Dehumidifier humidity intelligent adjustment method and system integrating width learning system and PID

Abstract

The present application relates to the technical field of dehumidifier intelligent control, and proposes a dehumidifier humidity intelligent adjustment method and system integrating a width learning system and a PID, the dehumidifier humidity intelligent adjustment method comprising: acquiring real-time monitoring data of a dehumidifier operating environment; constructing a multi-kernel extreme learning machine automatic encoder, performing feature conversion through a kernel function, and generating a mapping feature matrix; constructing enhanced nodes of a width learning system based on the matrix and integrating them to construct a humidity intelligent prediction model, thereby realizing accurate prediction of the environment humidity at a future time; calculating the control deviation of the predicted value and a target humidity value, inputting the control deviation into a PID controller, and adjusting the frequency of a dehumidifier compressor in real time.The present application extracts data features through a multi-kernel extreme learning machine, combines the rapid modeling capability of a width learning system with the stability of PID control, significantly improves the humidity adjustment accuracy and response speed of a dehumidifier in a complex environment, and realizes intelligent and efficient control of the dehumidification process.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for dehumidifiers, and in particular to an intelligent humidity adjustment method and system for dehumidifiers that integrates a wide-range learning system and a PID controller. Background Technology

[0002] Currently, the humidity control field for dehumidifiers mainly employs three modes: start-stop control, PID control, and intelligent control. Start-stop control triggers the device's on / off state by setting a humidity threshold, causing the ambient humidity to oscillate repeatedly around the target value, making it difficult to maintain a stable humidity within the ideal range. While PID control is simple in structure and highly adaptable, it is essentially a "post-correction" logic, relying on the measurement deviation at the current moment to adjust the output, resulting in a lag in response to dynamic changes in environmental parameters. Intelligent control methods based on fuzzy control or deep neural networks, although possessing predictive capabilities, have complex model structures and high computational costs, making them difficult to meet real-time control requirements. These control methods all have shortcomings in control accuracy, dynamic adaptability, and real-time response, limiting the application of dehumidifiers in precise environmental control scenarios. Summary of the Invention

[0003] To address the aforementioned shortcomings, the present invention aims to propose a dehumidifier intelligent humidity control method and system that integrates a width learning system and a PID controller. The method aims to predict the ambient humidity in the future by constructing a humidity prediction model based on a multi-core extreme learning machine autoencoder using a width learning system, and to adjust the compressor frequency in advance based on the prediction deviation using a PID controller, thereby achieving forward-looking control of the ambient humidity and improving the response speed and stability of humidity control.

[0004] To achieve this objective, the present invention adopts the following technical solution: A dehumidifier intelligent humidity control method integrating a width learning system and PID control includes the following steps: S1: Obtain real-time monitoring data of the dehumidifier's operating environment, including ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and compressor frequency; S2: Construct a multi-kernel extreme learning machine autoencoder, take the real-time monitoring data as input, perform feature transformation through multiple kernel functions, and generate a mapping feature matrix with nonlinear prediction capability; S3: Construct enhanced nodes for the width learning system based on the mapping feature matrix, and integrate the mapping feature matrix with the enhanced nodes to build a humidity intelligent prediction model. Predict the ambient humidity at future times using the humidity intelligent prediction model to obtain the humidity prediction value. S4: Calculate the control deviation between the predicted humidity value and the target humidity value, and input the control deviation to the PID controller, which then adjusts the compressor frequency of the dehumidifier in real time based on the control deviation.

[0005] Preferably, step S2 includes: The real-time monitoring data is used to construct the model input matrix. ,in express The parameters measured at each time point, and , , , and These represent ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and dehumidifier compressor frequency, respectively. Construct an extreme learning machine autoencoder and input the model into a matrix. As input and output, establish the hidden layer output matrix. With output weights The relationship between them satisfies the following relation: ; Among them, the hidden layer output matrix , This represents the output weights of the automatic encoder in the Extreme Learning Machine; Define activation function The following relation is satisfied: ; in, Indicates the size is The input weight matrix, Indicates the size is The input bias vector, This represents the activation function. Indicates the number of hidden layer nodes. Indicates the input variables of the function; By regularization parameters Solve for output weights The following relation is satisfied: ; in, Represents the identity matrix. This represents the regularization parameter.

[0006] Preferably, step S2 includes: Define a multi-core matrix Output weights Substitute the solution formula into the model input matrix The expression satisfies the following relation: ; in, Represents the hidden layer output matrix Its product with its transpose Represents the transformation matrix; Transformation matrix The expression satisfies the relation: ; in, Represents the identity matrix. Represents the regularization parameter. Represents the model input matrix; Multi-core matrix The matrix elements are designed by combining multiple kernel functions, satisfying the following relation: ; in, Represents a multi-core matrix The Line 1 Column elements, This represents the preset coefficients in the interval [0,1]. Represents the radial basis kernel function. Represents a linear kernel function; Radial basis kernel function With linear kernel function Satisfying the relation: ; in, Represents the natural constant. This represents the bandwidth of the kernel function. and They represent the first Time and the The vector of measured parameters at time; Obtain the transformation matrix Then, the mapping feature matrix The calculation satisfies the following relation: ; in, This represents the activation function. Represents the transformation matrix The transpose of .

[0007] Preferably, step S3 includes: Using the mapping feature matrix Calculate the first Group Enhanced Node Matrix The following relation is satisfied: ; in, Indicates the first Group enhancement node matrix, Represents the mapping feature matrix, Indicates the interval The first one is randomly generated in the middle. Group input weights, Indicates the interval The first one is randomly generated in the middle. Group input bias. This represents the activation function. Indicates the number of augmented node groups; Integrate all the enhanced node groups to obtain the enhanced node matrix. The following relation is satisfied: ; in, Represents an enhanced node matrix. They represent the first Group to No. Group-enhanced node matrix.

[0008] Preferably, step S3 includes: Map the feature matrix and enhanced node matrix Integration, constructing a joint matrix The following relation is satisfied: ; in, Represents the mapping feature matrix, Represents an enhanced node matrix; The objective function is introduced with the goal of minimizing the prediction error, satisfying the following relationship: ; in, This represents the output weights of the width learning system. This represents the output matrix of the width learning system. Represents the regularization parameter; By differentiating the objective function and setting the derivative to zero, the output weights can be obtained. The following relation is satisfied: ; in, Representing the joint matrix transpose, Represents the identity matrix. This represents the output matrix of the width learning system; Using the obtained output weights right Humidity of the external environment at any time Make a prediction that satisfies the following relationship: ; in, express Predicted humidity value at any given time. express The joint matrix at time points, ,and ,as well as ; in, express The mapping feature vector at time step, express The augmented node vector at time step, express The vector of measured parameters at time, This represents the transpose of the transformation matrix. This represents the activation function.

[0009] Preferably, step S4 includes: calculate Control deviation at time The following relation is satisfied: ; in, express Control deviation at any time This indicates the set target humidity value. express Predicted humidity value at any given time; Calculated based on PID control theory Input frequency of the dehumidifier compressor The following relation is satisfied: ; in, express The input frequency of the dehumidifier compressor at all times. Represents the proportionality coefficient. Represents the integral coefficient. Denotes the differential coefficient. Indicates the first Control deviation at any time Indicates the sampling period, i.e., the period from which the sensor... Time's up The time interval between sampling moments.

[0010] The dehumidifier's intelligent humidity control system integrates a width learning system and a PID controller, including: The data acquisition module is used to acquire real-time monitoring data of the dehumidifier's operating environment, including ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and compressor frequency. The feature transformation module is used to construct a multi-kernel extreme learning machine autoencoder. It takes the real-time monitoring data as input, performs feature transformation through multiple kernel functions, and generates a mapping feature matrix with nonlinear prediction capabilities. The humidity prediction module is used to construct the enhancement nodes of the width learning system based on the mapping feature matrix, and integrate the mapping feature matrix with the enhancement nodes to build a humidity intelligent prediction model. The humidity intelligent prediction model is used to predict the environmental humidity at future times to obtain the humidity prediction value. The control and adjustment module is used to calculate the control deviation between the predicted humidity value and the target humidity value, and input the control deviation to the PID controller, which adjusts the compressor frequency of the dehumidifier in real time based on the control deviation.

[0011] One of the above technical solutions has the following advantages or beneficial effects: This invention comprehensively captures the physical factors affecting humidity changes by acquiring real-time monitoring data, including ambient humidity, temperature, air pressure, wind speed, and compressor frequency. It then constructs a multi-core extreme learning machine autoencoder, using multiple kernel functions to perform nonlinear feature transformation on the input data, generating a mapping feature matrix with strong representational capabilities. This overcomes the unstable modeling performance problem caused by traditional random mapping. Subsequently, based on this mapping feature matrix, it constructs enhancement nodes for a wide learning system, integrating the mapping features with the enhancement nodes to build an intelligent humidity prediction model. This model rapidly learns the patterns of ambient humidity changes in a data-driven manner, achieving accurate prediction of ambient humidity at future times, thus solving the problems of complex modeling and poor real-time performance in traditional intelligent algorithms. Finally, it calculates the control deviation between the predicted humidity value and the target humidity value, and inputs this deviation into a PID controller. The PID controller adjusts the compressor frequency in advance based on the predicted deviation at future times, transforming the control action from traditional "post-event correction" to "pre-event correction." Therefore, this scheme enhances the model's adaptability to complex environmental conditions by extracting deep nonlinear features through a multi-core extreme learning machine autoencoder, and ensures the real-time response capability of the prediction model by leveraging the rapid training characteristics of the wide learning system. This enables the PID controller to make forward-looking adjustments based on future humidity deviations, effectively suppressing humidity oscillations and improving the adaptability of the control system to dynamic environmental changes and the stability of humidity maintenance. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0013] Figure 1 This is a flowchart of the intelligent humidity control method for dehumidifiers that integrates a width learning system and a PID controller, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the intelligent humidity control system for dehumidifiers that integrates a width learning system and a PID controller, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of the Extreme Learning Machine automatic encoder provided in an embodiment of the present invention. Detailed Implementation

[0014] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0015] In this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0016] A dehumidifier intelligent humidity control method integrating a width learning system and PID control, such as... Figure 1 As shown, a preferred embodiment of the present invention includes the following steps: S1: Obtain real-time monitoring data of the dehumidifier's operating environment, including ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and compressor frequency; It should be noted that ambient humidity refers to the water vapor content in the air, usually expressed as relative humidity. In step S1, it serves as the core target parameter for the dehumidification task, directly reflecting the current humidity level and providing fundamental data for subsequent humidity prediction. Ambient temperature refers to the air temperature, and in step S1, it is a key parameter affecting air moisture retention capacity and heat exchanger efficiency. Ambient air pressure refers to atmospheric pressure, and in step S1, it is an important parameter affecting air density and humid air quality. Ambient wind speed refers to the speed of airflow, and in step S1, it is a parameter affecting the amount of air flowing through the evaporator per unit time. Compressor frequency refers to the operating frequency of the dehumidifier compressor, and in step S1, it is a control parameter reflecting the current system's active dehumidification capacity. Real-time monitoring data refers to the data of the above five parameters continuously collected by sensors during operation, and in step S1, it serves as the input basis for the entire intelligent adjustment method.

[0017] Understandably, by acquiring five real-time monitoring data points—ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and compressor frequency—the system comprehensively captures the physical factors affecting the dehumidification process and future humidity changes, thereby achieving accurate perception of future humidity trends and providing complete data support for subsequent nonlinear feature learning and humidity prediction.

[0018] Specifically, during dehumidifier operation, humidity sensors, temperature sensors, air pressure sensors, wind speed sensors, and a compressor frequency detection device are deployed to continuously monitor the dehumidifier's operating environment according to a preset sampling period. The humidity sensor is installed at the dehumidifier's air inlet to collect ambient humidity; the temperature sensor is arranged adjacent to the humidity sensor to synchronously collect ambient temperature; the air pressure sensor is placed in the external ventilation area of ​​the equipment to collect ambient air pressure; the wind speed sensor is placed along the airflow path to collect ambient wind speed; and the compressor frequency detection device obtains the current compressor frequency by reading the output signal of the compressor driver. These five sensors convert the collected analog signals into digital signals, which are then transmitted to the controller via wired or wireless communication. The controller then integrates the monitoring data from each moment into model input data.

[0019] S2: Construct a multi-kernel extreme learning machine autoencoder, take the real-time monitoring data as input, perform feature transformation through multiple kernel functions, and generate a mapping feature matrix with nonlinear prediction capability; It should be noted that the multi-kernel extreme learning machine autoencoder refers to a neural network model that combines extreme learning machine theory with the structure of an autoencoder. The extreme learning machine refers to a single-layer feedforward neural network that randomly generates input weights and biases and solves for the output weights analytically. The autoencoder refers to a neural network structure that learns features using the input data itself as the output target, and is used in step S2 to extract deep nonlinear features from real-time monitoring data. The kernel function is a function that maps the original data to a high-dimensional feature space, and is used in step S2 to enhance the model's ability to capture nonlinear relationships. The mapped feature matrix is ​​the high-dimensional feature representation matrix obtained after feature transformation, and is used in step S2 as the nonlinear feature expression of the original monitoring data for the subsequent construction of the humidity prediction model. Feature transformation refers to the process of converting the original data into a new feature representation with greater representational power through mathematical transformation, and is used in step S2 to improve the separability and predictive ability of the data.

[0020] It is understandable that by constructing a multi-kernel extreme learning machine autoencoder, real-time monitoring data can be used as both input and output for feature learning. By leveraging the rapid training characteristics of extreme learning machines combined with the feature extraction capabilities of autoencoders, and then using multiple kernel functions to perform nonlinear mapping transformation on the hidden layer output, stable mapping features with strong nonlinear predictive capabilities can be extracted from the original monitoring data.

[0021] S3: Construct enhanced nodes for the width learning system based on the mapping feature matrix, and integrate the mapping feature matrix with the enhanced nodes to build a humidity intelligent prediction model. Predict the ambient humidity at future times using the humidity intelligent prediction model to obtain the humidity prediction value. It should be noted that the width learning system refers to a flattened neural network structure based on a random vector function connection network. It constructs a model through the lateral expansion of mapped features and augmenting nodes, and in step S3, it is used to quickly learn data patterns and achieve humidity prediction. Augmenting nodes are nodes obtained in the width learning system after performing a nonlinear transformation on the mapped features using random weights and biases. In step S3, they are used to further expand the feature space and improve the model's nonlinear fitting ability. The intelligent humidity prediction model is a mathematical model built based on the width learning system for predicting future environmental humidity. In step S3, it estimates the humidity at future times by learning the input-output relationship from historical data. The predicted humidity value is the estimated environmental humidity at future times output by the intelligent humidity prediction model, and in step S3, it serves as the feedforward input for PID control.

[0022] It is understandable that by constructing enhanced nodes based on the mapping feature matrix, and integrating the mapping features with the enhanced nodes to build a wide learning system, the output weights are solved with the goal of minimizing the prediction error, so as to achieve fast and accurate prediction of environmental humidity at future times. This overcomes the problems of high modeling complexity and slow training speed of traditional neural networks, and provides forward-looking humidity information for PID control.

[0023] S4: Calculate the control deviation between the predicted humidity value and the target humidity value, and input the control deviation to the PID controller, which then adjusts the compressor frequency of the dehumidifier in real time based on the control deviation.

[0024] It should be noted that the target humidity value refers to the pre-set desired ambient humidity level, which serves as the reference benchmark for humidity control in step S4. The control deviation is the difference between the predicted humidity value and the target humidity value, which serves as the input signal to the PID controller in step S4, reflecting the degree to which the humidity deviates from the target at future times. The PID controller is a feedback controller based on a combination of proportional, integral, and derivative control actions. In step S4, it is used to calculate the control quantity based on the control deviation to adjust the compressor frequency. The proportional coefficient is the gain parameter of the proportional term in the PID controller, used in step S4 to generate immediate control action based on the current deviation magnitude. The integral coefficient is the gain parameter of the integral term in the PID controller, used in step S4 to eliminate accumulated deviation. The derivative coefficient is the gain parameter of the derivative term in the PID controller, used in step S4 to predict the deviation change trend and suppress overshoot in advance. The sampling period is the time interval between two adjacent sensor data acquisitions, used as the time benchmark for the discretization calculation of PID control in step S4.

[0025] Understandably, by calculating the control deviation between the predicted humidity value and the target humidity value, the humidity deviation information at future moments is input into the PID controller. The PID controller then calculates the compressor frequency adjustment based on the three control actions: proportional, integral, and derivative. This allows for control actions to be taken in advance before the humidity deviation occurs, thereby transforming traditional post-event correction into pre-event correction, suppressing humidity oscillations, and improving the dynamic adaptability of the control system.

[0026] Preferably, step S2 includes: The real-time monitoring data is used to construct the model input matrix. ,in express The parameters measured at each time point, and , , , and These represent ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and dehumidifier compressor frequency, respectively. Construct an extreme learning machine autoencoder and input the model into a matrix. As input and output, establish the hidden layer output matrix. With output weights The relationship between them satisfies the following relation: ; Among them, the hidden layer output matrix , This represents the output weights of the automatic encoder in the Extreme Learning Machine; Define activation function The following relation is satisfied: ; in, Indicates the size is The input weight matrix, Indicates the size is The input bias vector, This represents the activation function. Indicates the number of hidden layer nodes. Indicates the input variables of the function; By regularization parameters Solve for output weights The following relation is satisfied: ; in, Represents the identity matrix. This represents the regularization parameter.

[0027] It should be noted that the model input matrix This refers to a matrix formed by arranging real-time monitoring data from multiple time points in a time series, where each row corresponds to five parameters at a sampling time. In step S2, this matrix serves as the input and output of the Extreme Learning Machine (ELM) autoencoder. The ELM autoencoder is a neural network combining the structures of an ELM and an autoencoder. An ELM is a single-layer feedforward neural network that randomly generates input weights and biases and directly solves for the output weights using analytical methods. An autoencoder is a network structure that uses the input data itself as the output target for unsupervised feature learning. In step S2, this autoencoder is used to learn mapped features from the real-time monitoring data. Hidden layer output matrix. This refers to the matrix formed by the outputs of the hidden layer of the Extreme Learning Machine's autoencoder for all input samples, used in step S2 to establish the mapping relationship between the input data and the output weights. Output weights This refers to the connection weight matrix from the hidden layer to the output layer of the autoencoder in the Extreme Learning Machine. It is obtained analytically in step S2 and is a key parameter for feature learning. (Input weight matrix) This refers to the weight matrix connecting the input layer and the hidden layer, which is randomly generated in the interval (0,1) in step S2 and used to perform a linear transformation on the input data. Input bias vector. This refers to the bias term vector of the hidden layer neurons, which is randomly generated in the interval (0,1) in step S2 and used to translate the result of the linear transformation. Activation function This refers to a function that weights neurons and applies a nonlinear transformation; in step S2, the Sigmoid function is used. This is used to introduce non-linear expressive power. Number of hidden layer nodes. This refers to the number of neurons in the hidden layer of the autoencoder in the Extreme Learning Machine, which is used as a hyperparameter of the network structure in step S2. Regularization parameter. These are constraint parameters used to prevent model overfitting, and are used in step S2 to balance training error and model complexity. Identity matrix This refers to a square matrix with 1s on the main diagonal and 0s on the rest, which is used for regularization calculation in step S2.

[0028] Understandably, by arranging real-time monitoring data in a time series to form a model input matrix, and constructing an extreme learning machine autoencoder that uses this matrix as both input and output, the input data is linearly transformed using randomly generated input weight matrices and input bias vectors. Then, nonlinearity is introduced through an activation function to obtain the hidden layer output matrix. Finally, by introducing regularization parameters to solve for the output weights, stable feature representations can be extracted from the original monitoring data. This overcomes the problem of unstable modeling performance caused by random mapping in traditional wide learning systems and lays the foundation for subsequent nonlinear feature transformation.

[0029] Specifically, the real-time monitoring data obtained in step S1 is first arranged in chronological order of sampling time to form a data set containing... Model input matrix for data at time point , of which OK The ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and compressor frequency at that time were respectively considered. It iterates over index variables, and their value range is usually [value range missing]. Then, construct the Extreme Learning Machine autoencoder, such as... Figure 3 The input and output layers are each set to 5 nodes, and the hidden layer has 1 node. Randomly generate a 5× size within the interval (0,1). Input weight matrix and size ×1 input bias vector For the input vector at each time step Calculate the hidden layer output ,in The activation function is sigmoid. The hidden layer outputs at all time points are arranged row-wise to form the hidden layer output matrix. To avoid the numerical stability problem caused by direct inversion, a regularization method is used to solve for the output weights, and a regularization parameter is set. Through formula Calculate the output weights, where for The identity matrix. This solution process directly obtains the global optimum through analytical calculation, without the need for iterative training.

[0030] Preferably, step S2 includes: Define a multi-core matrix Output weights Substitute the solution formula into the model input matrix The expression satisfies the following relation: ; in, Represents the hidden layer output matrix Its product with its transpose, Represents the transformation matrix; Transformation matrix The expression satisfies the relation: ; in, Represents the identity matrix. Represents the regularization parameter. Represents the model input matrix; Multi-core matrix The matrix elements are designed by combining multiple kernel functions, satisfying the following relation: ; in, Represents a multi-core matrix The Line 1 Column elements, This represents the preset coefficients in the interval [0,1]. Represents the radial basis kernel function. Represents a linear kernel function; Radial basis kernel function With linear kernel function Satisfying the relation: ; in, Represents the natural constant. This represents the bandwidth of the kernel function. and They represent the first Time and the The vector of measured parameters at time; Obtain the transformation matrix Then, the mapping feature matrix The calculation satisfies the following relation: ; in, This represents the activation function. Represents the transformation matrix The transpose of .

[0031] It should be noted that multi-core matrix This refers to a matrix composed of multiple kernel functions, used in step S2 to replace the inner product form of the hidden layer output matrix, thus achieving kernelized feature mapping. Transformation matrix This refers to the weight matrix that maps the original input data to the feature space, which is obtained analytically in step S2 through a multi-kernel matrix and regularization parameters. Radial basis function. This refers to a kernel function defined based on the Gaussian function, which measures similarity by calculating the exponential decay of the Euclidean distance between samples. In step S2, it is used to capture nonlinear local relationships between data. (Linear kernel function) This refers to the kernel function that directly calculates the inner product between samples, used in step S2 to preserve the global linear structure of the data. Preset coefficients. This refers to the parameter used to balance the contribution ratio of the radial basis function kernel function and the linear kernel function. In step S2, it is set to 0.5 to ensure that both kernel functions contribute equally. The bandwidth of the kernel function... This refers to the width parameter of the Gaussian function in the radial basis function kernel, which is set to 0.2 in step S2 to control the local effective range of the kernel function. Measurement parameter vector and They represent the first Time and the The vector formed by the five monitoring parameters at each time point is used as the input for the kernel function calculation in step S2. Mapping feature matrix This refers to the final feature representation matrix obtained after multi-kernel transformation and activation function processing, which serves as the input feature of the width learning system in step S2.

[0032] Understandably, by substituting the output weight solution formula into the model input matrix expression, the relationship between the multi-kernel matrix and the transformation matrix is ​​derived. The advantages of using the radial basis kernel function to capture the nonlinear local features of the data and the linear kernel function to preserve the global structure are utilized. The multi-kernel matrix is ​​constructed by weighting the data according to preset coefficients, and then the transformation matrix is ​​solved analytically through regularization. Finally, an activation function is applied to the product of the original input matrix and the transpose of the transformation matrix to generate the mapping feature matrix. This achieves the goal of enhancing the model's nonlinear learning ability while maintaining computational stability, thereby improving the model's adaptability to complex environmental conditions.

[0033] Specifically, firstly, based on the hidden layer output matrix obtained in step S2... ,calculate Obtain the inner product matrix of the hidden layer output. Then design the multi-core matrix. , its first Line 1 Column elements are Calculation, where preset coefficients Set to 0.5, radial basis kernel function ,bandwidth Taking 0.2, linear kernel function This multi-core design allows the matrix to simultaneously contain information about the linear correlation and non-linear similarity of the data. Then, regularization parameters are used... Through formula Solve the transformation matrix ,in for identity matrix Input the model matrix. After obtaining the transformation matrix, calculate... and The product of these terms is then applied element-wise with the Sigmoid activation function. The mapping feature matrix is ​​obtained. .

[0034] Preferably, step S3 includes: Using the mapping feature matrix Calculate the first Group Enhanced Node Matrix The following relation is satisfied: ; in, Indicates the first Group enhancement node matrix, Represents the mapping feature matrix, Indicates the interval The first one is randomly generated in the middle. Group input weights, Indicates the interval The first one is randomly generated in the middle. Group input bias. This represents the activation function. Indicates the number of augmented node groups; Integrate all the enhanced node groups to obtain the enhanced node matrix. The following relation is satisfied: ; in, Represents an enhanced node matrix. They represent the first Group to No. Group-enhanced node matrix.

[0035] It should be noted that the enhanced node matrix This refers to a matrix composed of multiple sets of enhancement nodes horizontally concatenated, used in step S3 to expand the feature space of the width learning system and improve the model's nonlinear fitting ability. Group Enhanced Node Matrix This refers to the node matrix obtained after performing the i-th random nonlinear transformation on the mapping feature matrix, which serves as the basic building block of the enhanced nodes in step S3. Group input weights This refers to the connection mapping feature and the first The weight matrix of the group of enhanced nodes is randomly generated in the interval (0,1) in step S3, and is used to linearly combine the mapping features. Group input bias It refers to the first The bias term matrix of the augmentation nodes is randomly generated in the interval (0,1) in step S3 and is used to shift the linear combination result. The number of augmentation node groups k refers to the number of groups of augmentation nodes, which is used as a structural parameter of the width learning system in step S3 to determine the size of the augmentation nodes.

[0036] It is understandable that by utilizing the mapping feature matrix and performing multiple nonlinear transformations on it using randomly generated input weights and biases, multiple sets of enhanced node matrices are generated. Then, all enhanced node sets are horizontally integrated into a complete enhanced node matrix, thereby expanding the feature space dimension and enhancing the nonlinear expressive power of the model, thus improving the ability of the wide learning system to fit complex humidity change patterns.

[0037] Specifically, firstly, based on the mapping feature matrix obtained in step S2... Set the number of enhanced node groups For each group Randomly generate input weights within the interval (0,1). and input bias ,calculate and The matrix product plus the bias Then, the Sigmoid activation function is applied to each element of the result. , obtained the Group Enhanced Node Matrix Generate sequentially in this manner. Group the enhanced node matrix, and then to Horizontally concatenate along the column direction to form an enhanced node matrix. Each group of augmentation nodes performs an independent nonlinear transformation on the mapped features using different random weights and biases. The augmentation nodes in each group are different from each other and together constitute a high-dimensional augmentation feature space.

[0038] Preferably, step S3 includes: Map the feature matrix and enhanced node matrix Integration, constructing a joint matrix The following relation is satisfied: ; in, Represents the mapping feature matrix, Represents an enhanced node matrix; The objective function is introduced with the goal of minimizing the prediction error, satisfying the following relationship: ; in, This represents the output weights of the width learning system. This represents the output matrix of the width learning system. Represents the regularization parameter; By differentiating the objective function and setting the derivative to zero, the output weights can be obtained. The following relation is satisfied: ; in, Representing the joint matrix transpose, Represents the identity matrix. This represents the output matrix of the width learning system; Using the obtained output weights right Humidity of the external environment at any time Make a prediction that satisfies the following relationship: ; in, express Predicted humidity value at any given time. express The joint matrix at time points, ,and ,as well as ; in, express The mapping feature vector at time step, express The augmented node vector at time step, express The vector of measured parameters at time, This represents the transpose of the transformation matrix. This represents the activation function.

[0039] It should be noted that the joint matrix This refers to the matrix formed by horizontally concatenating the mapping feature matrix and the augmentation node matrix, which serves as the input feature matrix for the width learning system in step S3. The output weights of the width learning system... This refers to the weight vector connecting the joint matrix and the output layer, obtained analytically in step S3, and is a core parameter of the humidity prediction model. (Wide learning system output matrix) This refers to a column vector composed of environmental humidity values ​​corresponding to historical moments, which serves as the labels for supervised learning in step S3. Regularization parameters. This refers to the parameter used to constrain the norm of the output weights, which takes a value in step S3. This is used to prevent the model from overfitting. Joint matrix at time It refers to the The row vector formed by concatenating the mapping feature vector and the enhanced node vector at each time step is used in step S3 to predict the humidity at the next time step. Mapping feature vector at time step This refers to the transformation matrix. The vector obtained after feature transformation of the time-measured parameters is used as the mapping feature input in step S3 for prediction. Time-based augmentation node vector It refers to The vector obtained after performing enhancement transformation on the time-mapping feature vector is used as the enhancement feature input during prediction in step S3.

[0040] Understandably, by integrating the mapping feature matrix and the enhancement node matrix to construct a joint matrix, and introducing an objective function with a regularization term with the goal of minimizing the prediction error, the derivative of the objective function is calculated and the output weights are analytically solved by setting the derivative to zero. Then, the output weights are combined with the mapping feature vector and the enhancement node vector at the current time to calculate the humidity prediction value at the next time. This enables the learning of humidity change patterns from historical data and the rapid prediction of future humidity, thus overcoming the problem of long iterative training time in traditional neural networks and improving the real-time performance and accuracy of humidity prediction.

[0041] Specifically, firstly, the mapping feature matrix obtained in step S2 is... and the enhanced node matrix obtained in step S3 Construct a joint matrix by horizontally concatenating the columns. Then, the ambient humidity from historical data is used as the output matrix. Construct the objective function The regularization parameter The objective function's first term measures the prediction error, and the second term constrains the weight norm. By taking the derivative and setting it to zero, we obtain the analytical solution. ,in A unit matrix of appropriate dimensions. During the prediction phase, when... Measurement parameter vector at time Then, first through calculate The mapping feature vector at time step, where This is the transpose of the transformation matrix obtained in step S2; then through... Calculate the augmented node vector at time t, where and The weights and biases are randomly generated in step S3; finally, and spliced ​​as ,pass calculate Predicted humidity value at any given time.

[0042] Preferably, step S4 includes: calculate Control deviation at time The following relation is satisfied: ; in, express Control deviation at any time This indicates the set target humidity value. express Predicted humidity value at any given time; Calculated based on PID control theory Input frequency of the dehumidifier compressor The following relation is satisfied: ; in, express The input frequency of the dehumidifier compressor at all times. This represents the proportionality coefficient. Represents the integral coefficient. Represents the differential coefficient. Indicates the first Control deviation at any time Indicates the sampling period, i.e., the period from which the sensor... Time's up The time interval between sampling moments.

[0043] It should be noted that the target humidity value This refers to the pre-set desired ambient humidity level, which serves as the reference benchmark for humidity control in step S4. Control deviation This refers to the difference between the target humidity value and the predicted humidity value, which is used as the input signal for the PID controller in step S4. Proportional coefficient. This refers to the gain parameter of the proportional term in the PID controller, which is set to 50 in step S4 and used to generate immediate control action based on the current deviation. Integral coefficient. This refers to the gain parameter of the integral term of the PID controller, which is used in step S4 to generate continuous control action based on the accumulation of historical deviations. (Derivative coefficient) This refers to the gain parameter of the derivative term of the PID controller, which takes a value of 0.1 in step S4 and is used to generate a lead control action based on the rate of change of the deviation. Sampling period. This refers to the time interval between two adjacent sensor data acquisitions, which serves as the time reference for PID discretization calculation in step S4.

[0044] Understandably, by calculating the control deviation between the target humidity value and the predicted humidity value, the deviation information at future moments is input into the PID controller. The PID controller responds to the current deviation with the proportional term, eliminates the accumulated deviation with the integral term, and suppresses the changing trend with the derivative term. It then comprehensively calculates the compressor frequency adjustment, thereby enabling control adjustment before the actual humidity deviation occurs. This achieves the goal of transforming traditional post-event correction into pre-event correction, suppressing humidity overshoot and oscillation, and improving the dynamic response capability of the control system.

[0045] Specifically, firstly, based on the results obtained in step S3... Predicted humidity values ​​at any time Compared with the preset target humidity value Perform the difference operation to obtain the control deviation. The control deviation is then input into the PID controller, which calculates the compressor input frequency according to the three control actions. The proportional term is... The proportionality coefficient This term generates an instantaneous control action proportional to the magnitude of the current control deviation; the integral term is... This term generates a control action based on the cumulative value of all control deviations from the initial time to the current time, used to eliminate the steady-state deviation of the system; the differential term is... The differential coefficients This term generates a control action opposite to the trend of change based on the rate of change of the control deviation, used to suppress potential overshoot and oscillations. Adding the proportional, integral, and derivative terms yields... compressor input frequency at any time The controller sends the frequency command to the compressor driver, which adjusts the compressor operating frequency in real time, changes the refrigerant flow rate and the evaporator dehumidification capacity, so that the future ambient humidity approaches the target humidity value.

[0046] A dehumidifier's intelligent humidity control system that integrates a width learning system and a PID controller, such as... Figure 2 As shown, it includes: The data acquisition module is used to acquire real-time monitoring data of the dehumidifier's operating environment, including ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and compressor frequency. The feature transformation module is used to construct a multi-kernel extreme learning machine autoencoder. It takes the real-time monitoring data as input, performs feature transformation through multiple kernel functions, and generates a mapping feature matrix with nonlinear prediction capabilities. The humidity prediction module is used to construct the enhancement nodes of the width learning system based on the mapping feature matrix, and integrate the mapping feature matrix with the enhancement nodes to build a humidity intelligent prediction model. The humidity intelligent prediction model is used to predict the environmental humidity at future times to obtain the humidity prediction value. The control and adjustment module is used to calculate the control deviation between the predicted humidity value and the target humidity value, and input the control deviation to the PID controller, which adjusts the compressor frequency of the dehumidifier in real time based on the control deviation.

[0047] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0048] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A dehumidifier intelligent humidity control method integrating a width learning system and PID control, characterized in that, Includes the following steps: S1: Obtain real-time monitoring data of the dehumidifier's operating environment, including ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and compressor frequency; S2: Construct a multi-kernel extreme learning machine autoencoder, take the real-time monitoring data as input, perform feature transformation through multiple kernel functions, and generate a mapping feature matrix with nonlinear prediction capability; S3: Construct enhanced nodes for the width learning system based on the mapping feature matrix, and integrate the mapping feature matrix with the enhanced nodes to build a humidity intelligent prediction model. Predict the ambient humidity at future times using the humidity intelligent prediction model to obtain the humidity prediction value. S4: Calculate the control deviation between the predicted humidity value and the target humidity value, and input the control deviation to the PID controller, which then adjusts the compressor frequency of the dehumidifier in real time based on the control deviation.

2. The intelligent humidity control method for dehumidifiers integrating a width learning system and PID control according to claim 1, characterized in that, Step S2 includes: The real-time monitoring data is used to construct the model input matrix. ,in express The parameters measured at each time point, and , , , and These represent ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and dehumidifier compressor frequency, respectively. Construct an extreme learning machine autoencoder and input the model into a matrix. As input and output, establish the hidden layer output matrix. With output weights The relationship between them satisfies the following relation: ; Among them, the hidden layer output matrix , This represents the output weights of the automatic encoder in the Extreme Learning Machine; Define activation function The following relation is satisfied: ； in, Indicates the size is The input weight matrix, Indicates the size is The input bias vector, This represents the activation function. Indicates the number of hidden layer nodes. Indicates the input variables of the function; By regularization parameters Solve for output weights The following relation is satisfied: ； in, Represents the identity matrix. This represents the regularization parameter.

3. The intelligent humidity control method for dehumidifiers integrating a width learning system and PID according to claim 2, characterized in that, Step S2 includes: Define a multi-core matrix Output weights Substitute the solution formula into the model input matrix The expression satisfies the following relation: ； in, Represents the hidden layer output matrix Its product with its transpose Represents the transformation matrix; Transformation matrix The expression satisfies the relation: ； in, Represents the identity matrix. Represents the regularization parameter. Represents the model input matrix; Multi-core matrix The matrix elements are designed by combining multiple kernel functions, satisfying the following relation: ； in, Represents a multi-core matrix The Line 1 Column elements, This represents the preset coefficients in the interval [0,1]. Represents the radial basis kernel function. Represents a linear kernel function; Radial basis kernel function With linear kernel function Satisfying the relation: ； in, Represents the natural constant. This represents the bandwidth of the kernel function. and They represent the first Time and the The vector of measured parameters at time; Obtain the transformation matrix Then, the mapping feature matrix The calculation satisfies the following relation: ； in, This represents the activation function. Represents the transformation matrix The transpose of .

4. The intelligent humidity control method for dehumidifiers integrating a width learning system and PID according to claim 1, characterized in that, Step S3 includes: Using the mapping feature matrix Calculate the first Group Enhanced Node Matrix The following relation is satisfied: ； in, Indicates the first Group enhancement node matrix, Represents the mapping feature matrix, Indicates the interval The first one is randomly generated in the middle. Group input weights, Indicates the interval The first one is randomly generated in the middle. Group input bias. This represents the activation function. Indicates the number of augmented node groups; Integrate all the enhanced node groups to obtain the enhanced node matrix. The following relation is satisfied: ； in, Represents an enhanced node matrix. They represent the first Group to No. Group-enhanced node matrix.

5. The intelligent humidity control method for a dehumidifier integrating a width learning system and PID according to claim 4, characterized in that, Step S3 includes: Map the feature matrix and enhanced node matrix Integration, constructing a joint matrix The following relation is satisfied: ； in, Represents the mapping feature matrix, Represents an enhanced node matrix; The objective function is introduced with the goal of minimizing the prediction error, satisfying the following relationship: ； in, This represents the output weights of the width learning system. This represents the output matrix of the width learning system. Represents the regularization parameter; By differentiating the objective function and setting the derivative to zero, the output weights can be obtained. The following relation is satisfied: ； in, Representing the joint matrix transpose, Represents the identity matrix. This represents the output matrix of the width learning system; Using the obtained output weights right Humidity of the external environment at any time Make a prediction that satisfies the following relationship: ； in, express Predicted humidity value at any given time. express The joint matrix at time points, ,and ,as well as ; in, express The mapping feature vector at time step, express The augmented node vector at time step, express The vector of measured parameters at time, This represents the transpose of the transformation matrix. This represents the activation function.

6. The intelligent humidity control method for a dehumidifier integrating a width learning system and PID according to claim 5, characterized in that, Step S4 includes: calculate Control deviation at time The following relation is satisfied: ； in, express Control deviation at any time This indicates the set target humidity value. express Predicted humidity value at any given time; Calculated based on PID control theory Input frequency of the dehumidifier compressor The following relation is satisfied: ； in, express The input frequency of the dehumidifier compressor at all times. Represents the proportionality coefficient. Represents the integral coefficient. Represents the differential coefficient. Indicates the first Control deviation at any time Indicates the sampling period, i.e., the period from which the sensor... Time's up The time interval between sampling moments.

7. A dehumidifier humidity intelligent adjustment system integrating a width learning system and PID control, characterized in that: include: The data acquisition module is used to acquire real-time monitoring data of the dehumidifier's operating environment, including ambient humidity, ambient temperature, ambient air pressure, ambient wind speed, and compressor frequency. The feature transformation module is used to construct a multi-kernel extreme learning machine autoencoder. It takes the real-time monitoring data as input, performs feature transformation through multiple kernel functions, and generates a mapping feature matrix with nonlinear prediction capabilities. The humidity prediction module is used to construct the enhancement nodes of the width learning system based on the mapping feature matrix, and integrate the mapping feature matrix with the enhancement nodes to build a humidity intelligent prediction model. The humidity intelligent prediction model is used to predict the environmental humidity at future times to obtain the humidity prediction value. The control and adjustment module is used to calculate the control deviation between the predicted humidity value and the target humidity value, and input the control deviation to the PID controller, which adjusts the compressor frequency of the dehumidifier in real time based on the control deviation.

Images