CT-SVM (Computed Tomography-Support Vector Machine)-based dehumidifier water level prediction method and system

Abstract

The invention relates to the technical field of dehumidifiers, in particular to a dehumidifier water level prediction method and system based on CT-SVM. According to the method, related data of historical dehumidifiers are collected, a mixed feature selection method combining a Filter method and a Wrapper method is adopted, a detection rate and a false detection rate are used as evaluation bases of feature selection, a Relief method is used for performing correlation sorting on features, and according to the sorted features, an SBS search strategy is adopted from back to front and an SVM is used as a classifier, so that the characteristics of the dehumidifiers are classified, and the characteristics of the dehumidifiers are classified. Feature screening is conducted on the collected relevant data of the dehumidifier, an optimal feature subset is selected, an adaptive coefficient is obtained through an adaptive feature operator extraction algorithm, and based on the optimal feature subset and the adaptive coefficient, a CT-SVM algorithm is adopted to conduct water level prediction on the dehumidifier; the CT-SVM algorithm is adopted for water level prediction, accuracy and adaptability are higher, and model training is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of dehumidifiers, and particularly to a method and system for predicting the water level of a dehumidifier based on CT-SVM. Background Art

[0002] As an important household appliance for regulating indoor humidity, dehumidifiers are widely used in daily life and industrial production. Traditional dehumidifiers are usually equipped with a water tank to collect the water condensed from the air during the dehumidification process. However, during actual use, users cannot determine when the water tank is full. This problem brings many inconveniences to users. For example, when users do not notice in time that the water tank is full, the dehumidifier may cause the surrounding environment to be soaked by water due to water tank overflow, damaging items such as furniture and floors; if users frequently check the water tank, it will consume unnecessary time and energy, reducing the convenience of the user experience.

[0003] To solve this problem, some dehumidifiers in the prior art adopt a water full indicator light or an alarm device. However, the water full indicator light requires users to actively observe the dehumidifier panel to know the status of the water tank. When users do not pay attention to the dehumidifier, they may still miss the reminder that the water tank is full; although the alarm device can emit a sound reminder when the water tank is full, in some noisy environments or when users are not present, it cannot ensure that users receive the notice in time. In addition, these reminder methods cannot let users know in advance that the water tank is about to be full, making it difficult to make preparations in advance, and there is a certain lag. Therefore, researching and developing a technology that can accurately and timely let users know the status of the water tank, especially predicting when the water tank will be full, is of great significance for improving the performance and user experience of dehumidifiers. Summary of the Invention

[0004] Aiming at the above deficiencies of the prior art, the present invention aims to provide a method and system for predicting the water level of a dehumidifier based on CT-SVM, so as to realize the prediction of the water level in the dehumidifier water tank and facilitate users to master the water level situation of the dehumidifier water tank.

[0005] To solve the above problems, the present invention adopts the following technical solutions:

[0006] On the one hand, the present invention provides a method for predicting the water level of a dehumidifier based on CT-SVM, including:

[0007] Collect relevant data of historical dehumidifiers, including indoor environmental data, outdoor environmental data, operating status data, time and location data, and the rate of change of the water level in the water tank.

[0008] A hybrid feature selection method combining Filter and Wrapper methods is adopted. The detection rate and false detection rate are used as the evaluation basis for feature selection. The Relief method is used to rank the features according to their correlation. Starting from the last feature in the sorted order, the SBS search strategy is adopted with SVM as the classifier to screen the features of the collected dehumidifier-related data, select the optimal feature subset, and obtain the adaptive coefficient through the adaptive feature operator extraction algorithm.

[0009] Based on the optimal feature subset and the adaptive coefficient, the CT-SVM algorithm is used to predict the water level of the dehumidifier.

[0010] As an implementable manner, the selection of the optimal feature subset includes:

[0011] S201. In the Filter stage, calculate the Relief weights of each feature for the original features in the dehumidifier-related data.

[0012] S202. According to the calculated Relief weights of each feature, sort the original features in the dehumidifier-related data in descending order to construct a feature sequence.

[0013] S203. In the Wrapper stage, adopt the SBS search strategy to sequentially remove one feature X from the back of the feature sequence. Using SVM as the classifier, calculate the detection rate, false detection rate, and feature evaluation value J of the remaining feature subset. If the feature evaluation value J is satisfied after removing feature X 剔除Xk is greater than J 剔除Xk+1 , then repeat S203 until J 剔除Xk is less than or equal to J 剔除Xk+1 Then determine the remaining features in the current feature sequence as the optimal feature subset, where k is the kth feature removed from the feature sequence.

[0014] As an implementable manner, the obtaining of the adaptive coefficient includes:

[0015] Based on the historical data of the optimal feature subset, use the random forest algorithm to calculate the prediction error α of each sample i , where α i =|Y i -yβ i |, where i is the sample number, Y i is the actual water tank water level change rate of the ith sample, y is the predicted water tank water level change rate, and β i is the adaptive coefficient of the ith sample.

[0016] If the prediction error is greater than the set threshold, adjust the weight of the sample and retrain to predict the adaptive coefficient; if the prediction error is less than or equal to the set threshold, the training is completed to obtain the final adaptive coefficient.

[0017] As an implementable manner, the water level prediction of the dehumidifier using the CT-SVM algorithm includes:

[0018] Using the Compressive Sensing (CT) technology to dynamically and real-time rank the importance of the features in the optimal feature subset based on a tree model, and screening out the features with importance lower than the importance threshold to obtain an optimal low-dimensional feature subset.

[0019] Combining the adaptive coefficient with the optimal low-dimensional feature subset to construct an enhanced input vector, inputting it into a non-linear support vector machine classification model, classifying the change rate of the water level in the water tank into multiple types as the output, determining the volume of the water tank according to the product model, and calculating the current water volume in the water tank and the estimated time to full water.

[0020] As an implementable manner, the classification decision function of the non-linear support vector machine classification model is:

[0021]

[0022]

[0023] .

[0024] Where N is the number of enhanced input vectors, is the optimal solution of the Lagrange multiplier, is the class label of the i-th training sample, is the kernel function, representing the vector in the original input space and the training sample vector mapped to a high-dimensional feature space, is the intercept term of the classification hyperplane, is the class label of the j-th training sample, is the kernel function, representing the inner product of the i-th training sample and the j-th training sample in the high-dimensional feature space, is 1 / 5.

[0025] On the other hand, the present invention provides a water level prediction system for a dehumidifier based on CT-SVM, including an acquisition module, a screening module, and a water level prediction module.

[0026] The acquisition module is used to collect relevant data of the historical dehumidifier, including indoor environmental data, outdoor environmental data, operating status data, time and location data, and the change rate of the water level in the water tank.

[0027] The screening module is used to adopt a hybrid feature selection method that combines the Filter and Wrapper methods. Taking the detection rate and false detection rate as the evaluation basis for feature selection, it uses the Relief method to rank the features according to their correlation. Starting from the last feature in the sorted feature list, it adopts the SBS search strategy with SVM as the classifier to screen the features of the collected dehumidifier-related data, select the optimal feature subset, and obtain the adaptive coefficient through the adaptive feature operator extraction algorithm.

[0028] The water level prediction module is used to predict the water level of the dehumidifier based on the optimal feature subset and the adaptive coefficient using the CT-SVM algorithm.

[0029] As an implementable manner, the selection of the optimal feature subset includes:

[0030] S201. In the Filter stage, calculate the Relief weight of each feature for the original features in the dehumidifier-related data.

[0031] S202. According to the calculated Relief weights of each feature, sort the original features in the dehumidifier-related data in descending order to construct a feature sequence.

[0032] S203. In the Wrapper stage, adopt the SBS search strategy to sequentially remove one feature X from the back of the feature sequence. Using SVM as the classifier, calculate the detection rate, false detection rate, and feature evaluation value J of the remaining feature subset. If the removal of feature X satisfies J 剔除Xk is greater than J 剔除Xk+1 , then repeat S203 until J 剔除Xk is less than or equal to J 剔除Xk+1 Then determine the remaining features in the current feature sequence as the optimal feature subset, where k is the kth feature removed from the feature sequence.

[0033] As an implementable manner, the obtaining of the adaptive coefficient includes:

[0034] Based on the historical data of the optimal feature subset, use the random forest algorithm to calculate the prediction error α of each sample i , where α i =|Y i -yβ i |, where i is the sample number, Y i is the true water tank water level change rate of the ith sample, y is the predicted water tank water level change rate, and β i is the adaptive coefficient of the ith sample.

[0035] If the prediction error is greater than the set threshold, adjust the weight of the sample and retrain the prediction adaptive coefficient; if the prediction error is less than or equal to the set threshold, the training is completed and the final adaptive coefficient is obtained.

[0036] As an implementable manner, the water level prediction of the dehumidifier using the CT-SVM algorithm includes:

[0037] Adopt the compressive tracking CT technology to dynamically and real-time rank the importance of the features in the optimal feature subset based on the tree model, screen out the features with importance lower than the importance threshold, and obtain the optimal low-dimensional feature subset.

[0038] Combine the adaptive coefficient with the optimal low-dimensional feature subset to construct an enhanced input vector, input it into the non-linear support vector machine classification model, divide the change rate of the water tank water level into multiple types as the output, determine the volume of the water tank according to the product model, and calculate the current water volume and the estimated time to full water of the water tank.

[0039] As an implementable manner, the classification decision function of the non-linear support vector machine classification model is:

[0040]

[0041]

[0042] .

[0043] Among them, N is the number of enhanced input vectors, is the optimal solution of the Lagrange multiplier, is the class label of the i-th training sample, is the kernel function, representing the vector in the original input space and the training sample vector are mapped to the high-dimensional feature space, is the intercept term of the classification hyperplane, is the class label of the j-th training sample, is the kernel function, representing the i-th training sample and the j-th training sample in the inner product of the high-dimensional feature space, is 1 / 5.

[0044] The beneficial effects of the present invention are as follows: The present invention adopts a hybrid feature selection method combining the Filter and Wrapper methods to screen the optimal feature subset and calculate the adaptive coefficient, and successfully constructs a water level prediction model through the CT-SVM algorithm model, realizing the prediction of the water level in the dehumidifier, which is more accurate than the general prediction model, and is suitable for various types of dehumidifiers, reducing model training. Description of the Drawings

[0045] Figure 1 Schematic diagram of the process of a dehumidifier water level prediction method based on CT - SVM according to the present invention.

[0046] Figure 2 Schematic diagram of the process of selecting the optimal feature subset according to the present invention.

[0047] Figure 3 Schematic diagram of a dehumidifier water level prediction system based on CT - SVM according to the present invention. Detailed implementation manners

[0048] The present invention will be further described in detail below in conjunction with specific embodiments.

[0049] It should be noted that these embodiments are only used to illustrate the present invention, rather than limiting the present invention. Any simple improvement of this method under the premise of the concept of the present invention falls within the scope of protection required by the present invention.

[0050] See Figure 1 , which is a dehumidifier water level prediction method based on CT - SVM, including:

[0051] S100. Collect relevant data of historical dehumidifiers, including indoor environmental data, outdoor environmental data, operating status data, time - area data, and water tank water level change rate.

[0052] Original data capture

[0053] Based on the Internet of Things platform, upload the relevant data of historical dehumidifiers to the big data cloud, thus constituting a big data source. The data source includes: indoor environmental data, outdoor environmental data, operating status data, time - area data, and water tank water level change rate.

[0054] The indoor environmental data includes sensing data such as indoor temperature, indoor humidity, and air volume at the air inlet. Such data records the indoor environment where the dehumidifier operates.

[0055] The outdoor environmental data includes sensing and meteorological data such as outdoor temperature, outdoor humidity, and outdoor weather. Such data records the outdoor environment where the dehumidifier operates.

[0056] The operating status data includes data such as water level volume, operating time, operating power, and operating wind speed during the operation of the dehumidifier. Such data records the operating status of the dehumidifier.

[0057] Time - area data: area information, mainly referring to the geographical location where the dehumidifier is located, such as: country - state; time information, mainly including data such as month and moment.

[0058] S200 adopts a hybrid feature selection method that combines the Filter and Wrapper methods. It uses the detection rate and false detection rate as the evaluation basis for feature selection, ranks the features according to their relevance using the Relief method, and then uses the SBS search strategy from back to front based on the sorted features with SVM as the classifier to screen the features of the collected dehumidifier-related data, select the optimal feature subset, and obtain the adaptive coefficient through the adaptive feature operator extraction algorithm.

[0059] The hybrid feature selection method that combines the Filter and Wrapper methods evaluates the quality of a certain feature according to the internal relationship of the data. Usually, it uses an evaluation criterion to calculate the corresponding value of each feature, compares it with a pre-set threshold, and selects the features higher than the threshold. For the specific process, see Figure 2 :

[0060] S201. In the Filter stage, calculate the Relief weights of each feature for the original features in the dehumidifier-related data.

[0061] S202. According to the calculated Relief weights of each feature, sort the original features in the dehumidifier-related data in descending order to construct a feature sequence.

[0062] S203. In the Wrapper stage, adopt the SBS search strategy to sequentially remove one feature X from the back to the front of the feature sequence, use SVM as the classifier, calculate the detection rate, false detection rate, and feature evaluation value J of the remaining feature subset. If the removal of feature X satisfies J 剔除Xk is greater than J 剔除Xk+1 (indicating that a subset of size k is found, but its performance is better than the original subset. Similarly, subsets of size k - 1, k - 2,... with better performance can also be found), then repeat S203 until J 剔除Xk is less than or equal to J 剔除Xk+1 Then determine the remaining features in the current feature sequence as the optimal feature subset, where k is the kth feature removed from the feature sequence.

[0063] Considering the influence of the usage time, motor performance fluctuations, refrigerant leakage, etc. of the dehumidifier during operation on the dehumidification performance, the dehumidification performance decays. Therefore, the adaptive feature extraction operator technology is introduced. Among them, the adaptive feature operator extraction technology takes the historical data of the above optimal feature subset of the dehumidifier as input, trains a machine learning algorithm model to obtain the adaptive feature β, and introduces it into the subsequent model input to reduce errors. The specific process includes:

[0064] Based on the historical data of the optimal feature subset, use the random forest algorithm to calculate the prediction error α of each sample i , where α i =|Yi -yβ i |, where i is the sample number, Y i is the actual water tank water level change rate of the i-th sample, y is the predicted water tank water level change rate, and β i is the adaptive coefficient of the i-th sample.

[0065] If the prediction error is greater than the set threshold, adjust the weight of the sample and retrain the prediction adaptive coefficient; if the prediction error is less than or equal to the set threshold, the training is completed and the final adaptive coefficient is obtained.

[0066] Based on the above process, examples of the optimal feature subset and adaptive coefficient are selected, as shown in Table 1.

[0067] Table 1 Optimal Feature Subset Table

[0068]

[0069] S300 Based on the optimal feature subset and adaptive coefficient, the CT-SVM algorithm is used to predict the water level of the dehumidifier.

[0070] The present invention proposes a CT (Compressive Tracking)-SVM algorithm for water level prediction. Its core idea is to first use a tree model to select the importance of features and eliminate the influence of invalid features on the model accuracy. Then use the SVM model to predict the water level. Specifically, it includes:

[0071] Adopt the compressive tracking CT technology to dynamically and real-time rank the features in the optimal feature subset based on the tree model, screen out the features with importance lower than the importance threshold, and obtain the optimal low-dimensional feature subset. The example is shown in Table 2.

[0072] Table 2 Optimal Low-Dimensional Feature Subset

[0073]

[0074] Combine the adaptive coefficient with the optimal low-dimensional feature subset to construct an enhanced input vector, input it into the non-linear support vector machine classification model, divide the water tank water level change rate into multiple types as the output, determine the volume of the water tank according to the product model, and calculate the current water volume and the estimated time to full water of the water tank.

[0075] The enhanced input vector is, for example: indoor humidity h1Humidy is denoted as x1, operating power P is denoted as x2, operating wind speed F is denoted as x3, indoor temperature t1Temp is denoted as x4, and adaptive coefficient β is denoted as x5. The enhanced input vector X = [x1, x2, x3, x4, x5].

[0076] The support vector machine (SVM) is a binary classification model. Its basic model is a linear classifier with the largest margin defined in the feature space. The largest margin makes it different from the perceptron; the SVM also includes the kernel trick, which makes it a virtually non-linear classifier. The basic idea of SVM learning is to solve the separating hyperplane that can correctly divide the training data set and has the largest geometric margin. w•x + b = 0 is the separating hyperplane. For a linearly separable data set, there are infinitely many such hyperplanes (i.e., perceptrons), but the separating hyperplane with the largest geometric margin is unique.

[0077] The present invention adopts a non-linear support vector machine classification model. For the non-linear classification problem in the input space, it is transformed into a linear classification problem in a certain-dimensional feature space through non-linear transformation, and a linear support vector machine is learned in the high-dimensional feature space. Since in the dual problem of linear support vector machine learning, the objective function and the classification decision function only involve the inner product between instances. The kernel function is used to represent the inner product between two instances after a non-linear transformation.

[0078] Input: Training data set constructed from enhanced input vectors Wherein, , , is the enhanced input vector set, n is the number of heavy features of the enhanced input vector, and N is the number of enhanced input vectors.

[0079] The objective function is:

[0080] .

[0081] w is the weight vector, β i is the adaptive coefficient of the i-th enhanced input vector, and ξ i is the slack variable of the i-th enhanced input vector.

[0082] Output: The category of the water tank water level change rate.

[0083] Select the kernel function and the penalty parameter C > 0, and construct and solve the convex quadratic programming problem:

[0084]

[0085] .

[0086] Get the optimal solution:

[0087] .

[0088] Select a component of , satisfying the condition , calculate .

[0089] Then the classification decision function is as follows:

[0090] .

[0091] The kernel function is as follows:

[0092] .

[0093] Among them, N is the number of enhanced input vectors, is the optimal solution of the Lagrange multiplier, is the class label of the i-th training sample, is the kernel function, representing the vector in the original input space and the training sample vector mapped to the high-dimensional feature space, is the intercept term of the classification hyperplane, is the class label of the j-th training sample, is the kernel function, representing the inner product of the i-th training sample and the j-th training sample in the high-dimensional feature space, is 1 / 5.

[0094] For example, a preliminary test is conducted on four products with different ability segments sold in the market. The specific product ability segments and parameter details are shown in Table 3, and the data from the laboratory test is used as the initial value of the prediction data.

[0095] Table 3 Parameter Details of the Models in the Laboratory Preliminary Test

[0096]

[0097] Indoor humidity environment: 30% - 90%, with a minimum interval threshold of 5%.

[0098] Indoor temperature environment: 5°C - 35°C, with a minimum interval threshold of 5°C.

[0099] Operating wind speed: low, medium, high, with a minimum interval threshold of one wind level.

[0100] This prediction algorithm combines feature selection and the non-linear SVM algorithm to perform CT-SVM algorithm prediction calculations on individual differences such as the use of dehumidifiers in different regions, fluctuations in the performance of compressor motors, and refrigerant leakage. Four features selected based on feature importance are used as the input of the SVM, and the change rate of the water level in the water tank is divided into multiple categories as the output to perform SVM multi-classification model prediction. The App calculates the water full time of the water tank and the current water volume in the water tank based on the change rate of the water level in the water tank predicted by the cloud big data and the volume of the water tank of the product model.

[0101] Theoretical calculations show that: 1. The MAE (Mean Absolute Error) of the water full prediction model of CT-SVM (after introducing β) adopted in the present invention is 19.6% lower than that of the linear regression model algorithm and 8.4% lower than that of the SVM model algorithm; 2. The RMAE (Root Mean Square Error) of the water full prediction model of CT-SVM is 72% lower than that of the linear regression model algorithm and 18% lower than that of the SVM model algorithm. In addition, β dynamically responds to equipment aging and environmental changes to achieve "one machine, one model"; the β value can be updated in the cloud without retraining the SVM, which is suitable for Internet of Things deployment.

[0102] See Figure 3 , which is a dehumidifier water level prediction system based on CT-SVM, including a collection module 100, a screening module 200, and a water level prediction module 300.

[0103] The collection module 100 is used to collect relevant data of historical dehumidifiers, including indoor environmental data, outdoor environmental data, operating status data, time and region data, and the change rate of the water level in the water tank.

[0104] The screening module 200 is used to adopt a hybrid feature selection method that combines the Filter and Wrapper methods, uses the detection rate and false detection rate as the evaluation basis for feature selection, ranks the features using the Relief method, and uses the SVM as the classifier to adopt the SBS search strategy from back to front according to the sorted features to screen the relevant data of the collected dehumidifiers, select the optimal feature subset, and obtain the adaptive coefficient through the adaptive feature operator extraction algorithm.

[0105] The water level prediction module 300 is used to perform water level prediction of the dehumidifier using the CT-SVM algorithm based on the optimal feature subset and the adaptive coefficient.

[0106] Among them, the selection of the optimal feature subset includes:

[0107] S201. In the Filter stage, calculate the Relief weights of each feature for the original features in the relevant data of the dehumidifier.

[0108] S202. Sort the original features in the dehumidifier-related data in descending order according to the Relief weights of each calculated feature, and construct a feature sequence.

[0109] S203. In the Wrapper stage, adopt the SBS search strategy to sequentially remove one feature X from the end of the feature sequence. Use SVM as the classifier to calculate the detection rate, false detection rate, and feature evaluation value J of the remaining feature subsets. If the condition J 剔除Xk is greater than J 剔除Xk+1 is satisfied after removing feature X, then repeat S203 until J 剔除Xk is less than or equal to J 剔除Xk+1 Then determine the remaining features in the current feature sequence as the optimal feature subset, where k is the kth feature removed from the feature sequence.

[0110] Among them, obtaining the adaptive coefficient includes:

[0111] Based on the historical data of the optimal feature subset, use the random forest algorithm to calculate the prediction error α of each sample i , where α i =|Y i -yβ i |, where i is the sample number, Y i is the actual water tank water level change rate of the ith sample, y is the predicted water tank water level change rate, and β i is the adaptive coefficient of the ith sample.

[0112] If the prediction error is greater than the set threshold, adjust the weight of the sample and retrain the prediction adaptive coefficient; if the prediction error is less than or equal to the set threshold, the training is completed, and the final adaptive coefficient is obtained.

[0113] Among them, using the CT-SVM algorithm for water level prediction of the dehumidifier includes:

[0114] Adopt the compressive sampling pursuit CT technology to dynamically and real-time rank the importance of the features in the optimal feature subset based on the tree model, and screen out the features with importance lower than the importance threshold to obtain the optimal low-dimensional feature subset.

[0115] Combine the adaptive coefficient with the optimal low-dimensional feature subset to construct an enhanced input vector, input it into the non-linear support vector machine classification model, divide the water tank water level change rate into multiple types as the output, determine the volume of the water tank according to the product model, and calculate the current water volume and the estimated time to full water of the water tank.

[0116] Among them, the classification decision function of the non-linear support vector machine classification model is:

[0117]

[0118]

[0119] 。

[0120] Among them, N is the number of enhanced input vectors, is the optimal solution of the Lagrange multiplier, is the class label of the i-th training sample, is the kernel function, representing the vector in the original input space and the training sample vector mapped to the high-dimensional feature space, is the intercept term of the classification hyperplane, is the class label of the j-th training sample, is the kernel function, representing the i-th training sample and the j-th training sample inner product in the high-dimensional feature space, is 1 / 5.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described by referring to the preferred embodiments of the present invention, those of ordinary skill in the art should understand that various changes can be made in form and details without departing from the spirit and scope of the present invention defined by the appended claims.

Claims

1. A method for predicting the water level of a dehumidifier based on CT-SVM, characterized in that, Including: Collecting relevant data of historical dehumidifiers, including indoor environmental data, outdoor environmental data, operating status data, time and location data, and the rate of change of the water tank level; Adopting a hybrid feature selection method that combines the Filter and Wrapper methods, using the detection rate and false detection rate as the evaluation basis for feature selection, using the Relief method to rank the features according to their correlation, and adopting the SBS search strategy from back to front based on the sorted features with SVM as the classifier to screen the features of the collected dehumidifier-related data, select the optimal feature subset, and obtain the adaptive coefficient through the adaptive feature operator extraction algorithm; Based on the optimal feature subset and the adaptive coefficient, using the CT-SVM algorithm to predict the water level of the dehumidifier.

2. The method for predicting the water level of a dehumidifier based on CT-SVM according to claim 1, wherein The selection of the optimal feature subset includes: S201. In the Filter stage, calculate the Relief weights of each feature for the original features in the dehumidifier-related data; S202. According to the calculated Relief weights of each feature, sort the original features in the dehumidifier-related data in descending order to construct a feature sequence; S203. In the Wrapper stage, adopt the SBS search strategy to sequentially remove one feature X from the feature sequence from back to front. Use SVM as the classifier to calculate the detection rate, false detection rate, and feature evaluation value J of the remaining feature subset. If the condition J 剔除Xk is greater than J 剔除Xk+1 , then repeat S203 until J 剔除Xk is less than or equal to J 剔除Xk+1 Then determine the remaining features in the current feature sequence as the optimal feature subset, where k is the k-th feature removed from the feature sequence.

3. The method for predicting the water level of a dehumidifier based on CT-SVM according to claim 1, characterized in that, The obtaining of the adaptive coefficient includes: Based on the historical data of the optimal feature subset, the prediction error α of each sample is calculated using the random forest algorithm i , where α i =|Y i -yβ i |, where i is the sample number, Y i is the true water tank water level change rate of the i-th sample, y is the predicted water tank water level change rate, and β i is the adaptive coefficient of the i-th sample; If the prediction error is greater than the set threshold, adjust the weight of the sample and retrain the prediction adaptive coefficient; if the prediction error is less than or equal to the set threshold, the training is completed to obtain the final adaptive coefficient.

4. The dehumidifier water level prediction method based on CT-SVM according to claim 1, characterized in that The using of the CT-SVM algorithm to predict the water level of the dehumidifier includes: Adopting the compressive tracking CT technology to dynamically and real-timely rank the features in the optimal feature subset based on the tree model, screening out the features with importance lower than the importance threshold to obtain the optimal low-dimensional feature subset; Combining the adaptive coefficient with the optimal low-dimensional feature subset to construct an enhanced input vector, inputting it into the nonlinear support vector machine classification model, dividing the rate of change of the water tank level into multiple types as the output, determining the volume of the water tank according to the product model, and calculating the current water volume of the water tank and the estimated time to full water.

5. The dehumidifier water level prediction method based on CT-SVM according to claim 4, wherein The classification decision function of the nonlinear support vector machine classification model is: ； where N is the number of enhanced input vectors, is the optimal solution of the Lagrange multiplier, is the class label of the i-th training sample, is the kernel function, representing the vector in the original input space and the training sample vector mapped to the high-dimensional feature space, is the intercept term of the classification hyperplane, is the class label of the j-th training sample, is the kernel function, representing the i-th training sample and the j-th training sample inner product in the high-dimensional feature space, is 1 / 5.

6. A dehumidifier water level prediction system based on CT-SVM, characterized in that, Including a collection module, a screening module, and a water level prediction module; The collection module is used to collect relevant data of historical dehumidifiers, including indoor environmental data, outdoor environmental data, operating status data, time and location data, and the rate of change of the water tank level; The screening module is used to adopt a hybrid feature selection method that combines the Filter and Wrapper methods, using the detection rate and false detection rate as the evaluation basis for feature selection, using the Relief method to rank the features according to their correlation, and adopting the SBS search strategy from back to front based on the sorted features with SVM as the classifier to screen the features of the collected dehumidifier-related data, select the optimal feature subset, and obtain the adaptive coefficient through the adaptive feature operator extraction algorithm; The water level prediction module is used to predict the water level of the dehumidifier based on the optimal feature subset and the adaptive coefficient by using the CT-SVM algorithm.

7. The dehumidifier water level prediction system based on CT - SVM according to claim 6, characterized in that, The selection of the optimal feature subset includes: S201. In the Filter stage, calculate the Relief weights of each feature for the original features in the dehumidifier-related data; S202. Sort the original features in the dehumidifier-related data in descending order according to the Relief weights of the calculated features, and construct a feature sequence; S203. In the Wrapper stage, adopt the SBS search strategy to sequentially remove one feature X from the feature sequence from back to front. Use SVM as the classifier to calculate the detection rate, false detection rate, and feature evaluation value J of the remaining feature subset. If the condition J 剔除Xk is greater than J 剔除Xk+1 is satisfied after removing feature X, then repeat S203 until J 剔除Xk is less than or equal to J 剔除Xk+1 At this time, determine the remaining features in the current feature sequence as the optimal feature subset, where k is the k-th feature removed from the feature sequence.

8. The dehumidifier water level prediction system based on CT-SVM according to claim 6, characterized in that, The obtaining of the adaptive coefficient includes: Based on the historical data of the optimal feature subset, the prediction error α of each sample is calculated using the random forest algorithm i , where α i =|Y i -yβ i |, where i is the sample number, Y i is the actual water tank water level change rate of the i-th sample, y is the predicted water tank water level change rate, and β i is the adaptive coefficient of the i-th sample; If the prediction error is greater than the set threshold, adjust the weight of the sample and retrain the prediction adaptive coefficient; if the prediction error is less than or equal to the set threshold, the training is completed to obtain the final adaptive coefficient.

9. The dehumidifier water level prediction system based on CT - SVM according to claim 6, wherein The water level prediction of the dehumidifier using the CT-SVM algorithm includes: Adopt the compressive tracking CT technology to dynamically and real-timely rank the features in the optimal feature subset based on the tree model, screen out the features with importance lower than the importance threshold, and obtain the optimal low-dimensional feature subset; Combine the adaptive coefficient with the optimal low-dimensional feature subset to construct an enhanced input vector, input it into the non-linear support vector machine classification model, divide the change rate of the water tank water level into multiple types as the output, determine the volume of the water tank according to the product model, and calculate the current water volume and the estimated time to full of the water tank.

10. The dehumidifier water level prediction system based on CT-SVM according to claim 9, characterized in that, The classification decision function of the non-linear support vector machine classification model is: ； where N is the number of enhanced input vectors, is the optimal solution of the Lagrange multiplier, is the class label of the i-th training sample, is the kernel function, representing the vector in the original input space and the training sample vector mapped to the high-dimensional feature space, is the intercept term of the classification hyperplane, is the class label of the j-th training sample, is the kernel function, representing the i-th training sample and the j-th training sample in the inner product of the high-dimensional feature space, is 1 / 5.

Images