Centrifugal air compression station building control method and device based on load characteristics and AI optimization

Abstract

The invention discloses a centrifugal air compression station building control method and device based on load characteristics and AI optimization, and the method comprises the steps: obtaining environmental parameters in a ferrous metallurgy plant, carrying out the preprocessing of the noise of the environmental parameters, and obtaining a preprocessed signal matrix; a fitting model of dust concentration and filtering resistance change in environmental parameters is utilized to extract filter resistance characteristics, a common spatial pattern algorithm is called to extract load spectrum characteristics corresponding to frequency converter load response, and a characteristic matrix is generated. And calculating a filter fatigue index according to the relationship between the filter resistance and the dust concentration, and calculating a load speed regulation response index based on the load characteristics of the frequency converter so as to comprehensively calculate a system fatigue index. And taking the feature matrix as the input of the convolutional neural network, and performing feature fusion on the feature matrix. And training the long short-term memory network by fusing the feature vectors, calling the trained long short-term memory network to carry out time sequence prediction on the system fatigue trend according to the historical environment parameters, and generating an optimization control strategy.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy-saving control of production workshops, and more specifically, relates to a centrifugal air compressor station control method and device based on load characteristics and AI optimization. Background Art

[0002] In the large production workshops of steel and metallurgical plants, smelting, continuous casting, steel rolling and other processes all require stable, clean high-pressure air to drive various pneumatic equipment, control valves and instrument actuators. The centrifugal air compressor station energy-saving system can be used to provide a centralized compressed air source for the entire plant.

[0003] At present, the existing centrifugal air compressor station energy-saving system in the steel and metallurgical plant environment has high dust concentration (typical dust concentration range can reach 0.5~1 mg / m³), high temperature, and air containing metal fines may accelerate filter clogging, which to a certain extent reduces the efficiency of the air intake pretreatment module and increases the maintenance frequency. In addition, in the metallurgical industry, the production conditions change quickly (such as the sudden need for a large number of pneumatic actions during steel rolling), and the variable frequency drive module may find it difficult to adapt to the frequent transient flow changes in time, which leads to a certain degree of energy loss and delayed response. Summary of the invention

[0004] In order to solve the deficiencies in the prior art, the purpose of the present invention is to solve the above-mentioned defects and further propose a centrifugal air compressor station control method and device based on load characteristics and AI optimization.

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of the present invention discloses a centrifugal air compressor station control method based on load characteristics and AI optimization, the method comprising: Acquire environmental parameters in a steel metallurgical plant through a multi-channel sensor, and pre-process the noise of the environmental parameters based on notch, filtering, and segmented time window interception technology to obtain a pre-processed signal matrix; Extracting filter resistance characteristics by using a fitting model of dust concentration and filter resistance change in the environmental parameters, and extracting load spectrum characteristics corresponding to the inverter load response by calling a common space mode algorithm to generate a feature matrix; The filter fatigue index is calculated based on the relationship between the filter resistance and the dust concentration, and the load speed regulation response index is calculated based on the inverter load characteristics to comprehensively calculate the system fatigue index; Using the feature matrix as an input of a convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector; The long short-term memory network is trained by the fused feature vector, and the trained long short-term memory network is called to perform time series prediction on the fatigue trend of the system according to the historical environmental parameters, so as to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station; The environmental parameters include dust concentration, temperature, humidity, air flow, filter resistance and inverter load response in the environment of the steel metallurgical plant.

[0007] Furthermore, the environmental parameters in the iron and steel metallurgical plant are obtained by multi-channel sensors, and the noise of the environmental parameters is preprocessed based on notch, filtering and segmented time window interception technology to obtain a preprocessed signal matrix, including: The dust concentration in the environment of the iron and steel metallurgical plant is monitored in real time by a particle concentration sensor, the temperature and humidity in the environment of the iron and steel metallurgical plant are monitored in real time by a temperature sensor and a humidity sensor, and the air flow, filter resistance and inverter load response are monitored in real time by a flow meter, a pressure sensor and an inverter built-in sensor; Converting the analog signal output by the multi-channel sensor into a digital signal, and filtering the digital signal using an FIR filter to filter out noise signals in the digital signal; Dividing the continuous filtered digital signal into a plurality of time windows according to fixed time intervals to obtain the signal matrix; Wherein, the multi-channel sensor includes the particle concentration sensor, temperature sensor, humidity sensor, flow meter, pressure sensor and frequency converter built-in sensor.

[0008] Furthermore, the filter resistance characteristics are extracted by using the fitting model of the dust concentration and filter resistance change in the environmental parameters, and the load spectrum characteristics corresponding to the inverter load response are extracted by calling the common space mode algorithm to generate a feature matrix, including: A fitting model between filter resistance and dust concentration is established based on the average filter resistance and dust concentration in the preprocessed signal matrix, and the filter resistance characteristics are extracted through the fitting model. The expression of the fitting model is:

[0009] In the formula, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitting model, which is used to indicate the influence of dust concentration change on filter resistance. is the intercept of the fitting model, which is used to represent the filter resistance when there is no dust concentration. The intercept and slope are obtained by fitting the environmental data of multiple time windows by the least square method; The filter efficiency decreasing trend of each time window is calculated based on the filter resistance characteristics, and a filter resistance characteristic vector is output based on the filter efficiency decreasing trend, wherein each element in the filter resistance characteristic vector corresponds to the filter efficiency decreasing trend of a time window.

[0010] Further, the method of extracting filter resistance characteristics by using the fitting model of the dust concentration and filter resistance change in the environmental parameters, and extracting load spectrum characteristics corresponding to the inverter load response by calling the common space mode algorithm to generate a feature matrix also includes: Dividing the inverter load feature vector in the preprocessed signal matrix into a plurality of overlapping sub-windows, each sub-window containing a plurality of continuous time window data, and performing a fast Fourier transform on each sub-window to obtain a frequency domain representation; Calculating the power spectral density of each spectrum in the frequency domain representation, and extracting the spatial pattern associated with the load response from the PSD of all sub-windows by the OVR-CSP algorithm to generate a dynamic load feature matrix to output the inverter load spectrum feature vector; The characteristic matrix is ​​composed of the filter resistance characteristic vector and the inverter load spectrum characteristic vector.

[0011] Furthermore, the filter fatigue index is calculated according to the relationship between the filter resistance and the dust concentration, and the load speed regulation response index is calculated based on the inverter load characteristics to comprehensively calculate the system fatigue index, including: The filter fatigue index is calculated based on the fitting model, and a filter fatigue index vector is obtained based on the filter fatigue index. The filter fatigue index vector is composed of a plurality of filter fatigue indexes, and the number of the filter fatigue indexes is the total number of time windows. The calculation expression of the filter fatigue index is:

[0012] In the formula, is the filter fatigue index of the mth time window, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitted model, is the intercept of the fitted model; The inverter load spectrum characteristics are extracted for each time window through the OVR-CSP algorithm, and the weighted sum of the inverter load spectrum characteristics is calculated according to the preset weight coefficient to obtain the load speed regulation response index. The system fatigue index is the product of the filter fatigue index and the load speed regulation response index. The calculation expression of the load speed regulation response index is:

[0013] In the formula, is the load speed regulation response index of the mth time window, is the number of extracted inverter load spectrum features, For the The weight coefficient of the frequency spectrum characteristics of the inverter load, is the first The amplitude of the frequency spectrum characteristic of the inverter load.

[0014] Furthermore, the step of using the feature matrix as an input of a convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector includes: Converting the system fatigue index vector formed by the system fatigue index into a two-matrix form of a load convolution operation, and calling the convolution layer of the convolutional neural network in combination with the ReLU activation function to perform a convolution operation on the system fatigue index vector in the two-matrix form to obtain a convolution output of any filter at any position; Calling the pooling layer in combination with the ReLU activation function to perform mean pooling on the system fatigue index vector, obtaining a pooling output of any filter at any point in the pooling window, wherein the pooling output is an intermediate representation of the fused feature vector, and flattening and connecting the pooled feature maps to obtain the fused feature vector; The fused feature vector contains comprehensive information corresponding to the filter resistance and load speed regulation.

[0015] Furthermore, the fused feature vector is used to train the long short-term memory network, and the trained long short-term memory network is called to perform time series prediction on the fatigue trend of the system according to historical environmental parameters to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station, including: The LSTM model is trained by using the historical fusion feature vector corresponding to the historical environment parameter and the corresponding system fatigue index, and the mean square error is used as the loss function of the LSTM model, and the Adam optimizer is used as the optimization algorithm to minimize the loss function, so as to obtain a trained LSTM model; The historical fusion feature vector sequence with time order corresponding to the historical environmental parameters is used as the input of the trained LSTM model to output the system fatigue index vector of the first length time step in the future, and the filter maintenance recommendation vector is output based on the system fatigue index vector to optimize the energy saving control of the centrifugal air compressor station; Among them, the prediction expression of the trained LSTM model is:

[0016] In the formula, is the predicted system fatigue index vector, with a length of , is the length of the predicted future time step, is the time step length or history window size of the LSTM model, For the prediction time.

[0017] The second aspect of the present invention discloses a centrifugal air compressor station energy-saving optimization control device, the device comprising: A data preprocessing module is used to obtain environmental parameters in the steel metallurgical plant through a multi-channel sensor, and preprocess the noise of the environmental parameters based on notch, filtering and segmented time window interception technology to obtain a preprocessed signal matrix; A feature extraction module, used to extract filter resistance characteristics by using a fitting model of dust concentration and filter resistance change in the environmental parameters, and to extract load spectrum characteristics corresponding to the inverter load response by calling a common space mode algorithm to generate a feature matrix; A fatigue index calculation module is used to calculate the filter fatigue index based on the relationship between the filter resistance and the dust concentration, and to calculate the load speed regulation response index based on the inverter load characteristics, so as to comprehensively calculate the system fatigue index; A feature fusion module, used to use the feature matrix as an input of a convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector; An optimization control strategy generation module is used to train the long short-term memory network through the fused feature vector, and call the trained long short-term memory network to perform time series prediction on the system fatigue trend according to historical environmental parameters, so as to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station; The environmental parameters include dust concentration, temperature, humidity, air flow, filter resistance and inverter load response in the environment of the steel metallurgical plant.

[0018] The third aspect of the present invention discloses a terminal, including a processor and a storage medium; the characteristics are: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method described in the first aspect.

[0019] The fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0020] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages: The environmental parameters in the iron and steel metallurgical plant are obtained through multi-channel sensors, and the noise of the environmental parameters is preprocessed based on notch, filtering and segmented time window interception technology to obtain the preprocessed signal matrix. The filter resistance characteristics are extracted using the fitting model of the dust concentration and filter resistance changes in the environmental parameters, and the common space mode algorithm is called to extract the load spectrum characteristics corresponding to the inverter load response to generate a feature matrix. Subsequently, the filter fatigue index is calculated based on the relationship between the filter resistance and the dust concentration, and the load speed regulation response index is calculated based on the inverter load characteristics to comprehensively calculate the system fatigue index. After that, the feature matrix is ​​used as the input of the convolutional neural network to perform feature fusion on the feature matrix and output the fused feature vector. Finally, the long short-term memory network is trained by fusing the feature vector, and the trained long short-term memory network is called to perform time series prediction of the system fatigue trend based on the historical environmental parameters to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station. This method divides the environmental parameters in the steel metallurgical plant into time windows, combines convolutional neural networks and long short-term memory networks to extract corresponding features and predict future operation optimization strategies of time series, so that relevant maintenance can be performed before a fault occurs, reducing the maintenance frequency, and providing more sufficient maintenance time for frequent transient flow changes, reducing energy consumption and providing a faster response. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the centrifugal air compressor station control method based on load characteristics and AI optimization provided by the present invention. DETAILED DESCRIPTION

[0022] The present application is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present application.

[0023] like Figure 1 As shown, in one embodiment, a centrifugal air compressor station control method based on load characteristics and AI optimization includes the following steps: Step S110, obtaining environmental parameters in the steel metallurgical plant through a multi-channel sensor, and preprocessing the noise of the environmental parameters based on notch, filtering and segmented time window interception technology to obtain a preprocessed signal matrix.

[0024] Among them, environmental parameters include dust concentration, temperature, humidity, air flow, filter resistance and inverter load response in the environment of the steel metallurgical plant.

[0025] In some embodiments, the centrifugal air compressor station control method based on load characteristics and AI optimization provided by the present invention obtains environmental parameters in a steel metallurgical plant through a multi-channel sensor, and pre-processes the noise of the environmental parameters based on notch filtering and segmented time window interception technology to obtain a pre-processed signal matrix, which specifically includes the following steps: Step S111, monitor the dust concentration in the environment of the steel metallurgical plant in real time through a particle concentration sensor, monitor the temperature and humidity in the environment of the steel metallurgical plant in real time through a temperature sensor and a humidity sensor, and monitor the air flow, filter resistance and inverter load response in real time through a flow meter, a pressure sensor and a built-in sensor of the inverter.

[0026] Step S112, converting the analog signal output by the multi-channel sensor into a digital signal, and filtering the digital signal using an FIR filter to filter out noise signals in the digital signal.

[0027] Step S113, dividing the continuous filtered digital signal into multiple time windows according to fixed time intervals to obtain a signal matrix.

[0028] Among them, the multi-channel sensors include particle concentration sensors, temperature sensors, humidity sensors, flow meters, pressure sensors and inverter built-in sensors.

[0029] Step S120, extracting filter resistance characteristics using a fitting model of dust concentration and filter resistance change in environmental parameters, and calling a common space mode algorithm to extract load spectrum characteristics corresponding to the inverter load response to generate a feature matrix.

[0030] In some embodiments, the centrifugal air compressor station control method based on load characteristics and AI optimization provided by the present invention uses a fitting model of dust concentration and filter resistance change in environmental parameters to extract filter resistance characteristics, and calls a common space mode algorithm to extract load spectrum characteristics corresponding to the inverter load response to generate a feature matrix, specifically including the following steps: Step S121, a fitting model between filter resistance and dust concentration is established based on the average value of filter resistance and the average value of dust concentration in the preprocessed signal matrix, and the filter resistance characteristics are extracted through the fitting model. The expression of the fitting model is:

[0031] In the formula, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitting model, which is used to indicate the influence of dust concentration change on filter resistance. is the intercept of the fitted model, which is used to represent the filter resistance in the absence of dust concentration. The intercept and slope are obtained by fitting the environmental data of multiple time windows by the least squares method.

[0032] Step S122, calculating the filtration efficiency decreasing trend of each time window based on the filter resistance characteristics, and outputting a filter resistance characteristic vector based on the filtration efficiency decreasing trend, wherein each element in the filter resistance characteristic vector corresponds to the filtration efficiency decreasing trend of a time window.

[0033] In some embodiments, the centrifugal air compressor station control method based on load characteristics and AI optimization provided by the present invention uses a fitting model of dust concentration and filter resistance change in environmental parameters to extract filter resistance characteristics, and calls a common space mode algorithm to extract load spectrum characteristics corresponding to the inverter load response to generate a feature matrix, which specifically also includes the following steps: Step S123, dividing the inverter load feature vector in the preprocessed signal matrix into a plurality of overlapping sub-windows, each sub-window containing a plurality of continuous time window data, and performing a fast Fourier transform on each sub-window to obtain a frequency domain representation.

[0034] Step S124, calculate the power spectral density of each spectrum in the frequency domain representation, and extract the spatial pattern associated with the load response from the PSD of all sub-windows through the OVR-CSP algorithm to generate a dynamic load feature matrix to output the inverter load spectrum feature vector.

[0035] The characteristic matrix is ​​composed of the filter resistance characteristic vector and the inverter load spectrum characteristic vector.

[0036] Step S130, calculating the filter fatigue index according to the relationship between the filter resistance and the dust concentration, and calculating the load speed regulation response index based on the inverter load characteristics, so as to comprehensively calculate the system fatigue index.

[0037] In some embodiments, the centrifugal air compressor station control method based on load characteristics and AI optimization provided by the present invention calculates the filter fatigue index according to the relationship between the filter resistance and the dust concentration, and calculates the load speed regulation response index based on the inverter load characteristics to comprehensively calculate the system fatigue index, specifically including the following steps: Step S131, calculate the filter fatigue index based on the fitting model, and obtain the filter fatigue index vector based on the filter fatigue index. The filter fatigue index vector is composed of multiple filter fatigue indexes, and the number of filter fatigue indexes is the total number of time windows. The calculation expression of the filter fatigue index is:

[0038] In the formula, is the filter fatigue index of the mth time window, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitted model, is the intercept of the fitted model.

[0039] Step S132, extract the inverter load spectrum characteristics for each time window through the OVR-CSP algorithm, and calculate the weighted sum of the inverter load spectrum characteristics according to the preset weight coefficient to obtain the load speed regulation response index. The system fatigue index is the product of the filter fatigue index and the load speed regulation response index. The calculation expression of the load speed regulation response index is:

[0040] In the formula, is the load speed regulation response index of the mth time window, is the number of extracted inverter load spectrum features, For the The weight coefficient of the frequency spectrum characteristics of the inverter load, is the first The amplitude of the frequency spectrum characteristic of the inverter load.

[0041] Step S140, using the feature matrix as the input of the convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector.

[0042] In some embodiments, the centrifugal air compressor station control method based on load characteristics and AI optimization provided by the present invention uses the feature matrix as the input of the convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector, specifically including the following steps: Step S141, converting the system fatigue index vector composed of the system fatigue index into a two-matrix form of a load convolution operation, and calling the convolution layer of the convolutional neural network in combination with the ReLU activation function to perform a convolution operation on the system fatigue index vector in the two-matrix form to obtain the convolution output of any filter at any position.

[0043] Step S142, call the pooling layer in combination with the ReLU activation function to perform mean pooling on the system fatigue index vector, obtain the pooling output of any filter at any point in the pooling window, the pooling output is the intermediate representation of the fused feature vector, and the pooled feature map is flattened and connected to obtain the fused feature vector.

[0044] Among them, the fused feature vector contains the comprehensive information corresponding to the filter resistance and load speed regulation.

[0045] Step S150, training the long short-term memory network by fusing feature vectors, and calling the trained long short-term memory network to perform time series prediction on system fatigue trend according to historical environmental parameters, so as to generate a control strategy for energy-saving optimization control of centrifugal air compressor station.

[0046] In some embodiments, the centrifugal air compressor station control method based on load characteristics and AI optimization provided by the present invention trains the long short-term memory network by fusing feature vectors, and calls the trained long short-term memory network to perform time series prediction on the system fatigue trend according to historical environmental parameters to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station, which specifically includes the following steps: Step S151, train the LSTM model through the historical fusion feature vector corresponding to the historical environment parameters and its corresponding system fatigue index, use the mean square error as the loss function of the LSTM model, and use the Adam optimizer as the optimization algorithm to minimize the loss function, so as to obtain a trained LSTM model.

[0047] Step S152, using the time-ordered historical fusion feature vector sequence corresponding to the historical environmental parameters as the input of the trained LSTM model to output the system fatigue index vector of the first length time step in the future, and output the filter maintenance recommendation vector based on the system fatigue index vector to optimize the energy-saving control of the centrifugal air compressor station.

[0048] Among them, the prediction expression of the trained LSTM model is:

[0049] In the formula, is the predicted system fatigue index vector, with a length of , is the length of the predicted future time step, is the time step length or history window size of the LSTM model, For the prediction time.

[0050] The above-mentioned centrifugal air compressor station control method based on load characteristics and AI optimization, by dividing the environmental parameters in the steel metallurgical plant into time windows, combines convolutional neural networks and long short-term memory networks to extract corresponding features and predict future operation optimization strategies of time series. It can perform relevant maintenance before a fault occurs, reducing the maintenance frequency, and providing more sufficient maintenance time for frequent transient flow changes, reducing energy consumption losses and responding faster.

[0051] In a specific embodiment, the centrifugal air compressor station control method based on load characteristics and AI optimization provided by the present invention includes steps 1 to 5: Step 1: Data collection and preprocessing.

[0052] Physical mechanism framework: Based on fluid dynamics and environmental parameter modeling, real-time collection of dust concentration (0.5~1 mg / m³), temperature, humidity, air flow, filter resistance, inverter load response and other parameters in the steel and metallurgical plant environment. Data is collected through multi-channel sensors (including particle concentration sensors and pressure sensors), and noise is processed by combining notch, filtering and segmented time window interception technology to finally obtain a pre-processed signal matrix.

[0053] Specifically, the sensors are installed and data is collected first. The dust concentration is monitored in real time by the particle concentration sensor, the unit is mg / m³, and the range is 0.5~1mg / m³. The temperature is monitored in real time by the ambient temperature sensor, the unit is ℃, and the range is 20~50℃. The humidity is monitored in real time by the ambient humidity sensor, the unit is %RH, and the range is 30%~70%RH. The intake flow is monitored in real time by the flow meter, the unit is m³ / s, and the range is 0.05~0.5m³ / s. The pressure and filter resistance are monitored in real time by the pressure sensor, the unit is Pa, and the pressure range is 10 5 ~10 6 Pa, filter resistance range is 10 2 ~10 3 Pa. The inverter load uses the inverter built-in sensor to obtain its load information.

[0054] In this embodiment, the data collected by the multi-pass sensor is firstly A / D converted. The function of A / D conversion is to convert the analog signal output by the sensor into a digital signal, which is convenient for subsequent digital processing and analysis. In the A / D conversion process, the analog signal collected by each sensor is A / D converted to obtain its corresponding digital signal. Subsequently, the collected environmental parameters and operating parameters are filtered, the purpose of which is to reduce high-frequency noise and transient pulse interference to obtain a smooth and stable signal. The filtering method uses a finite impulse response (FIR) low-pass filter to process the relevant digital signal after A / D conversion. In the filtering process, the filter length and filter coefficient are selected according to the system requirements. For example, the window function method, including the Hamming window and the Blackman window, is used to design the FIR low-pass filter. In addition, it is necessary to ensure that the filter meets the corresponding normalization conditions.

[0055] Subsequently, the time window is intercepted, and the continuous filtered signal is divided into multiple time windows at fixed time intervals for subsequent feature extraction and analysis. In the process of intercepting the time window, it is first necessary to determine the window parameters, including the window length, the sampling period, and the number of sampling points contained in each window. Secondly, each filtered signal is used as a time window in chronological order with the same number of sampling points, for example, every N sampling points in chronological order are a time window. Finally, statistical feature extraction is performed, the average value of each parameter in each time window is calculated, and a pre-processed signal matrix is ​​constructed. The signal matrix includes row index and column index, the row index represents the number of the time window, and the column index represents the average value of the corresponding parameter in any numbered time window.

[0056] Step 2: Feature extraction and state modeling.

[0057] Physical mechanism framework: Using the fitting model of dust concentration and filtration resistance change, the filter resistance characteristics are extracted and the filtration efficiency decline trend is calculated.

[0058] Algorithm OVR-CSP (Common Space Mode): For the inverter load signal, extract the spectrum characteristics related to the load response and generate a dynamic load feature matrix.

[0059] The filter resistance characteristics and load spectrum characteristics are comprehensively extracted to generate a corresponding feature matrix (used to include the filter resistance characteristics and load spectrum characteristics) for subsequent status monitoring.

[0060] Specifically, a fitting model between the filter resistance and the dust concentration is established by using the average value of the filter resistance and the average value of the dust concentration in the preprocessed signal matrix to extract the filter resistance characteristics.

[0061] The expression of the fitting model is:

[0062] In the formula, is the average filter resistance value in the mth time window, in Pa; is the average dust concentration in the mth time window, in mg / m³; is the slope of the fitting model, which is used to indicate the influence of dust concentration change on filter resistance. is the intercept of the fitted model, which is used to represent the filter resistance in the absence of dust concentration. The intercept and slope are obtained by fitting the environmental data of multiple time windows by the least squares method.

[0063] Then, the filtration efficiency decline trend of each time window was calculated based on the above fitting model. , whose expression is:

[0064] In the formula, is the decreasing trend of the filtration efficiency in the mth time window, expressed in dimensionless percentage, which reflects the increase in resistance of the filter due to dust accumulation in actual operation, and The larger the value is, the more obvious the decrease in filtration efficiency is. Other variables are the same as in the fitted model.

[0065] It should be noted that the filter resistance characteristic vector is composed of multiple filtration efficiency decrease trends, and each element in the filter resistance characteristic vector corresponds to a filtration efficiency decrease trend in a time window.

[0066] In this embodiment, the OVR-CSP algorithm is a feature extraction method for multi-category signal classification, which is mainly used to extract spatial patterns from multi-channel signals, and is mainly used to extract spectral features related to load response from inverter load signals. In the process of spectral feature extraction, the inverter load feature vector is first divided into multiple overlapping sub-windows, each sub-window containing multiple continuous time window data. Secondly, each sub-window is subjected to a fast Fourier transform to obtain a frequency domain representation, which is composed of the spectra of multiple sub-windows. Subsequently, the power spectral density of each spectrum is calculated, and then the spatial pattern related to the load response is extracted from the PSD of all sub-windows by the OVR-CSP algorithm to generate a dynamic load feature matrix, and then the inverter load spectrum feature vector is output, which is composed of multiple spectrum features, each element of which represents a spectrum feature.

[0067] The obtained filter resistance characteristics and inverter load spectrum characteristics are combined to form the final feature matrix for subsequent status monitoring. The dimension of the final feature matrix is ​​N×2, column 1 represents the filter resistance characteristics of multiple time windows, and column 2 represents the inverter load spectrum characteristics of multiple time windows.

[0068] In this embodiment, in order to improve the stability and accuracy of the model, the feature matrix needs to be standardized. The standardized processing expression is:

[0069] In the formula, is the standardized feature matrix, is the feature matrix before standardization, is the mean vector of the feature matrix, is the standard deviation vector of the feature matrix.

[0070] Step 3, fatigue state monitoring model.

[0071] Physical mechanism framework: The filter fatigue index (F_resist) is calculated in real time through the relationship formula between filter resistance and dust concentration. The load speed regulation response index (F_response) is calculated based on the inverter load characteristics to reflect the load lag degree and comprehensively calculate the system fatigue index.

[0072] Specifically, the filter fatigue index is used to measure the degree of filter fatigue caused by dust accumulation. The index is based on the relationship between the change in filter resistance and dust concentration, and is calculated in real time through a fitting model. The calculation formula for the filter fatigue index is:

[0073] In the formula, is the filter fatigue index of the mth time window, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitted model, is the intercept of the fitted model.

[0074] It should be noted that the filter fatigue index vector is composed of multiple filter fatigue indices, each element of which corresponds to the filter fatigue index of a time window, and the number of rows of the feature matrix is ​​the total number of time windows.

[0075] In this embodiment, the load speed response index is used to measure the response lag of the inverter when the load changes. The index is calculated based on the frequency spectrum characteristics of the inverter load through the frequency spectrum characteristics extracted by the algorithm OVR-CSP. The load speed response index reflects the response speed and stability of the system by analyzing the key frequency components in the frequency spectrum characteristics of the inverter load. Its calculation expression is:

[0076] In the formula, is the load speed regulation response index of the mth time window, and is the weighted sum of the spectrum features in the time window, is the number of extracted inverter load spectrum features, For the The weight coefficient of the frequency spectrum characteristics of the inverter load, and , is the first The amplitude of the frequency spectrum characteristic of the inverter load.

[0077] It should be noted that the load speed regulation response index vector is composed of multiple load speed regulation response indexes, each element of which corresponds to a response index of a time window.

[0078] The system fatigue index comprehensively considers the fatigue degree of the filter and the response lag of the inverter, and is used to comprehensively evaluate the health status of the air compressor station system. and Similarly, the system fatigue index vector is composed of multiple system fatigue indices, and each element corresponds to the system fatigue index of a time window.

[0079] Step 4: feature fusion and convolution processing.

[0080] CNN (Convolutional Neural Network): Input the filter and load response feature matrix into CNN for multimodal fusion: Convolutional layer: 64 filters (kernel size 25×1) are used to extract feature maps and capture multi-scale interactions.

[0081] Pooling layer: Mean pooling reduces data dimensions and reduces computational complexity.

[0082] The output fused feature vector integrates the filter and inverter load characteristics for optimizing control strategy generation.

[0083] Specifically, Convolutional Neural Network (CNN) is a deep learning model that is widely used in the fields of image and signal processing. Its core is to extract local features through the convolution layer and reduce the data dimension and computational complexity through the pooling layer. In order to meet the input requirements of CNN, the system fatigue index vector needs to be converted into a two-dimensional matrix form suitable for convolution operations. The activation function selects the ReLU activation function to introduce nonlinearity so that CNN can learn complex patterns.

[0084] In this embodiment, the pooling type of the pooling layer is mean pooling, and the pooling output of any filter at any point in the pooling window is output through the pooling operation, and the output of the pooling layer is the intermediate representation of the fused feature vector, which is used to capture multi-scale interactions. The fused feature vector is obtained by flattening and concatenating all pooled feature maps for subsequent optimization control strategy generation. As a high-dimensional feature representation after multimodal fusion, the fused feature vector contains comprehensive information about filter resistance and load speed regulation response.

[0085] In addition, the fused feature vector needs to be normalized to eliminate the dimensional effects between different features, improve the stability and convergence speed of model training, and the dimension of the normalized fused feature vector remains unchanged. The normalized fused feature vector contains comprehensive information about the filter resistance and the inverter load speed regulation response, which is suitable for subsequent optimization control strategy generation. Each element is a normalized pooled feature.

[0086] Step 5: Generate an optimized control strategy.

[0087] LSTM (Long Short-Term Memory Network): Use historical data to make time series predictions on system fatigue trends and generate optimized control strategies in advance: Filter maintenance advice: If the filter resistance exceeds the threshold, you will be prompted to clean or replace the filter in advance.

[0088] Inverter frequency optimization: Use LSTM to predict load fluctuation trends, dynamically adjust the inverter operating frequency, and reduce lag response.

[0089] Energy-saving mode switching: Dynamically adjust the system operating mode (such as entering energy-saving mode when the load decreases).

[0090] Specifically, the Long Short-Term Memory (LSTM) network is a special type of recurrent neural network that can effectively capture and learn long-range dependencies in time series data, and is used to predict future trends in the system fatigue index, thereby generating an optimized control strategy in advance.

[0091] The LSTM network consists of four gates: input gate, forget gate, output gate, and cell state. In the LSTM network prediction process, the trained LSTM model is used to input the historical fusion feature vector sequence to predict the system fatigue index of the next T time steps. The prediction expression is:

[0092] In the formula, is the predicted system fatigue index vector, with a length of , is the length of the predicted future time step, is the time step length or history window size of the LSTM model, For the prediction time.

[0093] During the LSTM network training process, the historical fusion feature vector and the corresponding system fatigue index are used to train the LSTM network model. The loss function adopts the mean square error, and the optimization algorithm adopts the Adam optimizer to minimize the loss function. After prediction, the system fatigue index vector is output, and then the filter maintenance recommendation vector is output, in which each element is a specific maintenance operation recommendation.

[0094] The load speed response index can be extracted from the predicted system fatigue index, and the inverter's motion frequency can be dynamically adjusted according to the load speed response index. In addition, the predicted system fatigue index can be directly used to determine the energy-saving mode switch, or the system operation mode can be dynamically adjusted according to the system fatigue index, and then the corresponding energy-saving mode switching vector and inverter frequency optimization vector are output, each element of which is a specific frequency adjustment value, and finally energy-saving optimization control is achieved.

[0095] The optimization control strategy is a comprehensive control instruction generated based on the prediction results and set thresholds, including filter maintenance recommendations, inverter frequency adjustment, and energy-saving mode switching. The various sub-strategies can be integrated into the final optimization control strategy. For example, when the filter prompts maintenance, arrange to clean or replace the filter; optimize energy use and system response by dynamically adjusting the operating frequency of the inverter; switch the system operating mode or normal mode according to the control instructions.

[0096] The energy-saving optimization control device for a centrifugal air compressor station provided by the present invention is described below. The energy-saving optimization control device for a centrifugal air compressor station described below and the control method for a centrifugal air compressor station based on load characteristics and AI optimization described above can be referenced to each other.

[0097] In one embodiment, a centrifugal air compressor station energy-saving optimization control device includes a data preprocessing module, a feature extraction module, a fatigue index calculation module, a feature fusion module and an optimization control strategy generation module.

[0098] The data preprocessing module is used to obtain the environmental parameters in the steel metallurgical plant through multi-channel sensors, and preprocess the noise of the environmental parameters based on notch, filtering and segmented time window interception technology to obtain the preprocessed signal matrix.

[0099] The feature extraction module is used to extract the filter resistance characteristics by using the fitting model of the dust concentration and filter resistance change in the environmental parameters, and to call the common space mode algorithm to extract the load spectrum characteristics corresponding to the inverter load response to generate a feature matrix.

[0100] The fatigue index calculation module is used to calculate the filter fatigue index according to the relationship between the filter resistance and the dust concentration, and calculate the load speed regulation response index based on the inverter load characteristics, so as to comprehensively calculate the system fatigue index.

[0101] The feature fusion module is used to take the feature matrix as the input of the convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector.

[0102] The optimization control strategy generation module is used to train the long short-term memory network by fusing feature vectors, and call the trained long short-term memory network to perform time series prediction on the system fatigue trend according to historical environmental parameters, so as to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station.

[0103] Among them, environmental parameters include dust concentration, temperature, humidity, air flow, filter resistance and inverter load response in the environment of the steel metallurgical plant.

[0104] In this embodiment, the energy-saving optimization control device for a centrifugal air compressor station provided by the present invention, the data preprocessing module is specifically used for: The dust concentration in the environment of the steel and metallurgical plant is monitored in real time through the particle concentration sensor, the temperature and humidity in the environment of the steel and metallurgical plant are monitored in real time through the temperature sensor and humidity sensor, and the air flow, filter resistance and inverter load response are monitored in real time through the flow meter, pressure sensor and inverter built-in sensor.

[0105] The analog signal output by the multi-channel sensor is converted into a digital signal, and the digital signal is filtered using a FIR filter to filter out the noise signal in the digital signal.

[0106] The continuous filtered digital signal is divided into multiple time windows according to fixed time intervals to obtain a signal matrix.

[0107] Among them, the multi-channel sensors include particle concentration sensors, temperature sensors, humidity sensors, flow meters, pressure sensors and inverter built-in sensors.

[0108] In this embodiment, the centrifugal air compressor station energy-saving optimization control device provided by the present invention, the feature extraction module is specifically used for: Based on the average filter resistance and dust concentration in the preprocessed signal matrix, a fitting model between filter resistance and dust concentration is established, and the filter resistance characteristics are extracted through the fitting model. The expression of the fitting model is:

[0109] In the formula, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitting model, which is used to indicate the influence of dust concentration change on filter resistance. is the intercept of the fitted model, which is used to represent the filter resistance in the absence of dust concentration. The intercept and slope are obtained by fitting the environmental data of multiple time windows by the least squares method.

[0110] The filter efficiency decreasing trend of each time window is calculated based on the filter resistance characteristics, and a filter resistance feature vector is output based on the filter efficiency decreasing trend. Each element in the filter resistance feature vector corresponds to the filter efficiency decreasing trend of a time window.

[0111] In this embodiment, the centrifugal air compressor station energy-saving optimization control device provided by the present invention, the feature extraction module is specifically used for: The inverter load feature vector in the preprocessed signal matrix is ​​divided into a plurality of overlapping sub-windows, each of which contains a plurality of continuous time window data, and a fast Fourier transform is performed on each sub-window to obtain a frequency domain representation.

[0112] The power spectral density of each spectrum in the frequency domain representation is calculated, and the spatial patterns associated with the load response are extracted from the PSD of all sub-windows through the OVR-CSP algorithm to generate a dynamic load feature matrix to output the inverter load spectrum feature vector.

[0113] The characteristic matrix is ​​composed of the filter resistance characteristic vector and the inverter load spectrum characteristic vector.

[0114] In this embodiment, the energy-saving optimization control device for a centrifugal air compressor station provided by the present invention, the fatigue index calculation module is specifically used for: The filter fatigue index is calculated based on the fitting model, and a filter fatigue index vector is obtained based on the filter fatigue index. The filter fatigue index vector is composed of multiple filter fatigue indexes, and the number of filter fatigue indexes is the total number of time windows. The calculation expression of the filter fatigue index is:

[0115] In the formula, is the filter fatigue index of the mth time window, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitted model, is the intercept of the fitted model.

[0116] The OVR-CSP algorithm is used to extract the inverter load spectrum characteristics for each time window, and the weighted sum of the inverter load spectrum characteristics is calculated according to the preset weight coefficient to obtain the load speed regulation response index. The system fatigue index is the product of the filter fatigue index and the load speed regulation response index. The calculation expression of the load speed regulation response index is:

[0117] In the formula, is the load speed regulation response index of the mth time window, is the number of extracted inverter load spectrum features, For the The weight coefficient of the frequency spectrum characteristics of the inverter load, is the first The amplitude of the frequency spectrum characteristic of the inverter load.

[0118] In this embodiment, the centrifugal air compressor station energy-saving optimization control device provided by the present invention, the feature fusion module is specifically used for: The system fatigue index vector composed of the system fatigue index is converted into a two-matrix form of load convolution operation, and the convolution layer of the convolutional neural network is called in combination with the ReLU activation function to perform a convolution operation on the system fatigue index vector in the two-matrix form to obtain the convolution output of any filter at any position.

[0119] The pooling layer is called in combination with the ReLU activation function to perform mean pooling on the system fatigue index vector to obtain the pooling output of any filter at any point in the pooling window. The pooling output is the intermediate representation of the fused feature vector, and the pooled feature map is flattened and connected to obtain the fused feature vector.

[0120] Among them, the fused feature vector contains the comprehensive information corresponding to the filter resistance and load speed regulation.

[0121] In this embodiment, the energy-saving optimization control device for a centrifugal air compressor station provided by the present invention, the optimization control strategy generation module is specifically used for: The LSTM model is trained by the historical fusion feature vector corresponding to the historical environmental parameters and its corresponding system fatigue index, and the mean square error is used as the loss function of the LSTM model, and the Adam optimizer is used as the optimization algorithm to minimize the loss function to obtain a trained LSTM model.

[0122] The historical fusion feature vector sequence with time order corresponding to the historical environmental parameters is used as the input of the trained LSTM model to output the system fatigue index vector of the first length time step in the future, and the filter maintenance recommendation vector is output based on the system fatigue index vector to optimize the energy-saving control of the centrifugal air compressor station.

[0123] Among them, the prediction expression of the trained LSTM model is:

[0124] In the formula, is the predicted system fatigue index vector, with a length of , is the length of the predicted future time step, is the time step length or history window size of the LSTM model, For the prediction time.

[0125] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0126] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0127] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0128] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0129] Various aspects of the present disclosure are described herein with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of boxes in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0130] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0131] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0132] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A centrifugal air compressor station control method based on load characteristics and AI optimization, characterized in that: The method comprises: Acquire environmental parameters in a steel metallurgical plant through a multi-channel sensor, and pre-process the noise of the environmental parameters based on notch, filtering, and segmented time window interception technology to obtain a pre-processed signal matrix; Extracting filter resistance characteristics by using a fitting model of dust concentration and filter resistance change in the environmental parameters, and extracting load spectrum characteristics corresponding to the inverter load response by calling a common space mode algorithm to generate a feature matrix; The filter fatigue index is calculated based on the relationship between the filter resistance and the dust concentration, and the load speed regulation response index is calculated based on the inverter load characteristics to comprehensively calculate the system fatigue index; Using the feature matrix as an input of a convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector; The long short-term memory network is trained by the fused feature vector, and the trained long short-term memory network is called to perform time series prediction on the fatigue trend of the system according to the historical environmental parameters, so as to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station; The environmental parameters include dust concentration, temperature, humidity, air flow, filter resistance and inverter load response in the environment of the steel metallurgical plant.

2. The centrifugal air compressor station control method based on load characteristics and AI optimization according to claim 1 is characterized in that: The environmental parameters in the steel metallurgical plant are obtained by using a multi-channel sensor, and the noise of the environmental parameters is preprocessed based on notch, filtering and segmented time window interception technology to obtain a preprocessed signal matrix, including: The dust concentration in the environment of the iron and steel metallurgical plant is monitored in real time by a particle concentration sensor, the temperature and humidity in the environment of the iron and steel metallurgical plant are monitored in real time by a temperature sensor and a humidity sensor, and the air flow, filter resistance and inverter load response are monitored in real time by a flow meter, a pressure sensor and an inverter built-in sensor; Converting the analog signal output by the multi-channel sensor into a digital signal, and filtering the digital signal using an FIR filter to filter out noise signals in the digital signal; Dividing the continuous filtered digital signal into a plurality of time windows according to fixed time intervals to obtain the signal matrix; Wherein, the multi-channel sensor includes the particle concentration sensor, temperature sensor, humidity sensor, flow meter, pressure sensor and frequency converter built-in sensor.

3. The centrifugal air compressor station control method based on load characteristics and AI optimization according to claim 2 is characterized in that: The filter resistance characteristics are extracted by using the fitting model of the dust concentration and filter resistance change in the environmental parameters, and the load spectrum characteristics corresponding to the inverter load response are extracted by calling the common space mode algorithm to generate a feature matrix, including: A fitting model between filter resistance and dust concentration is established based on the average filter resistance and dust concentration in the preprocessed signal matrix, and the filter resistance characteristics are extracted through the fitting model. The expression of the fitting model is: ； In the formula, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitting model, which is used to indicate the influence of dust concentration change on filter resistance. is the intercept of the fitting model, which is used to represent the filter resistance when there is no dust concentration. The intercept and slope are obtained by fitting the environmental data of multiple time windows by the least square method; The filter efficiency decreasing trend of each time window is calculated based on the filter resistance characteristics, and a filter resistance characteristic vector is output based on the filter efficiency decreasing trend, wherein each element in the filter resistance characteristic vector corresponds to the filter efficiency decreasing trend of a time window.

4. The centrifugal air compressor station control method based on load characteristics and AI optimization according to claim 3 is characterized in that: The method further comprises: extracting the filter resistance characteristics by using the fitting model of the dust concentration and the filter resistance change in the environmental parameters, and extracting the load spectrum characteristics corresponding to the inverter load response by calling the common space mode algorithm to generate a characteristic matrix, and further comprising: Dividing the inverter load feature vector in the preprocessed signal matrix into a plurality of overlapping sub-windows, each sub-window containing a plurality of continuous time window data, and performing a fast Fourier transform on each sub-window to obtain a frequency domain representation; Calculating the power spectral density of each spectrum in the frequency domain representation, and extracting the spatial pattern associated with the load response from the PSD of all sub-windows by the OVR-CSP algorithm to generate a dynamic load feature matrix to output the inverter load spectrum feature vector; The characteristic matrix is ​​composed of the filter resistance characteristic vector and the inverter load spectrum characteristic vector.

5. The centrifugal air compressor station control method based on load characteristics and AI optimization according to claim 4 is characterized in that: The filter fatigue index is calculated based on the relationship between the filter resistance and the dust concentration, and the load speed regulation response index is calculated based on the inverter load characteristics to comprehensively calculate the system fatigue index, including: The filter fatigue index is calculated based on the fitting model, and a filter fatigue index vector is obtained based on the filter fatigue index. The filter fatigue index vector is composed of a plurality of filter fatigue indexes, and the number of the filter fatigue indexes is the total number of time windows. The calculation expression of the filter fatigue index is: ； In the formula, is the filter fatigue index of the mth time window, is the average filter resistance of the mth time window, is the average dust concentration in the mth time window, is the slope of the fitted model, is the intercept of the fitted model; The inverter load spectrum characteristics are extracted for each time window through the OVR-CSP algorithm, and the weighted sum of the inverter load spectrum characteristics is calculated according to the preset weight coefficient to obtain the load speed regulation response index. The system fatigue index is the product of the filter fatigue index and the load speed regulation response index. The calculation expression of the load speed regulation response index is: ； In the formula, is the load speed regulation response index of the mth time window, is the number of extracted inverter load spectrum features, For the The weight coefficient of the frequency spectrum characteristics of the inverter load, is the first The amplitude of the frequency spectrum characteristic of the inverter load.

6. The centrifugal air compressor station control method based on load characteristics and AI optimization according to claim 5 is characterized in that: The step of using the feature matrix as an input of a convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector includes: Converting the system fatigue index vector formed by the system fatigue index into a two-matrix form of a load convolution operation, and calling the convolution layer of the convolutional neural network in combination with the ReLU activation function to perform a convolution operation on the system fatigue index vector in the two-matrix form to obtain a convolution output of any filter at any position; Calling the pooling layer in combination with the ReLU activation function to perform mean pooling on the system fatigue index vector, obtaining a pooling output of any filter at any point in the pooling window, wherein the pooling output is an intermediate representation of the fused feature vector, and flattening and connecting the pooled feature maps to obtain the fused feature vector; The fused feature vector contains comprehensive information corresponding to the filter resistance and load speed regulation.

7. The centrifugal air compressor station control method based on load characteristics and AI optimization according to claim 6 is characterized in that: The long short-term memory network is trained by the fusion feature vector, and the trained long short-term memory network is called to perform time series prediction on the fatigue trend of the system according to the historical environmental parameters to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station, including: The LSTM model is trained by using the historical fusion feature vector corresponding to the historical environment parameter and the corresponding system fatigue index, and the mean square error is used as the loss function of the LSTM model, and the Adam optimizer is used as the optimization algorithm to minimize the loss function, so as to obtain a trained LSTM model; The historical fusion feature vector sequence with time order corresponding to the historical environmental parameters is used as the input of the trained LSTM model to output the system fatigue index vector of the first length time step in the future, and the filter maintenance recommendation vector is output based on the system fatigue index vector to optimize the energy saving control of the centrifugal air compressor station; Among them, the prediction expression of the trained LSTM model is: ； In the formula, is the predicted system fatigue index vector, with a length of , is the length of the predicted future time step, is the time step length or history window size of the LSTM model, For the prediction time.

8. A centrifugal air compressor station energy-saving optimization control device, characterized in that: The device comprises: A data preprocessing module is used to obtain environmental parameters in the steel metallurgical plant through a multi-channel sensor, and preprocess the noise of the environmental parameters based on notch, filtering and segmented time window interception technology to obtain a preprocessed signal matrix; A feature extraction module, used to extract filter resistance characteristics by using a fitting model of dust concentration and filter resistance change in the environmental parameters, and to extract load spectrum characteristics corresponding to the inverter load response by calling a common space mode algorithm to generate a feature matrix; A fatigue index calculation module is used to calculate the filter fatigue index based on the relationship between the filter resistance and the dust concentration, and to calculate the load speed regulation response index based on the inverter load characteristics, so as to comprehensively calculate the system fatigue index; A feature fusion module, used to use the feature matrix as an input of a convolutional neural network to perform feature fusion on the feature matrix and output a fused feature vector; An optimization control strategy generation module is used to train the long short-term memory network through the fused feature vector, and call the trained long short-term memory network to perform time series prediction on the system fatigue trend according to historical environmental parameters, so as to generate a control strategy for energy-saving optimization control of the centrifugal air compressor station; The environmental parameters include dust concentration, temperature, humidity, air flow, filter resistance and inverter load response in the environment of the steel metallurgical plant.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images