Industrial electric cabinet cooling fan energy efficiency control method

Abstract

The invention discloses an energy efficiency control method for a cooling fan of an industrial electric cabinet, and the method comprises the steps: firstly collecting temperature, humidity, equipment current / voltage and other multi-mode sensing data, carrying out the PCA algorithm dimension reduction and feature extraction, combining with Granger causality test, thermodynamic principle and machine learning algorithm, and carrying out the building and fusion, thereby obtaining a cooling efficiency model. Thirdly, preprocessing current and voltage data of the equipment, training and optimizing an LSTM model to predict load power, and calculating a fan rotating speed control value through a heat dissipation efficiency model; then, a PWM speed regulation signal is generated by adopting a fuzzy PID control algorithm, and a pulse width modulation and smooth speed regulation mixed strategy is applied to the duty ratio; and finally, energy efficiency indexes are collected to compare and update the weight of the LSTM model, knowledge distillation is introduced to optimize new model training, and self-adaptive adjustment of the rotating speed of the fan is achieved. The rotating speed of the fan can be accurately adjusted, energy consumption is reduced, the energy utilization efficiency is improved, performance reduction of equipment due to overheating is effectively avoided, and the service life of the equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of fan control, and in particular to an energy efficiency control method for a cooling fan of an industrial electric cabinet. Background Art

[0002] In modern industrial production, industrial electrical cabinets are widely used in many fields such as power, automation control, and communication. Various electrical devices such as controllers, drivers, and power modules are integrated inside the electrical cabinet, and these devices will inevitably generate heat during operation. Take a typical automated production line control cabinet as an example. When the PLC (programmable logic controller) and inverter are running at full load, they will release a lot of heat. If this heat cannot be dissipated in time, the temperature inside the electrical cabinet will continue to rise, which will lead to a decline in the performance of the equipment, such as reduced working accuracy and shortened service life of electronic components. In severe cases, it may even cause equipment failure, resulting in production interruption and huge economic losses to the company.

[0003] At present, the cooling fan control methods used by most industrial cabinets are relatively simple and extensive. The following are the common types:

[0004] Fixed speed control: This is the most traditional control method. The cooling fan will continue to run at a fixed speed after the cabinet is started. Although this method is simple in structure and low in cost, it cannot be flexibly adjusted according to the actual heat generation inside the cabinet. When the equipment load is low and the heat generated is low, the fan still runs at a high speed, which not only wastes electricity, but also increases the mechanical wear of the fan and shortens its service life. For example, in the cabinets of some intermittent industrial equipment, when the equipment is in standby or light load operation, the unnecessary high-speed operation of the cooling fan will cause a lot of energy consumption.

[0005] Temperature switch control: This method sets a temperature switch in the electric cabinet. When the temperature inside the electric cabinet reaches the preset threshold, the fan starts; when the temperature drops to a certain level, the fan stops. However, this control method has obvious hysteresis. Since the action of the temperature switch requires a certain temperature change range, before the fan starts, the temperature inside the electric cabinet may have risen to a level that has a certain impact on the equipment; and after the fan stops, the temperature may rise rapidly. In addition, this simple switch control method cannot achieve fine adjustment of the fan speed, and it is difficult to meet the heat dissipation requirements under different working conditions.

[0006] With the rapid development of Industry 4.0 and intelligent manufacturing, industrial production has put forward higher requirements for the intelligence, efficiency and energy saving of equipment. In the field of heat dissipation of industrial cabinets, there is also an urgent need for a control method that can intelligently adjust the speed of the cooling fan according to the actual operating conditions of the cabinet to achieve energy efficiency optimization. On the one hand, in order to reduce production costs and improve production efficiency, enterprises pay more and more attention to the energy consumption control of equipment. By optimizing the energy efficiency of the cooling fan, the overall energy consumption of the industrial cabinet can be significantly reduced, saving a lot of electricity expenses for enterprises. On the other hand, with the continuous advancement of electronic technology, the integration of equipment in the cabinet is getting higher and higher, and the heat density is getting higher and higher, which poses higher challenges to the heat dissipation capacity and precise control of the cooling system. The traditional cooling fan control method can no longer meet the needs of modern industrial production. Therefore, it is imperative to develop a new cooling fan energy efficiency control method. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide an energy efficiency control method for the cooling fan of an industrial electric cabinet. By real-time and accurate monitoring of multi-dimensional parameters during the operation of the industrial electric cabinet and using advanced intelligent algorithms to perform in-depth analysis and processing of these data, dynamic and adaptive adjustment of the cooling fan speed can be achieved, thereby minimizing the energy consumption of the cooling fan and improving energy utilization efficiency.

[0008] To solve the above technical problems, an embodiment of the present invention provides the following technical solution: an industrial electric cabinet cooling fan energy efficiency control method, comprising the following steps:

[0009] Collect multi-modal sensor data related to the cooling fan, analyze the correlation of the multi-modal sensor data through data processing algorithms, and establish a cooling efficiency model;

[0010] Build and train a lightweight machine learning model LSTM to predict the load power change trend in the future, and calculate the fan speed control value through the heat dissipation efficiency model based on the prediction results;

[0011] Taking the fan speed control value as the target, the fuzzy PID control algorithm is used to generate the PWM speed regulation signal;

[0012] Collect the energy efficiency index of the cooling fan, compare it with the LSTM model, update the LSTM model weight, and adaptively adjust the cooling fan speed.

[0013] Preferably, the multimodal sensor data related to the cooling fan includes temperature, humidity, device current / voltage, ambient air pressure, and fan speed.

[0014] Preferably, the multimodal sensing data is analyzed for correlation through a data processing algorithm to establish a heat dissipation efficiency model, specifically:

[0015] The multimodal data is reduced in dimension using the PCA algorithm, and time domain / frequency domain features are extracted;

[0016] The data after feature extraction is defined as the relationship between variables and independent variables, and a linear relationship model between multiple independent variables and dependent variables is established through Granger causality test. The influence coefficient of each independent variable on the heat dissipation efficiency is determined by fitting historical data; wherein the independent variables include temperature, humidity, equipment current / voltage, air pressure, fan speed, load power, and the dependent variable is heat dissipation efficiency;

[0017] Based on the basic principles of thermodynamics, a physical model is established to describe the heat transfer process inside the electrical cabinet, which includes heat generation, heat conduction, heat convection and heat radiation;

[0018] A data-driven model is established using a machine learning algorithm, in which the feature weights of each independent variable in the model are determined according to its influence coefficient on the heat dissipation efficiency. Multimodal sensor data is used as input and heat dissipation efficiency is used as output for training. The weights and biases of the network are continuously adjusted through the back-propagation algorithm.

[0019] The heat dissipation efficiency model is obtained by weighted fusion of the thermodynamic model and the data-driven model.

[0020] Preferably, the lightweight machine learning model LSTM is constructed and trained to predict the load power change trend in the future, specifically:

[0021] Preprocess the equipment working current and voltage data in the multimodal sensor data;

[0022] The preprocessed data is divided into a training set and a test set, and the training set is used to train the machine learning model LSTM model;

[0023] Use the test set to evaluate the trained LSTM model, calculate the prediction error of the model, and optimize the model based on the evaluation results.

[0024] Preferably, the fan speed control value is calculated by a heat dissipation efficiency model according to the prediction result, specifically:

[0025]

[0026] Among them, P r is the load power in the future predicted by the LSTM model; t is the time interval; c p is the specific heat capacity of air; ρ is the air density; A is the effective ventilation area of ​​the fan; k1 is the proportional coefficient of air flow rate and fan speed; ΔT is the temperature difference between the inside and outside of the electric cabinet, which can be calculated from the measurement values ​​of the internal temperature sensor of the electric cabinet and the external ambient temperature sensor; k2 is the error correction coefficient of the heat dissipation efficiency model.

[0027] Preferably, the fan speed control value is targeted and a fuzzy PID control algorithm is used to generate a PWM speed regulation signal, specifically:

[0028] The PID control algorithm consists of three parts: proportional, integral, and differential. The formula is:

[0029]

[0030] Among them, K p is the proportionality coefficient, K i is the integration coefficient, K d is the differential coefficient; e(t) is the error at the current moment, that is, the difference between the fan speed control value and the current fan speed; the error e and the error change rate ec are used as the input of the fuzzy controller, where the error change rate Δt is the sampling time interval;

[0031] Fuzzify e and ec, map them to the corresponding fuzzy domain, divide the fuzzy subsets, each subset corresponds to a membership function, and calculate the membership of e and ec to each fuzzy subset through the membership function;

[0032] According to the expert experience, a fuzzy rule base is formulated, and based on the fuzzy rules and the fuzzified input, the fuzzy output is calculated by using the fuzzy reasoning method;

[0033] Defuzzify the fuzzy output obtained by fuzzy reasoning to obtain the accurate adjustment value ΔK p , ΔK i , ΔK d According to the adjustment amount obtained by defuzzification, the PID parameters are updated in real time:

[0034]

[0035] Substitute the updated PID parameters into the PID control algorithm formula and calculate the control quantity u(t). Assume that the value range of the control quantity u(t) is [u min ,u max ], the value range of PWM signal duty cycle D is [0,1], then:

[0036]

[0037] Preferably, the collecting of the cooling fan energy efficiency index, comparing it with the LSTM model and updating the LSTM model weights are specifically as follows:

[0038] Collect the actual speed n of the cooling fan and the real-time temperature T in the electrical cabinet a 、Ambient temperature T e , Equipment load power Pl And the real-time power consumption P of the fan fan ;

[0039] Calculate energy efficiency indicators, including heat dissipation efficiency and energy efficiency ratio, where:

[0040] Heat dissipation efficiency Energy efficiency ratio Where c is the specific heat capacity of air, and m is the mass of the air in the cabinet;

[0041] The LSTM model predicts the parameters of the cooling fan speed and the temperature in the cabinet for a period of time in the future based on historical data, not limited to equipment load power and environmental parameters, and calculates the predicted cooling efficiency E based on the predicted parameters. p and energy efficiency ratio COP p ;

[0042] Calculate the heat dissipation efficiency error and energy efficiency ratio error, and assign weights to calculate the comprehensive error e;

[0043] Assume that the weight of the LSTM model is W, the bias is b, and the loss function L is defined as the mean square value of the comprehensive error e, that is, Where N is the number of samples; the gradient of the loss function with respect to the weight W and bias b is calculated according to the chain rule, and the weight and bias are updated; the weight and bias gradients are obtained by back propagation.

[0044] Preferably, the feature-extracted data is defined as the relationship between the variable and the independent variable, and a linear relationship model between multiple independent variables and the dependent variable is established through the Granger causality test, specifically:

[0045] Unit root test is performed on time series data. If the data is non-stationary, it is made stationary through difference or logarithmic transformation. The independent variables and dependent variables are aligned at the timestamp to form a synchronous observation data set.

[0046] Use Akaike Information Criterion or Bayesian Information Criterion to determine the optimal lag order p;

[0047] For each independent variable X i Perform the following tests in order:

[0048] Restricted Model Only the impact of the historical information of the dependent variable itself on the current value is considered, where α0 is a constant term, α k is the coefficient of the lagged term of the dependent variable, ε t is the random error term;

[0049] Unrestricted Model β0 is a constant term, β k is the coefficient of the lagged term of the dependent variable, γ k is the coefficient of the lagged term of the independent variable, u tis the random error term;

[0050] By performing regression estimation on the restricted model and the unrestricted model, the residual sum of squares of the restricted model and the residual sum of squares of the unrestricted model are obtained, and the test statistics are calculated to perform hypothesis testing.

[0051] Preferably, the duty cycle adopts a hybrid control strategy of pulse width modulation and smooth speed regulation, specifically:

[0052] When the duty cycle D is less than 5%, a fixed period T is used for intermittent start and stop, and the on time t on =D×T;

[0053] When 5%D<10%, smooth transition mode control is adopted, and the PWM signal frequency f increases linearly with the duty cycle D. The calculation formula is:

[0054]

[0055] f max and f min They correspond to the maximum and minimum values ​​of the PWM signal frequency under the smooth transition mode control respectively;

[0056] When the duty cycle D>10%, switch back to standard PWM speed regulation;

[0057] Among them, hysteresis interval buffering is adopted between each mode switching.

[0058] Preferably, the method further includes introducing knowledge distillation to use the historical model output as a soft label to constrain the new model training process during the adaptive adjustment of the cooling fan speed, specifically:

[0059] The collected data is input into the historical model, and the prediction results of the cooling fan speed and the temperature parameters in the electric cabinet are output as soft labels;

[0060] During the training of the new model, a comprehensive loss function including soft label loss and hard label loss is constructed; the soft label loss is used to measure the difference between the new model output and the historical model output; the hard label loss is used to measure the difference between the new model output and the actual measurement value;

[0061] Use the collected data and the constructed comprehensive loss function to train the new model, and continuously adjust the parameters of the new model through the back propagation algorithm to minimize the value of the comprehensive loss function.

[0062] The beneficial effects of the above technical solution of the present invention are as follows:

[0063] 1. The present invention collects multimodal sensor data and uses the LSTM model to predict load power, and combines the heat dissipation efficiency model to accurately calculate the fan speed control value, thereby realizing dynamic adaptive adjustment of the fan speed. Compared with fixed speed control, the fan will not continue to run at high speed under conditions of low equipment load and low heat generation, thus avoiding energy waste; compared with temperature switch control, the speed can be adjusted more timely and accurately according to the actual heat generation of the electric cabinet, thus reducing unnecessary energy consumption.

[0064] 2. The present invention adopts fuzzy PID control algorithm to generate PWM speed regulation signal according to fan speed control value, combines error and error change rate to perform fuzzy processing, formulates rule base and inference calculation, updates PID parameters in real time, and realizes accurate regulation of fan speed. It has fast response speed and high regulation accuracy, can quickly adapt to heat changes in the electric cabinet, maintain stable temperature in the electric cabinet, and meet the heat dissipation requirements under different working conditions.

[0065] 3. When establishing the heat dissipation efficiency model, the present invention performs PCA dimensionality reduction and feature extraction on multimodal data, combines Granger causality test to determine the relationship between independent variables and dependent variables, and integrates thermodynamic models and data-driven models. The advantages of physical principles and data-driven are utilized to improve the accuracy of the heat dissipation efficiency model, and more accurately predict and evaluate the heat dissipation efficiency. During the LSTM model training process, the accuracy and generalization ability of the model in predicting the load power change trend are improved through data preprocessing, data set division, evaluation optimization, and the introduction of knowledge distillation technology, so that it can better adapt to the complex and changeable operating environment of industrial electrical cabinets.

[0066] 4. The fan control signal duty cycle of the present invention adopts a hybrid control strategy of pulse width modulation and smooth speed regulation. By limiting the minimum duty cycle, it avoids the complete stop of the motor due to quantization error. Different speed regulation methods are adopted in different duty cycle ranges, and a hysteresis interval buffer is set to avoid frequent sudden changes in fan speed and frequent mode switching, reduce the impact on the fan and equipment in the electrical cabinet, and enhance the stability and reliability of the heat dissipation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of the energy efficiency control method of the industrial electric cabinet cooling fan of the present invention. DETAILED DESCRIPTION

[0068] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0069] like Figure 1 As shown, the present invention proposes an industrial electric cabinet cooling fan energy efficiency control method, comprising the following steps:

[0070] S1. Collect multimodal sensor data related to the cooling fan, analyze the correlation of the multimodal sensor data through a data processing algorithm, and establish a cooling efficiency model;

[0071] S2. Build and train a lightweight machine learning model LSTM to predict the load power change trend in the future, and calculate the fan speed control value through the heat dissipation efficiency model based on the prediction results;

[0072] S3, taking the fan speed control value as the target, using fuzzy PID control algorithm to generate PWM speed regulation signal;

[0073] S4. Collect the energy efficiency index of the cooling fan, compare it with the LSTM model, update the LSTM model weight, and adaptively adjust the speed of the cooling fan.

[0074] The working process of the present invention is described below in conjunction with specific embodiments:

[0075] In this embodiment, the multimodal sensor data related to the cooling fan in step S1 includes temperature, humidity, device current / voltage, ambient air pressure, and fan speed. Then, the multimodal sensor data is analyzed for correlation through a data processing algorithm to establish a cooling efficiency model, specifically:

[0076] (1) Reduce the dimension of multimodal data using the PCA algorithm. The steps of PCA dimensionality reduction are as follows:

[0077] 1. Standardized data:

[0078] Because PCA is very sensitive to the scale of the data, it is usually necessary to standardize the data (also called normalization) before performing PCA. That is, each feature is subtracted from its mean and divided by its standard deviation so that the mean of each feature is 0 and the variance is 1.

[0079] 2. Calculate the covariance matrix:

[0080] The covariance matrix measures the correlation between the features in the data set. For n-dimensional data, its covariance matrix is ​​an n×n matrix, where each element C ij is the covariance between the ith feature and the jth feature.

[0081] 3. Calculate the eigenvalues ​​and eigenvectors of the covariance matrix:

[0082] The eigenvalue represents the variance contribution of each principal component in the data set, while the eigenvector defines the direction of the new coordinate axis.

[0083] 4. Select the main components:

[0084] According to the size of the eigenvalue, select the eigenvectors corresponding to the first k largest eigenvalues. These eigenvectors will form a new feature space, where k is the number of dimensions after dimensionality reduction.

[0085] 5. Project the original data into the new feature space:

[0086] Using the selected eigenvectors as a basis, the original data is projected into the new feature space to obtain the reduced-dimensional data.

[0087] Then, extract the time domain / frequency domain features; for multimodal data such as temperature and current that change over time, extract time domain features such as mean, variance, maximum, and minimum. For example, calculating the mean of temperature data over a period of time can reflect the average temperature level in the cabinet during that period; calculating the variance of current data can reflect the current fluctuation. The larger the variance, the worse the current stability. You can also extract the zero-crossing rate, that is, the number of times the signal crosses the zero value per unit time. This is important for analyzing periodically changing data such as fan speed, and can reflect its frequency characteristics.

[0088] Frequency domain feature extraction: First, perform Fourier transform on the data to convert the time domain signal into a frequency domain signal. For the fan speed data, the frequency component can be obtained through Fourier transform, and the main frequency and its corresponding amplitude can be found. The main frequency component can reflect the basic frequency of the fan rotation, and the amplitude indicates the energy intensity of the frequency component. The power spectral density (PSD) can also be calculated. PSD represents the distribution of signal power in frequency, which can more intuitively observe the contribution of different frequency components to the signal power, helping to analyze the operating status of the equipment in the cabinet and the working characteristics of the cooling fan.

[0089] (2) The data after feature extraction is defined as the relationship between variables and independent variables, and a linear relationship model between multiple independent variables and dependent variables is established through Granger causality test. The influence coefficient of each independent variable on the heat dissipation efficiency is determined by fitting the historical data; wherein the independent variables include temperature, humidity, equipment current / voltage, air pressure, fan speed, load power, and the dependent variable is heat dissipation efficiency; the linear relationship model is constructed specifically as follows:

[0090] Unit root test is performed on time series data. If the data is non-stationary, it is made stationary through difference or logarithmic transformation. The independent variables and dependent variables are aligned at the timestamp to form a synchronous observation data set.

[0091] Use Akaike Information Criterion or Bayesian Information Criterion to determine the optimal lag order p;

[0092] For each independent variable X i Perform the following tests in order:

[0093] Restricted Model Only the impact of the historical information of the dependent variable itself on the current value is considered, where α0 is a constant term, α k is the coefficient of the lagged term of the dependent variable, ε t is the random error term;

[0094] Unrestricted Model β0 is a constant term, β k is the coefficient of the lagged term of the dependent variable, γ k is the coefficient of the lagged term of the independent variable, u t is the random error term;

[0095] By performing regression estimation on the restricted model and the unrestricted model, the residual sum of squares of the restricted model and the residual sum of squares of the unrestricted model are obtained, and the test statistics are calculated to perform hypothesis testing.

[0096] (3) Based on the basic principles of thermodynamics, a physical model describing the heat transfer process inside the electric cabinet is established. The heat transfer process includes heat generation, heat conduction, heat convection and heat radiation. Then, a data-driven model is established using a machine learning algorithm. The characteristic weights of each independent variable in the model are determined according to its influence coefficient on the heat dissipation efficiency. Multimodal sensor data is used as input and heat dissipation efficiency is used as output for training. The weights and biases of the network are continuously adjusted through the back propagation algorithm. Finally, the heat dissipation efficiency model is obtained by weighted fusion of the thermodynamic model and the data-driven model. The physical model established based on the basic principles of thermodynamics accurately describes the heat transfer process inside the electric cabinet from a theoretical level, covering various links such as heat generation, conduction, convection and radiation, and ensures the accuracy of the basic understanding of the heat dissipation mechanism. The data-driven model uses a machine learning algorithm to explore the potential relationship between multimodal sensor data and heat dissipation efficiency. The weights and biases are continuously adjusted through the back propagation algorithm to adaptively fit complex nonlinear relationships. The combination of the two can give full play to the theoretical nature of the physical model and the flexibility of the data-driven model, make up for each other's shortcomings, and thus significantly improve the accuracy of the heat dissipation efficiency model, and more accurately predict and evaluate the heat dissipation efficiency.

[0097] In this embodiment, the step S2 constructs and trains a lightweight machine learning model LSTM to predict the load power change trend in the future, specifically:

[0098] The working current and voltage data of the equipment in the multimodal sensor data are preprocessed. The data preprocessing includes data cleaning and normalization. Data cleaning: There may be abnormal values ​​in the working current and voltage data of the equipment, such as mutation values ​​caused by sensor failure or electromagnetic interference. Identify abnormal values ​​by setting a reasonable threshold range. For example, determine a reasonable upper and lower limit based on the rated current and voltage range of the equipment and the statistical analysis of historical data. For data points out of the range, mean interpolation, median interpolation or machine learning-based outlier repair algorithms can be used for processing. At the same time, check the integrity of the data and fill in the possible missing values. Linear interpolation, spline interpolation or model-based (such as K nearest neighbor algorithm) interpolation methods can be used to ensure the continuity and reliability of the data. Normalization: Due to the different dimensions and value ranges of current and voltage data, normalization is required to avoid certain features dominating the training process during model training and affecting model performance. Common normalization methods include minimum-maximum normalization.

[0099] The preprocessed data is divided into a training set and a test set, and the training set is used to train the machine learning model LSTM model. The preprocessed data is divided into a training set and a test set according to a certain ratio, and the common division ratio is 70%-30% or 80%-20%. When dividing, the randomness and representativeness of the data must be ensured to avoid the situation where the data distribution of the training set and the test set is too different.

[0100] Use the test set to evaluate the trained LSTM model, calculate the prediction error of the model, and optimize the model based on the evaluation results. First, initialize the LSTM model and set the model's hyperparameters, such as the number of hidden layer neurons, the number of layers, the learning rate, and the batch size. The choice of the number of hidden layer neurons and the number of layers will affect the complexity and learning ability of the model, and generally needs to be determined through experiments and tuning. The learning rate controls the step size of parameter updates during model training. Too large a learning rate may cause unstable model training, while too small a learning rate will slow down the training process. The batch size determines the number of samples used in each training. A suitable batch size can improve training efficiency and the generalization ability of the model. Input the training set data into the LSTM model, use the back propagation algorithm to calculate the error between the predicted value and the true value, and update the model's weights and biases. Continuously iterate the training until the model converges or reaches the preset number of training rounds. Finally, use the test set data to evaluate the trained LSTM model. Commonly used prediction error indicators include mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE). In addition, optimize the model based on the evaluation results. If the prediction error is large, you can try to adjust the model's hyperparameters, such as increasing the number of hidden layer neurons or layers to improve the model's fitting ability; adjusting the learning rate to make the model training more stable and efficient; changing the batch size to find the optimal combination of training parameters. You can also consider increasing the amount of training data, improving the diversity of data, and enhancing the generalization ability of the model. In addition, you can also try to use regularization methods (such as L1 and L2 regularization) to prevent the model from overfitting and improve the performance of the model.

[0101] Then, the fan speed control value is calculated based on the prediction results through the heat dissipation efficiency model, specifically:

[0102]

[0103] Among them, P r is the load power in the future predicted by the LSTM model; t is the time interval; c p is the specific heat capacity of air; ρ is the air density; A is the effective ventilation area of ​​the fan; k1 is the proportional coefficient of air flow rate and fan speed; ΔT is the temperature difference between the inside and outside of the electric cabinet, which can be calculated from the measurement values ​​of the internal temperature sensor of the electric cabinet and the external ambient temperature sensor; k2 is the error correction coefficient of the heat dissipation efficiency model.

[0104] In this embodiment, the step S3 is aimed at the fan speed control value and uses a fuzzy PID control algorithm to generate a PWM speed regulation signal, specifically:

[0105] The PID control algorithm consists of three parts: proportional, integral, and differential. The formula is:

[0106]

[0107] Among them, K p is the proportionality coefficient, K i is the integration coefficient, K d is the differential coefficient; e(t) is the error at the current moment, that is, the difference between the fan speed control value and the current fan speed; the error e and the error change rate ec are used as the input of the fuzzy controller, where the error change rate Δt is the sampling time interval;

[0108] Fuzzify e and ec, map them to the corresponding fuzzy domain, divide the fuzzy subsets, each subset corresponds to a membership function, and calculate the membership of e and ec to each fuzzy subset through the membership function;

[0109] According to the expert experience, a fuzzy rule base is formulated, and based on the fuzzy rules and the fuzzified input, the fuzzy output is calculated by using the fuzzy reasoning method;

[0110] Defuzzify the fuzzy output obtained by fuzzy reasoning to obtain the accurate adjustment value ΔK p , ΔK i , ΔK d According to the adjustment amount obtained by defuzzification, the PID parameters are updated in real time:

[0111]

[0112] Substitute the updated PID parameters into the PID control algorithm formula and calculate the control quantity u(t). Assume that the value range of the control quantity u(t) is [u min ,u max ], the value range of PWM signal duty cycle D is [0,1], then:

[0113]

[0114] In this embodiment, the step S4 collects the energy efficiency index of the cooling fan, compares it with the LSTM model and updates the LSTM model weight, specifically:

[0115] Collect the actual speed n of the cooling fan and the real-time temperature T in the electrical cabinet a 、Ambient temperature T e , Equipment load power P l And the real-time power consumption P of the fan fan ;

[0116] Calculate energy efficiency indicators, including heat dissipation efficiency and energy efficiency ratio, where:

[0117] Heat dissipation efficiency Energy efficiency ratio Where c is the specific heat capacity of air, and m is the mass of the air in the cabinet;

[0118] The LSTM model predicts the parameters of the cooling fan speed and the temperature in the cabinet for a period of time in the future based on historical data, not limited to equipment load power and environmental parameters, and calculates the predicted cooling efficiency E based on the predicted parameters. p and energy efficiency ratio COP p ;

[0119] Calculate the heat dissipation efficiency error and energy efficiency ratio error, and assign weights to calculate the comprehensive error e;

[0120] Assume that the weight of the LSTM model is W, the bias is b, and the loss function L is defined as the mean square value of the comprehensive error e, that is, Where N is the number of samples; the gradient of the loss function with respect to the weight W and bias b is calculated according to the chain rule, and the weight and bias are updated; the weight and bias gradients are obtained by back propagation.

[0121] In addition, the above duty cycle generation adopts a hybrid control strategy of pulse width modulation and smooth speed regulation, specifically:

[0122] When the duty cycle D is less than 5%, a fixed period T is used for intermittent start and stop, and the on time t on =D×T; for example, when D=3%, t on The minimum duty cycle is limited to 30ms to avoid the motor from completely stopping due to quantization error.

[0123] When 5%<D<10%, smooth transition mode control is adopted, and the PWM signal frequency f increases linearly with the duty cycle D. The calculation formula is:

[0124]

[0125] f max and f min They correspond to the maximum and minimum values ​​of the PWM signal frequency under the smooth transition mode control. The smooth transition mode control makes the PWM signal frequency increase linearly with the duty cycle. This allows the fan speed to change smoothly, avoids mechanical shock and noise caused by sudden speed changes, prolongs the fan's service life, and provides a stable heat dissipation effect to ensure the relative stability of the temperature inside the cabinet.

[0126] When the duty cycle D>10%, switch back to standard PWM speed regulation;

[0127] Among them, a hysteresis interval buffer is used between each mode switch, for example, the duty cycle is temporarily inactive within the range of ±1%, so as to avoid frequent switching at the mode transition.

[0128] In addition, in the adaptive adjustment of the cooling fan speed, knowledge distillation is introduced to use the historical model output as a soft label to constrain the new model training process, specifically:

[0129] The collected data is input into the historical model, and the prediction results of the cooling fan speed and the temperature parameters in the electric cabinet are output as soft labels;

[0130] During the training of the new model, a comprehensive loss function including soft label loss and hard label loss is constructed; the soft label loss is used to measure the difference between the new model output and the historical model output; the hard label loss is used to measure the difference between the new model output and the actual measurement value;

[0131] Use the collected data and the constructed comprehensive loss function to train the new model, and continuously adjust the parameters of the new model through the back propagation algorithm to minimize the value of the comprehensive loss function.

[0132] The new model not only fits the actual measurements (hard labels), but also learns the knowledge of the historical model (soft labels), gradually optimizing its own performance during the training process. As training progresses, the new model will continuously adjust its parameters to better match the output of the historical model while maintaining its ability to fit the actual data.

[0133] In practical applications, the present invention collects multimodal sensor data in real time by installing sensors such as temperature, humidity, equipment current / voltage, and ambient air pressure sensors in the control cabinet. The PCA algorithm is used to reduce the dimension of the data and extract features. The relationship between the independent variable and the dependent variable is established through the Granger causality test. The model is established and integrated by combining the thermodynamic principles and the machine learning algorithm to obtain an accurate heat dissipation efficiency model. For example, when the PLC and the inverter are running at full load, the model can accurately evaluate the relationship between the heat dissipation efficiency and various factors.

[0134] Then, the working current and voltage data of the equipment are collected, and the LSTM model is trained and optimized after preprocessing to predict the load power change trend. The fan speed control value is calculated based on the prediction results and the heat dissipation efficiency model. If it is predicted that the equipment will enter a high-load operation state, the fan speed is increased in advance. The fuzzy PID control algorithm generates a PWM speed regulation signal based on the speed control value and adjusts the fan speed in real time. When the temperature in the cabinet rises, it can respond quickly and accurately adjust the fan speed.

[0135] The duty cycle adopts a hybrid control strategy. When the duty cycle is less than 5%, the fan starts and stops intermittently at a fixed period; when it is 5%-10%, the PWM signal frequency increases linearly with the duty cycle; when it is greater than 10%, it switches back to standard PWM speed regulation, and there is a hysteresis interval buffer when the mode is switched. This effectively avoids sudden changes in fan speed and frequent mode switching, ensuring the stability of the cooling system.

[0136] At the same time, we continuously collect energy efficiency indicators such as the actual fan speed and the temperature inside the cabinet, compare them with the LSTM model prediction value, calculate the error, and update the model weight. We introduce knowledge distillation technology to allow the new model to learn historical model knowledge, improve prediction accuracy, and achieve adaptive adjustment of fan speed.

[0137] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet, characterized in that: The following steps are involved: Collect multi-modal sensor data related to the cooling fan, analyze the correlation of the multi-modal sensor data through data processing algorithms, and establish a cooling efficiency model; Build and train a lightweight machine learning model LSTM to predict the load power change trend in the future, and calculate the fan speed control value through the heat dissipation efficiency model based on the prediction results; Taking the fan speed control value as the target, the fuzzy PID control algorithm is used to generate the PWM speed regulation signal; Collect the energy efficiency index of the cooling fan, compare it with the LSTM model, update the LSTM model weight, and adaptively adjust the cooling fan speed.

2. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 1, characterized in that: The multimodal sensor data related to the cooling fan includes temperature, humidity, device current / voltage, ambient air pressure, and fan speed.

3. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 1, characterized in that: The multi-modal sensor data is analyzed for correlation through a data processing algorithm to establish a heat dissipation efficiency model, specifically: The multimodal data is reduced in dimension using the PCA algorithm, and time domain / frequency domain features are extracted; The data after feature extraction is defined as the relationship between variables and independent variables, and a linear relationship model between multiple independent variables and dependent variables is established through Granger causality test. The influence coefficient of each independent variable on the heat dissipation efficiency is determined by fitting historical data; wherein the independent variables include temperature, humidity, equipment current / voltage, air pressure, fan speed, load power, and the dependent variable is heat dissipation efficiency; Based on the basic principles of thermodynamics, a physical model is established to describe the heat transfer process inside the electrical cabinet, which includes heat generation, heat conduction, heat convection and heat radiation; A data-driven model is established using a machine learning algorithm, in which the feature weights of each independent variable in the model are determined according to its influence coefficient on the heat dissipation efficiency. Multimodal sensor data is used as input and heat dissipation efficiency is used as output for training. The weights and biases of the network are continuously adjusted through the back-propagation algorithm. The heat dissipation efficiency model is obtained by weighted fusion of the thermodynamic model and the data-driven model.

4. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 1, characterized in that: The lightweight machine learning model LSTM is constructed and trained to predict the load power change trend in the future, specifically: Preprocess the equipment working current and voltage data in the multimodal sensor data; The preprocessed data is divided into a training set and a test set, and the training set is used to train the machine learning model LSTM model; Use the test set to evaluate the trained LSTM model, calculate the prediction error of the model, and optimize the model based on the evaluation results.

5. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 1, characterized in that: The fan speed control value is calculated by the heat dissipation efficiency model according to the prediction result, specifically: Among them, P r is the load power in the future predicted by the LSTM model; t is the time interval; c p is the specific heat capacity of air; ρ is the air density; A is the effective ventilation area of ​​the fan; k1 is the proportional coefficient of air flow rate and fan speed; ΔT is the temperature difference between the inside and outside of the electric cabinet, which can be calculated from the measurement values ​​of the internal temperature sensor of the electric cabinet and the external ambient temperature sensor; k2 is the error correction coefficient of the heat dissipation efficiency model.

6. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 1, characterized in that: The fan speed control value is targeted and a fuzzy PID control algorithm is used to generate a PWM speed regulation signal, specifically: The PID control algorithm consists of three parts: proportional, integral, and differential. The formula is: Among them, K p is the proportionality coefficient, K i is the integration coefficient, K d is the differential coefficient; e(t) is the error at the current moment, that is, the difference between the fan speed control value and the current fan speed; the error e and the error change rate ec are used as the input of the fuzzy controller, where the error change rate Δt is the sampling time interval; Fuzzify e and ec, map them to the corresponding fuzzy domain, divide the fuzzy subsets, each subset corresponds to a membership function, and calculate the membership of e and ec to each fuzzy subset through the membership function; According to the expert experience, a fuzzy rule base is formulated, and based on the fuzzy rules and the fuzzified input, the fuzzy output is calculated by using the fuzzy reasoning method; Defuzzify the fuzzy output obtained by fuzzy reasoning to obtain the accurate adjustment value ΔK p , ΔK i , ΔK d According to the adjustment amount obtained by defuzzification, the PID parameters are updated in real time: Substitute the updated PID parameters into the PID control algorithm formula and calculate the control quantity u(t). Assume that the value range of the control quantity u(t) is [u min ,u max ], the value range of PWM signal duty cycle D is [0,1], then:

7. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 1, characterized in that: The energy efficiency index of the cooling fan is collected, compared with the LSTM model and the weight of the LSTM model is updated, specifically: Collect the actual speed n of the cooling fan and the real-time temperature T in the electrical cabinet a 、Ambient temperature T e , Equipment load power P l And the real-time power consumption P of the fan fan ; Calculate energy efficiency indicators, including heat dissipation efficiency and energy efficiency ratio, where: Cooling efficiency Energy efficiency ratio Where c is the specific heat capacity of air, and m is the mass of the air in the cabinet; The LSTM model predicts the parameters of the cooling fan speed and the temperature in the cabinet for a period of time in the future based on historical data, not limited to equipment load power and environmental parameters, and calculates the predicted cooling efficiency E based on the predicted parameters. p and energy efficiency ratio COP p ; Calculate the heat dissipation efficiency error and energy efficiency ratio error, and assign weights to calculate the comprehensive error e; Assume that the weight of the LSTM model is W, the bias is b, and the loss function L is defined as the mean square value of the comprehensive error e, that is, Where N is the number of samples; the gradient of the loss function with respect to the weight W and bias b is calculated according to the chain rule, and the weight and bias are updated; the weight and bias gradients are obtained by back propagation.

8. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 3, characterized in that: The feature-extracted data is defined as the relationship between variables and independent variables, and a linear relationship model between multiple independent variables and dependent variables is established through Granger causality test, specifically: Unit root test is performed on time series data. If the data is non-stationary, it is made stationary through difference or logarithmic transformation. The independent variables and dependent variables are aligned at the timestamp to form a synchronous observation data set. Use Akaike Information Criterion or Bayesian Information Criterion to determine the optimal lag order p; For each independent variable X i Perform the following tests in order: Restricted Model Only the influence of the historical information of the dependent variable itself on the current value is considered, where α0 is a constant term, α k is the coefficient of the lagged term of the dependent variable, ε t is the random error term; Unrestricted Model β0 is a constant term, β k is the coefficient of the lagged term of the dependent variable, γ k is the coefficient of the lagged term of the independent variable, u t is the random error term; By performing regression estimation on the restricted model and the unrestricted model, the residual sum of squares of the restricted model and the residual sum of squares of the unrestricted model are obtained, and the test statistics are calculated to perform hypothesis testing.

9. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 6, characterized in that: The duty cycle adopts a hybrid control strategy of pulse width modulation and smooth speed regulation, specifically: When the duty cycle D is less than 5%, a fixed period T is used for intermittent start and stop, and the on time t on =D×T; When 5%D<10%, smooth transition mode control is adopted, and the PWM signal frequency f increases linearly with the duty cycle D. The calculation formula is: f max and f min They correspond to the maximum and minimum values ​​of the PWM signal frequency under the smooth transition mode control respectively; When the duty cycle D>10%, switch back to standard PWM speed regulation; Among them, hysteresis interval buffering is adopted between each mode switching.

10. The method for controlling the energy efficiency of a cooling fan of an industrial electric cabinet according to claim 1, characterized in that: It also includes the introduction of knowledge distillation in the adaptive adjustment of the cooling fan speed, using the historical model output as a soft label to constrain the new model training process, specifically: The collected data is input into the historical model, and the prediction results of the cooling fan speed and the temperature parameters in the electric cabinet are output as soft labels; During the training of the new model, a comprehensive loss function including soft label loss and hard label loss is constructed; the soft label loss is used to measure the difference between the new model output and the historical model output; the hard label loss is used to measure the difference between the new model output and the actual measurement value; Use the collected data and the constructed comprehensive loss function to train the new model, and continuously adjust the parameters of the new model through the back propagation algorithm to minimize the value of the comprehensive loss function.

Images