Detergent dosage calculation method and device and washing equipment
Through the improved time domain convolution network TCN model, combined with the normalized mutual information calculation of washing parameters, the problem of inaccurate detergent dosage in washing equipment is solved, and more intelligent detergent dosage calculation is achieved, improving the efficiency and user experience of the laundry process.
Patent Information
- Application Number
- CN202510415265.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-03
- Publication Date
- 2025-05-23
AI Technical Summary
It is difficult for existing washing equipment to accurately calculate the amount of detergent based on specific washing scenarios, resulting in too much or insufficient foam during the washing process, affecting the user experience.
Using the improved time domain convolution network TCN model, the corresponding relationship between multiple washing parameters and detergent usage is obtained, normalized mutual information is calculated, key parameters are selected, and network training is used using the Swish function and the improved expansion convolutional structure to calculate the real-time detergent usage.
It realizes the calculation of the appropriate amount of detergent usage based on specific washing scenarios, improves the intelligence of washing equipment, reduces foam generation and sewage discharge, and improves the user experience.
Smart Images
Figure CN120026461A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of washing equipment, and in particular to a method and device for calculating the amount of detergent used, and washing equipment. Background Art
[0002] At present, for washing equipment, before washing, users put detergent into the washing equipment. If too much detergent is put in, more foam will be generated during the washing process. In order to eliminate the foam, the amount of water will be increased, and the washing machine will discharge more sewage. If too little detergent is put in, the clothes will not be washed clean, which will affect the user experience. Therefore, how to put the right amount of detergent according to the specific washing scene is particularly important for the intelligence of washing equipment. However, in the current existing technology, the amount of detergent put into the washing equipment is generally a fixed amount or put through a pre-set relationship comparison table, which is difficult to match the washing parameters well, and the amount of detergent put in is still not accurate. Summary of the invention
[0003] The present application provides a detergent dosage calculation method, device and washing equipment, which have the advantage of being able to calculate the appropriate amount of detergent according to the specific washing scenario, which is conducive to improving the intelligence level of the washing equipment.
[0004] This application plan is:
[0005] In one aspect, the present application provides a method for calculating the amount of detergent, comprising the following steps:
[0006] S1. Obtaining a correspondence between a plurality of washing parameters and a detergent dosage for a washing device;
[0007] S2, for the data collected in S1, respectively calculating the normalized mutual information of multiple washing parameters and detergent dosage, and selecting several washing parameters that are highly correlated with detergent dosage;
[0008] S3, using the selected washing parameters as input features for data processing and then feeding them into the detergent dosage calculation model for network training;
[0009] The detergent dosage calculation model is an improvement of the time domain convolutional network TCN, which is improved as follows:
[0010] (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN;
[0011] (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially.
[0012] S4. Deploy the trained model to the washing equipment controller. The controller collects the real-time washing parameters of the washing equipment and calculates the amount of detergent required for washing.
[0013] Furthermore, in step S1, the washing parameters include water temperature parameters, water intake parameters, washing mode parameters, laundry weight parameters, washing equipment type parameters, clothing cleanliness, foam volume, and outlet water quality, wherein the washing mode parameters include rotation speed, number of washing beats, and laundry material.
[0014] Furthermore, in step S1, a correspondence between a plurality of washing parameters and the amount of detergent is obtained through a full factorial experiment.
[0015] Further, in step S2, for the washing parameter X={x 1 ,x 2 ,…,x k}, then the entropy of X is:
[0016]
[0017] Where p(x k ) is x k The probability density of two variables (X, Y) is:
[0018]
[0019] Where p(x k ,y l ) is x k and l The relationship between mutual information entropy and joint entropy can be expressed as:
[0020] I(X,Y)=H(X)+H(Y)-H(X,Y)
[0021] Normalized mutual information scales the mutual information to between [0,1], and its formula is:
[0022] NMI(X,Y)=2I(X,Y) / [H(X)+H(Y)].
[0023] Furthermore, in step S3, the improved time domain convolutional network TCN adopts dilated convolution, and its formula is as follows:
[0024]
[0025] Where f is the convolution kernel, {0,…,k-1}∈R, k is the convolution kernel size, x is the input sequence, and d is the dilation factor;
[0026] The function formulas of the two expansion factors are as follows:
[0027]
[0028] C is a constant, representing the expansion factor The upper limit of , k is a parameter;
[0029] The improved TCN dilated convolutional layer is described as follows:
[0030]
[0031]
[0032] H i =Drop(Swish(F i ))+H i-1
[0033] in With expansion factor The dilated convolution, For bias, is the weight, H i is the output of the i-th layer, Drop is the Dropout layer, F i for and The convolution kernels of all expansion layers are set to 3.
[0034] In another aspect, the present application provides a detergent dosage calculation device, comprising:
[0035] A washing parameter acquisition module, used to acquire washing parameters of a washing device;
[0036] The calculation module deploys a trained detergent dosage calculation model, calculates the detergent dosage according to the washing parameters of the washing equipment and sends it to the washing equipment;
[0037] The detergent dosage calculation model is an improvement of the time domain convolutional network TCN, which is improved as follows:
[0038] (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN;
[0039] (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially.
[0040] In another aspect, the present application provides a washing device, comprising
[0041] A washing parameter acquisition module, used to acquire washing parameters of a washing device;
[0042] The calculation module is deployed with a trained detergent dosage calculation model, and the detergent dosage is calculated according to the washing parameters of the washing equipment; wherein the detergent dosage calculation model is an improvement of the time domain convolutional network TCN, and the improvement is:
[0043] (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN;
[0044] (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially.
[0045] And a detergent adding module, used for adding a corresponding amount of detergent according to the output result of the calculation module.
[0046] In summary, the beneficial effects of the present application include: being able to calculate the appropriate amount of detergent according to specific washing parameters, which is beneficial to improving the intelligence level of washing equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 It is a schematic diagram of the overall process of the detergent dosage calculation method of the present application;
[0048] Figure 2 It is the ReLU and Swish activation unit curve graph;
[0049] Figure 3 is the exponential expansion factor and constrained exponential expansion factor curve. DETAILED DESCRIPTION
[0050] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.
[0051] Example 1: A method for calculating the amount of detergent used, referring to Figure 1 , including the following steps:
[0052] S1. Obtaining a correspondence between multiple washing parameters and detergent dosage for a washing device. The washing parameters of the washing device include water temperature parameters, water inlet parameters, washing mode parameters, laundry weight parameters, washing device type parameters, laundry cleanliness, foam volume, and water quality. The washing mode parameters include rotation speed, number of washing beats, and laundry material.
[0053] The correspondence between multiple washing parameters and detergent dosage was obtained through full factorial experiments.
[0054] In the experiment, ensure that the value of each parameter covers its actual possible range (for example, water temperature must include low / medium / high extreme values); Sample size requirement: Based on the variable dimension, the sample size must satisfy n≥10×k (k is the number of parameters).
[0055] A full factorial experimental design is used, where all level combinations of all factors (variables) are to be experimented with.
[0056] For example: study detergent type (2 types) × water temperature (3 levels) × speed (3 levels);
[0057] Number of full factorial experiments = 2 × 3 × 3 = 18 combinations;
[0058] Suppose we study the following three factors:
[0059] Detergent dosage: 5ml, 10ml, 15ml (3 levels);
[0060] Water temperature: 30℃, 40℃, 60℃ (3 levels);
[0061] Washing mode: Gentle, Standard, Strong (3 levels);
[0062] The full factorial experiment requires 3×3×3=27 experiments. Some examples of combinations are shown in Table 1.
[0063] Table 1: Some experimental combinations
[0064]
[0065] S2. Calculate the normalized mutual information of the data collected in S1 respectively, and select several washing parameters that are highly correlated with the target parameters; specifically, collect all the data through a full factorial experiment and then perform a mutual information test.
[0066] In step S2, for the washing parameter X={x 1 ,x 2 ,…,x k}, then the entropy of X is:
[0067]
[0068] Where p(x k ) is x k The probability density of two variables (X, Y) is:
[0069]
[0070] Where p(x k ,y l ) is x k and lThe relationship between mutual information entropy and joint entropy can be expressed as:
[0071] I(X,Y)=H(X)+H(Y)-H(X,Y)
[0072] Normalized mutual information scales the mutual information to between [0,1], and its formula is:
[0073] NMI(X,Y)=2I(X,Y) / [H(X)+H(Y)].
[0074] In this embodiment, the parameters finally selected through mutual information calculation are: water temperature, water inlet, washing machine speed, laundry material, laundry weight parameter, and outlet foam density value. The above parameters are used as input sequences to train the subsequent time domain convolutional network. The input and output data of the time domain convolutional network are shown in Table 2.
[0075] Table 2: Example of input and output data of the time domain convolutional network
[0076]
[0077] The laundry material uses One-Hot coding, and the laundry material conversion table is shown in Table 3.
[0078] Table 3: Conversion table of laundry materials
[0079] cowboy wool Silk Cotton and Linen down jacket baby 32 16 8 4 2 1
[0080] S3, using the selected washing parameters as input features for data processing and then feeding them into the detergent dosage calculation model for network training;
[0081] The detergent dosage calculation model is an improvement of the time domain convolutional network TCN, which is improved as follows:
[0082] (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN;
[0083] (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially.
[0084] The time domain convolutional network uses dilated convolution, and its formula is as follows:
[0085]
[0086] Where f is the convolution kernel, {0,…,k-1}∈R, k is the convolution kernel size, x is the input sequence, d is the expansion factor, and its model structure is as follows Figure 2 shown.
[0087] TCN has a wider receptive field and powerful parallel computing capabilities. However, traditional TCN still has difficulty understanding irregularities and complex relationships in sequences. Therefore, this authorization makes two improvements to TCN. The first aspect: improving the activation function to overcome the problems of discontinuous information extraction and gradient vanishing of TCN. The second aspect: improving the dilated convolution so that TCN always has a more appropriate receptive field and an intermediate layer near a smaller receptive field. Each expansion layer has two expansion factors, one expansion factor grows exponentially and the other expansion factor grows linearly. The basic structure of the extended convolution layer is shown in the figure.
[0088] The Swish function activation unit is the key to the neural network learning nonlinear functions. Compared with the ReLU function, the Swish function has a smoother curve when it is close to zero, such as Figure 2 As shown, and because it uses the sigmoid function, the output range of the network can be between 0 and 1. This makes Swish perform better than ReLU in some applications, and Swish works better than ReLU on deep models. It can be regarded as a smoothed ReLU activation function. The Swish formula is as follows:
[0089]
[0090] As the number of network layers increases, the traditional exponential expansion factor will no longer be reasonable, such as Figure 3 As shown in the figure, when the number of network layers i reaches 15, the value of the expansion factor d has reached 16000, and its receptive field does not optimize the network according to the actual situation of the data. Therefore, each expansion layer of the TCN network improved by this application has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially. The function formula is as follows:
[0091]
[0092]
[0093] C is a constant, representing the expansion factor The upper limit of , k is a parameter. It starts to grow from x=0, and as x increases, the growth rate gradually slows down, but it will never reach the upper limit C. The parameter k controls the growth rate. The combination of parameters C and k makes Limited exponential growth.
[0094] The improved TCN dilated convolutional layer can be described as:
[0095]
[0096] Hi =Drop(Swish(F i ))+H i-1
[0097] in With expansion factor The dilated convolution, For bias, is the weight, H i is the output of the i-th layer, Drop is the Dropout layer, F i for and The convolution kernels of all expansion layers are set to 3.
[0098] The model performance was evaluated using the following method:
[0099] MSE: measures the mean of the squared errors and is given by the following equation:
[0100]
[0101] MAPE: is the average or mean of the absolute percentage errors of the predictions:
[0102]
[0103] Where Y(i) is the predicted value, y(i) is the actual value, and is the average of the predicted values.
[0104] S4. Deploy the trained model to the washing equipment controller. The controller collects the real-time washing parameters of the washing equipment and calculates the amount of detergent required for washing.
[0105] Embodiment 2, a detergent dosage calculation device, comprising:
[0106] A washing parameter acquisition module, used to acquire washing parameters of a washing device;
[0107] The calculation module deploys a trained detergent dosage calculation model, calculates the detergent dosage according to the washing parameters of the washing equipment and sends it to the washing equipment;
[0108] The detergent dosage calculation model is an improvement of the time domain convolutional network TCN, which is improved as follows:
[0109] (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN;
[0110] (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially.
[0111] Embodiment 3, a washing device, comprising
[0112] A washing parameter acquisition module, used to acquire washing parameters of a washing device;
[0113] The calculation module is deployed with a trained detergent dosage calculation model, and the detergent dosage is calculated according to the washing parameters of the washing equipment; wherein the detergent dosage calculation model is an improvement of the time domain convolutional network TCN, and the improvement is:
[0114] (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN;
[0115] (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially.
[0116] And a detergent adding module, used for adding a corresponding amount of detergent according to the output result of the calculation module.
[0117] The above is only a preferred implementation of the present application. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept of the present application, and these all fall within the scope of protection of the present application.
Claims
1. A method for calculating the amount of detergent, characterized in that: The following steps are involved: S1. Obtaining a correspondence between a plurality of washing parameters and a detergent dosage for a washing device; S2, for the data collected in S1, respectively calculating the normalized mutual information of multiple washing parameters and detergent dosage, and selecting several washing parameters that are highly correlated with detergent dosage; S3, using the selected washing parameters as input features for data processing and then feeding them into the detergent dosage calculation model for network training; The detergent dosage calculation model is an improvement of the time domain convolutional network TCN, which is improved as follows: (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN; (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially. S4. Deploy the trained model to the washing equipment controller. The controller collects the real-time washing parameters of the washing equipment and calculates the amount of detergent required for washing.
2. The method for calculating the amount of detergent according to claim 1, characterized in that: In step S1, the washing parameters include water temperature parameters, water intake parameters, washing mode parameters, laundry weight parameters, washing equipment type parameters, clothing cleanliness, foam volume, and outlet water quality, wherein the washing mode parameters include rotation speed, number of washing beats, and laundry material.
3. The method for calculating the amount of detergent according to claim 2, characterized in that: In step S1, a correspondence between a plurality of washing parameters and the amount of detergent is obtained through a full factorial experiment.
4. The method for calculating the amount of detergent according to claim 1, characterized in that: In step S2, for the washing parameters X={x1,x2,…,x k }, then the entropy of X is: Where p(x k ) is x k The probability density of two variables (X, Y) is: Where p(x k ,y l ) is x k and l The relationship between mutual information entropy and joint entropy can be expressed as: I(X,Y)=H(X)+H(Y)-H(X,Y) Normalized mutual information scales the mutual information to between [0,1], and its formula is: NMI(X,Y)=2I(X,Y) / [H(X)+H(Y)] 5. The method for calculating the amount of detergent according to claim 1, characterized in that: In step S3, the improved time domain convolutional network TCN adopts dilated convolution, and its formula is as follows: Where f is the convolution kernel, {0,…,k-1}∈R, k is the convolution kernel size, x is the input sequence, and d is the dilation factor; The function formulas of the two expansion factors are as follows: C is a constant, representing the expansion factor The upper limit of , k is a parameter; The improved TCN dilated convolutional layer is described as follows: H i =Drop(Swish(F i ))+H i-1 in With expansion factor The dilated convolution, For bias, is the weight, H i is the output of the i-th layer, Drop is the Dropout layer, F i for and The convolution kernels of all expansion layers are set to 3.
6. A detergent dosage calculation device, characterized in that: include: A washing parameter acquisition module, used to acquire washing parameters of a washing device; The calculation module deploys a trained detergent dosage calculation model, calculates the detergent dosage according to the washing parameters of the washing equipment and sends it to the washing equipment; The detergent dosage calculation model is an improvement of the time domain convolutional network TCN, which is improved as follows: (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN; (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially.
7. A washing device, characterized in that: include A washing parameter acquisition module, used to acquire washing parameters of a washing device; The calculation module is deployed with a trained detergent dosage calculation model, and the detergent dosage is calculated according to the washing parameters of the washing equipment; wherein the detergent dosage calculation model is an improvement of the time domain convolutional network TCN, and the improvement is: (1) Use the Swish function to replace the ReLU function in the time domain convolutional network TCN; (2) Improve the dilated convolution structure in the time domain convolutional network (TCN). In the improved time domain convolutional network (TCN), each expansion layer has two expansion factors, one expansion factor grows linearly, and the other expansion factor grows exponentially. And a detergent adding module, used for adding a corresponding amount of detergent according to the output result of the calculation module.