down.
  6. 6. The improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion according to claim 1, wherein in the step 3, the energy consumption estimation calculation process is as follows: (1) For time series data, extracting local dependence on time dimension by adopting one-dimensional convolution, for the first And the characteristic output calculation expression of the convolution layer is as follows: Wherein, the Representing the time position of an input feature Is used for the value of (a) and (b), Represent the first The convolution kernel is at The weight of the individual positions is determined, Representing the length of the convolution kernel, The term of the bias is indicated, Representing the convolved feature output; (2) The convolution output is input to a pooling layer for downsampling, and the pooling output expression is as follows: Wherein, the The size of the pooled window is indicated, Representing the pooled output characteristics; (3) Flattening the characteristics of the multilayer convolution and pooling, inputting the flattened characteristics into a full-connection layer for fusion, and outputting an energy consumption estimated value, wherein the energy consumption estimated value is calculated as follows: Wherein, the Representing the high-level feature vector obtained by flattening after convolution and pooling, The weight matrix is represented by a matrix of weights, As a result of the bias term, Represent the first Energy consumption estimates for each sampling instant.
  7. 7. The improved online estimation algorithm for the energy consumption of the CNN air-cooled refrigerator combined with dynamic feature expansion according to claim 6, wherein the number of the convolution layers is2 to 4, the number of convolution cores in each convolution layer is 16, and the length of the convolution cores is 3; the pooling window size of the pooling layer is 2, and the neuron number of the full-connection layer is 16.
  8. 8. The improved CNN air-cooled refrigerator energy consumption online estimation algorithm according to claim 1, wherein in the step 4, the quantization process is to convert floating point data into integer data with a lower bit width, and the quantized data is converted by the following formula: Wherein, the The data representing the floating point is displayed, The quantization step size is represented as, Representing the quantized integer value.
  9. 9. The improved CNN air-cooled refrigerator energy consumption online estimation algorithm combining dynamic feature expansion according to claim 1 is characterized in that in the step 4, different quantization modes are adopted for a convolution layer and a full-connection layer by quantization treatment, specifically, the strategy is that (1) 8-bit quantization is adopted for the convolution layer, and (2) 4-bit quantization is adopted for the full-connection layer.
  10. 10. The improved CNN air-cooled refrigerator energy consumption online estimation algorithm according to claim 1, wherein in the step 5, the difference loss between the predicted value and the actual value is calculated by the following expression: Wherein, the The number of samples is represented and the number of samples, The true value is represented by a value that is true, Representing the predicted value; The Adam is adopted for parameter optimization, and the optimal parametric calculation expression is as follows: Wherein, the And Representing first and second moment estimates of the gradient respectively, The learning rate is indicated as being indicative of the learning rate, To prevent a small constant of zero removal.

Description

Improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion Technical Field The application belongs to the technical field of artificial intelligence, and particularly relates to an improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion. Background With the wide application of the air-cooled refrigerator in the fields of household appliances and business, how to efficiently estimate the running energy consumption of the refrigerator becomes an important research direction for optimizing the energy saving and control strategies of the whole refrigerator. The prior art energy consumption estimation methods mostly rely on traditional mathematical statistical models or sensor-based data analysis. In addition, there are some conventional neural network methods, such as a multi-layer perceptron and a simple convolutional neural network, which can improve the accuracy, but still have the following technical problems: (1) The mathematical statistical model does not fully consider dynamic characteristics, the time dependence between the working state of the refrigerator and the external environment cannot be effectively captured, and the accuracy of the model is difficult to guarantee. (2) And the energy consumption is calculated by adopting a sensor, and the cost is high. (3) The existing energy consumption prediction method has the defects of complex model and large calculated amount. Therefore, a new method is needed to combine the historical data of the refrigerator with the current data by introducing a dynamic characteristic expansion technology, and the adaptability of the model to the running state of the refrigerator and the change of the external environment is enhanced, so that the accuracy of energy consumption prediction is improved. Disclosure of Invention In order to solve the technical problems in the prior art, the invention aims to improve the energy consumption prediction precision of the refrigerator, and the technical scheme is as follows: An improved CNN air-cooled refrigerator energy consumption online estimation algorithm combined with dynamic feature expansion comprises the following steps: step 1, acquiring running state data of a refrigerator to form a running state vector, preprocessing the running state vector to form continuous and stable time sequence data, and normalizing the time sequence data; Step 2, performing feature extraction on the standardized time series data by adopting a convolutional neural network, splicing the standardized data at the current moment and the historical moment by adopting a sliding time window to form an extended input feature, calculating the change amplitude of the running state, and determining the length of the time window; step 3, constructing a lightweight improved CNN, carrying out one-dimensional convolution on time sequence data in a time dimension to extract local time sequence characteristics, carrying out convolution and pooling to form high-level characteristics, and carrying out regression output energy consumption estimation through a full-connection layer; Step 4, carrying out quantization processing on the model parameters and the intermediate features, and converting floating point data into integer data with lower bit width; Step 5, constructing a training sample by using historical operation data, calculating the difference loss between the energy consumption estimated predicted value and the energy consumption actual value by using a mean square error as a loss function, and performing parameter optimization by using Adam to meet a preset condition, and judging that the energy consumption estimated predicted value and the energy consumption actual value are converged; Step 6, solidifying and deploying the model parameters after training convergence to a refrigerator controller, circularly executing the model parameters during online operation, and outputting the model parameters in real time As a result of the online energy consumption estimation. Further, in the step 1, the operation state data includes a compressor operation frequencyRotation speed of evaporating fanRotational speed of condensing fanTemperature of air in cabinetAmbient temperature; The expression of the running state vector is: Wherein, the Is shown in the firstA state vector formed by the sampling moments. Further, in the step 1, the preprocessing operation includes data synchronization, outlier processing, missing value compensation and smoothing operation. Further, in the step 1, the normalization is to convert the running state parameters of the refrigerator with different physical meanings and different dimensions into time series input data with uniform dimensions, and the normalized processing calculation expression is as follows: Wherein, the Representing the sensor data after normalization,The mean value of the data is repre