Operation control method of dust removal fan

Abstract

The invention relates to an operation control method of a dust removal fan, and belongs to the technical field of automatic control. Comprising the following steps of firstly establishing and forming a historical operation data set, then establishing a prediction model according to an FEDform time sequence prediction algorithm, inputting data in the historical operation data set into the dust removal fan operation parameter prediction model for model training, and finally establishing an objective function of dust removal fan operation control. And performing optimization calculation on the target function through a cuckoo search optimization algorithm to obtain an optimal negative pressure set value in the operation control time period. According to the method, a time sequence prediction algorithm in an FEDform mode is improved, a structure capable of extracting time domain information of fan operation data is added to an encoder part of an existing time sequence prediction algorithm, and meanwhile, a multi-query attention mechanism structure is adopted to replace a multi-head self-attention structure in the existing algorithm; therefore, the prediction result of the improved FEDform-form time series prediction algorithm is more accurate, and the optimization accuracy is improved.

Description

Technical Field

[0001] The invention relates to an operation control method of a dust removal fan, and belongs to the technical field of automatic control. Background Art

[0002] The blast furnace ironmaking process generates large amounts of flue gas and dust, containing harmful substances such as iron, sulfur, nitrogen, and oxygen, which pose serious risks to the environment and human health. To mitigate the emission of these harmful substances, dust removal fans have become indispensable equipment in blast furnace ironmaking systems. The main function of dust removal fans is to extract flue gas and dust from the blast furnace ironmaking system and transport them to dust collectors or other purification facilities for treatment, achieving effective separation and recovery. The purified flue gas is then discharged or reused. The operating state of the dust removal fan (i.e., dust removal efficiency and energy consumption level) is affected by the negative pressure value set before operation. Therefore, how to reasonably set the negative pressure value of the dust removal fan to control its operation so that the dust removal fan achieves the optimal operating state for the current dust removal requirements (i.e., the dust removal fan generates the lowest energy consumption while maintaining the required dust removal efficiency) is a key issue in the blast furnace ironmaking dust removal process.

[0003] Most existing technologies employ intelligent control methods. Based on the blast furnace ironmaking process conditions and environmental parameters, intelligent algorithms or models are used to automatically calculate a negative pressure setpoint. At this negative pressure setpoint, the dust removal fan's operating state is expected to achieve the optimal effect for the current dust removal requirements. Operational control is then implemented using a frequency converter or other control device based on the negative pressure setpoint. However, existing technologies have the following shortcomings: 1. These intelligent algorithms or models often overly focus on short-term or even ultra-short-term effects, ignoring long-term energy savings. Furthermore, they result in overly frequent manipulation of the frequency converter or other control device, significantly increasing the risk of failure of the control equipment. 2. Because the dust removal fan's operating state is not only affected by its own negative pressure value, but also by operating conditions, the equipment's operating environment, and blast furnace output, the intelligent algorithms employed in existing technologies often only consider the relationship between operating state and negative pressure. Consequently, there is a discrepancy between the calculated negative pressure setpoint and the actual optimal effect achieved. Existing technologies still have significant room for improvement in dust removal fan operational control. In order to overcome the above problems, it is necessary to develop a dust removal fan operation control method that can more comprehensively consider various influencing factors while taking into account long-term energy-saving benefits. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: how to control the working state of the dust removal fan over a long period of time, avoid multiple controls in a short period of time resulting in an increased probability of equipment failure, and at the same time improve the effect of the dust removal fan operation control.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is: a dust removal fan operation control method, comprising the following steps:

[0006] Step 1: First, collect all the historical operation data of the dust removal fan. The historical operation data are divided into four types: continuous operation data, discrete operation data, equipment capacity data and equipment environment data. The continuous operation data includes the operating parameters and negative pressure setting value of the dust removal fan during operation. The operating parameters include fan current, fan pressure difference and fan speed. The discrete operation data includes the switch status of the iron mouth top suction electric valve, iron mouth side suction electric valve and iron mouth slag skimmer electric valve when the dust removal fan is running. The equipment capacity data includes fan power consumption and blast furnace capacity. The equipment environment data includes date, holiday information, temperature and weather. Then, align the four types of historical operation data by the difference method. Finally, eliminate all unreasonable data in the historical operation data by the 3σ method, and collect the remaining historical operation data to form a historical operation data set.

[0007] Step 2: Establish a dust removal fan operating parameter prediction model based on the FEDformer time series prediction algorithm, as shown in the following formula (1):

[0008]

[0009] In formula (1), S en_f is the frequency domain sequence obtained by encoder decomposition in the FEDformer time series prediction algorithm; Decomp is the seasonality-trend decomposition algorithm; SCI (self-Channel Interaction) is the self-channel interaction network in the FEDformer time series prediction algorithm; the MQA (Multi-Query Attention) is the multi-query attention mechanism algorithm; the X en is an input data sequence input to the encoder; (Fourier Transform) is the Fourier transform algorithm; S en_t is the time domain sequence decomposed by the encoder in the FEDformer time series prediction algorithm; S en is the output of the encoder; S de_1 、S de_2 and S de are respectively the first seasonal sequence result, the second seasonal sequence result and the final seasonal sequence result obtained by the decoder decomposition in the FEDformer time series prediction algorithm; T de is the final trend sequence result obtained by the decoder; T de_1 、T de_2 and T de_3are respectively the first trend sequence result, the second trend sequence result and the third trend sequence result obtained by the decoder decomposition; X de is the input of the decoder; W s is the training parameter of the dust removal fan operating parameter prediction model;

[0010] Step 3: Divide all the data in the historical operation data set in order of arrangement, divide the first 70% of the data into a training set, and divide the last 30% of the data into a test set; input the data in the training set and the test set into the dust removal fan operation parameter prediction model for model training; after the training is completed, the input of the dust removal fan operation parameter prediction model is the negative pressure set value of the dust removal fan during the period when operation control is required, and the output is the operating parameter of the dust removal fan during the period when operation control is required;

[0011] Step 4: Establish the objective function of dust removal fan operation control, as shown in the following formula (2):

[0012]

[0013] In formula (2), α, β, and γ are different weight parameters in the objective function, which are empirical values; N is the duration of the period in minutes during which the dust removal fan needs to be controlled; M is the number of dust removal fans in the fan system; P j,i is the negative pressure of the j-th fan in the system at the i-th moment; I (P j,i ,X j,i ) is the P j,i Substitute the fan current value of the j-th fan in the system at the i-th moment obtained from the dust removal fan operation parameter prediction model after training; n (P j,i ,X j,i ) is the P j,i Substitute the fan speed value of the jth fan in the system at the i-th moment obtained from the dust removal fan operation parameter prediction model after training; ΔP set is the pressure difference setting value of the dust removal fan, which is an empirical value; ΔP (P j,i ,X j,i ) is the P j,i Substitute the fan pressure difference value of the j-th fan in the system at the i-th moment obtained from the dust removal fan operation parameter prediction model after training;

[0014] Step 5: Set the time length T for the dust removal fan to perform one operation control; select the distribution range of the negative pressure setting value in the historical operation data within 30 days from the start of the operation control as the P in the objective functionj,i The optimal negative pressure setting value in the operation control period is obtained by optimizing the objective function through the cuckoo search optimization algorithm; the optimal negative pressure setting value is compared with the current negative pressure setting value of the dust removal fan. If the difference exceeds the threshold, the optimal negative pressure setting value is sent to the dust removal fan for operation control; if the difference does not exceed the threshold, the dust removal fan operates normally according to the current negative pressure setting value, and no operation control is performed.

[0015] Furthermore, the model training steps of the dust removal fan operating parameter prediction model in step 3 are as follows:

[0016] Step 3.1: Set the input data types of the encoder and decoder; the input data types of the encoder are the dust removal fan operating parameters in the continuous operation data, all the equipment production capacity data, and all the discrete operation data; the input data types of the decoder are the negative pressure set value in the continuous operation data and all the equipment environment data;

[0017] Step 3.2: Use a sliding window method to perform sliding window division on the data in the training set and the test set; the sliding window length in the sliding window method is three times the time period required for the dust removal fan operating parameter prediction model to perform prediction; input the data belonging to the encoder input data type in the first two-thirds of the data obtained from the sliding window division into the dust removal fan operating parameter prediction model, and input the data belonging to the decoder input data type in the last one-third of the data into the dust removal fan operating parameter prediction model; at this time, the fan current, fan speed, and fan pressure difference of the dust removal fan within the time period required for prediction can be calculated;

[0018] Step 3.3: According to the data division method described in step 3.2, the training set data is input into the dust removal fan operating parameter prediction model for model training. The training adopts 20 epochs, that is, the complete training set is input into the model 20 times. The model parameters are updated by the back propagation algorithm. The optimizer selected is the adaptive moment estimation algorithm (Adam). The loss function used is:

[0019]

[0020] In the above formula, Represents the predicted value of each indicator, y _,i represents the actual value of each indicator, and α, β, and γ are the weight parameters of current loss, speed loss, and pressure difference loss, respectively.

[0021] Step 3.4: Use the test set divided in step 3.2 to evaluate the effect of the dust removal fan operating parameter prediction model. When the R-squared value of the model's prediction of fan current, fan speed, and fan pressure difference is greater than 0.8, it means that the ideal prediction effect has been achieved and the prediction model training can be terminated.

[0022] Furthermore, the specific steps of the cuckoo search optimization algorithm in step 5 for optimizing the objective function are as follows:

[0023] Step 5.1: The objective function of the cuckoo search optimization process is Equation (2) in Step 4. The optimization range is the distribution range of the negative pressure set value in the historical operation data within 30 days when the operation control starts. During initialization, the negative pressure value is obtained by Levy flight as the initial value, and then the objective function value f is obtained. i ;

[0024] Step 5.2: Randomly select a negative pressure value within the negative pressure optimization range and obtain the objective function value f j , compare f i and f j If the value of f i Less than f j , then f j The negative pressure value at f i The negative pressure value is replaced;

[0025] Step 5.3: Within the negative pressure optimization range, discard some bad negative pressure values, such as those with a large number of negative pressure values ​​or those with unreasonable values. The new values ​​of the discarded positions are generated by the Levy flight formula. The current negative pressure values ​​are arranged and the negative pressure value that minimizes the objective function is retained.

[0026] Step 5.4: Repeat steps 5.2 and 5.3, set the maximum number of loops to 50, and stop the optimization when the maximum number of loops is reached.

[0027] The beneficial effects of the present invention are as follows: 1. The present invention uses a time series prediction in the form of FEDformer to establish a dust removal fan operating parameter prediction model; then, the model's predicted value is used to design the operating control objective function, and a cuckoo search algorithm is used to find the recommended control value that optimizes the operating control function within the actual control parameter fluctuation range of the past 30 days; after obtaining the recommended control value, it is sent to the control system to complete the entire control process; the present invention ensures that the recommended control value can maintain the optimal control effect for a period of time, reducing the number of pushes. 2. The present invention improves the FEDformer-based time series prediction algorithm, adds a structure that can extract time domain information of fan operating data to the encoder part of the existing FEDformer-based time series prediction algorithm, and uses a multi-query attention mechanism structure to replace the multi-head self-attention structure in the existing algorithm, making the prediction results of the improved FEDformer-based time series prediction algorithm more accurate, thereby improving the accuracy of the operating control objective function designed using the model's predicted value when searching for the optimal value. DETAILED DESCRIPTION

[0028] The operation control method of a dust removal fan of the present invention is further described below in conjunction with a specific embodiment of a dual fan system.

[0029] A method for controlling the operation of a dust removal fan comprises the following steps:

[0030] Step 1: First, collect all historical operation data of the dust removal fan. The historical operation data is divided into four types: continuous operation data, discrete operation data, equipment capacity data, and equipment environment data. Continuous operation data includes the operating parameters and negative pressure set value of the dust removal fan during operation. The operating parameters include fan current, fan pressure difference, and fan speed. Discrete operation data includes the switch status of the iron mouth top suction electric valve, iron mouth side suction electric valve, and iron mouth slag skimmer electric valve during the operation of the dust removal fan. Equipment capacity data includes fan power consumption and blast furnace capacity. Equipment environment data includes date, holiday information, temperature, and weather. Then, align the four types of historical operation data using the difference method. Finally, use the 3σ method to eliminate all unreasonable data in the historical operation data, and collect the remaining historical operation data to form a historical operation data set.

[0031] Step 2: Establish a dust removal fan operating parameter prediction model based on the FEDformer time series prediction algorithm, as shown in the following formula (1):

[0032]

[0033] In formula (1), S en_fIt is the frequency domain sequence obtained by encoder decomposition in the FEDformer time series prediction algorithm; Decomp is the seasonality-trend decomposition algorithm; SCI (self-Channel Interaction) is the self-channel interaction network in the FEDformer time series prediction algorithm; MQA (Multi-Query Attention) is the multi-query attention mechanism algorithm; X en is the input data sequence of the input encoder; (Fourier Transform) is the Fourier transform algorithm; S en_t It is the time domain sequence decomposed by the encoder in the FEDformer time series prediction algorithm; S en is the output of the encoder; S de_1 、S de_2 and S de They are the first seasonal sequence results, the second seasonal sequence results, and the final seasonal sequence results obtained by decoder decomposition in the FEDformer time series prediction algorithm; T de is the final trend sequence result obtained by the decoder; T de_1 、T de_2 and T de_3 are the first trend sequence results, the second trend sequence results, and the third trend sequence results obtained by the decoder decomposition; X de is the input of the decoder; W s are the training parameters of the dust removal fan operation parameter prediction model;

[0034] Step 3: Divide all the data in the historical operation data set into the order of arrangement, with the first 70% of the data divided into the training set and the last 30% of the data divided into the test set; input the data in the training set and the test set into the dust removal fan operation parameter prediction model for model training; after the training, the input of the dust removal fan operation parameter prediction model is the negative pressure set value of the dust removal fan during the period requiring operation control, and the output is the operating parameters of the dust removal fan during the period requiring operation control;

[0035] The specific model training steps are as follows:

[0036] Step 3.1: Set the input data types of the encoder and decoder; the input data types of the encoder are the dust removal fan operating parameters in the continuous operation data, all the equipment production capacity data, and all the discrete operation data; the input data types of the decoder are the negative pressure set value in the continuous operation data and all the equipment environment data;

[0037] Step 3.2: Use a sliding window method to perform sliding window division on the data in the training set and the test set; the sliding window length in the sliding window method is three times the time period required for the dust removal fan operating parameter prediction model to perform prediction; input the data belonging to the encoder input data type in the first two-thirds of the data obtained from the sliding window division into the dust removal fan operating parameter prediction model, and input the data belonging to the decoder input data type in the last one-third of the data into the dust removal fan operating parameter prediction model; at this time, the fan current, fan speed, and fan pressure difference of the dust removal fan within the time period required for prediction can be calculated;

[0038] Step 3.3: According to the data division method described in step 3.2, the training set data is input into the dust removal fan operating parameter prediction model for model training. The training adopts 20 epochs, that is, the complete training set is input into the model 20 times. The model parameters are updated by the back propagation algorithm. The optimizer selected is the adaptive moment estimation algorithm (Adam). The loss function used is:

[0039]

[0040] In the above formula, Represents the predicted value of each indicator, y _,i represents the actual value of each indicator, and α, β, and γ are the weight parameters of current loss, speed loss, and pressure difference loss, respectively.

[0041] Step 3.4: Use the test set divided in step 3.2 to evaluate the effect of the dust removal fan operating parameter prediction model. When the R-squared value of the model's prediction of fan current, fan speed, and fan pressure difference is greater than 0.8, it means that the ideal prediction effect has been achieved and the prediction model training can be terminated.

[0042] Step 4: Establish the objective function of dust removal fan operation control, as shown in the following formula (2):

[0043]

[0044] In formula (2), α, β, and γ are different weight parameters in the objective function, which are empirical values; N is the duration of the period in minutes during which the dust removal fan needs to be controlled; P 1,i and P 2,i They are respectively the negative pressure of main pipe 1 and main pipe 2 of the dust removal fan; I (P 1,i ,X 1,i ) and O I (P 2,j ,X2,j ) are P 1,i and P 2,j Substitute the current values ​​of the two fans at the i-th moment obtained from the dust removal fan operation parameter prediction model after training; n (P 1,i ,X 1,i ) and O n (P 2,i ,X 2,i ) are P 1,i and P 2,i Substitute the fan speed values ​​of the two fans at the i-th moment obtained from the dust removal fan operation parameter prediction model after training; ΔP set is the pressure difference setting value of the dust removal fan, which is an empirical value; ΔP (P 1,i ,X 1,i ) and O ΔP (P 2,i ,X 2,i ) are P 1,i and P 2,i Substitute the fan pressure difference value of the two fans at the i-th moment obtained from the dust removal fan operation parameter prediction model after training;

[0045] Step 5: Set the time length T for the dust removal fan to perform one operation control; select the distribution range of the negative pressure set value in the historical operation data within the last 30 days from the start of the operation control as the P in the objective function. 1,i and P 2,i The optimal negative pressure setting value in the operation control period is obtained by optimizing the objective function through the cuckoo search optimization algorithm; the optimal negative pressure setting value is compared with the current negative pressure setting value of the dust removal fan. If the difference exceeds a certain threshold, the optimal negative pressure setting value is sent to the dust removal fan for operation control; if the difference does not exceed a certain threshold, the dust removal fan operates normally according to the current negative pressure setting value, and no operation control is performed.

[0046] The specific steps of the cuckoo search optimization algorithm to optimize the objective function are as follows:

[0047] Step 5.1: The objective function of the cuckoo search optimization process is Equation (2) in Step 4. The optimization range is the distribution range of the negative pressure set value in the historical operation data within 30 days when the operation control starts. During initialization, the negative pressure value is obtained by Levy flight as the initial value, and then the objective function value fi is obtained. i ;

[0048] Step 5.2: Randomly select a negative pressure value within the negative pressure optimization range and obtain the objective function value f j , compare f iand f j If the value of f i Less than f j , then f j The negative pressure value at f i The negative pressure value is replaced;

[0049] Step 5.3: Within the negative pressure optimization range, discard some bad negative pressure values, such as those with a large number of negative pressure values ​​or those with unreasonable values. The new values ​​of the discarded positions are generated by the Levy flight formula. The current negative pressure values ​​are arranged and the negative pressure value that minimizes the objective function is retained.

[0050] Step 5.4: Repeat steps 5.2 and 5.3, set the maximum number of loops to 50, and stop the optimization when the maximum number of loops is reached.

[0051] The following is the operation control of a specific dust removal fan

[0052] The data of the dust removal fan within the current hour is shown in the following table:

[0053]

[0054]

[0055]

[0056] After aligning the data and removing unreasonable data, the input sample was obtained. The mean of the actual negative pressure value within 30 days was calculated to be -1.95 kPa and the variance was 0.125 kPa. According to the 1σ criterion distribution, the optimal range of the set negative pressure value was [-2.079, -1.829].

[0057] Based on the trained dust removal fan operating parameter prediction model and input samples, the cuckoo optimization algorithm is used to obtain the optimal set negative pressure recommendation value of -1.81kPa that minimizes the objective function of the dust removal fan operation control within the set negative pressure optimization range. The current set negative pressure value is -1.93kPa. The difference between the two exceeds the threshold of 0.1kPa, and down-control processing can be performed.

[0058] Based on the dust removal fan operation data in the past month, the comparison between the existing control method and the method of the present invention is as follows:

[0059]

[0060] By comparing the data, it can be seen that the energy-saving control method of the present invention is significantly stronger than the existing control method. Compared with the existing control method, the method of the present invention pays more attention to the influence of equipment working conditions, equipment operating environment, and blast furnace output, pays attention to long-term energy-saving benefits, avoids excessively frequent manipulation of the control device, and reduces the risk of failure of the control equipment.

Claims

1. A method for controlling the operation of a dust removal fan, characterized in that: The following steps are involved: Step 1: First, collect all historical operation data of the dust removal fan, which are divided into four types: continuous operation data, discrete operation data, equipment production capacity data, and equipment environment data; The continuous operation data includes the operating parameters and negative pressure setting value of the dust removal fan during operation, and the operating parameters include fan current, fan pressure difference and fan speed; the discrete operation data includes the switch status of the iron mouth top suction electric valve, iron mouth side suction electric valve and iron mouth slag skimmer electric valve when the dust removal fan is running; the equipment production capacity data includes fan power consumption and blast furnace production capacity; the equipment environment data includes date, holiday information, temperature and weather; then, the four types of historical operation data are aligned by the difference method; finally, the 3σ method is used to eliminate all unreasonable data in the historical operation data, and the remaining historical operation data are collected to form a historical operation data set; Step 2: Establish a dust removal fan operating parameter prediction model based on the FEDformer time series prediction algorithm, as shown in the following formula (1): In formula (1), S en_f is the frequency domain sequence obtained by encoder decomposition in the FEDformer time series prediction algorithm; Decomp is the seasonality-trend decomposition algorithm; SCI (self-Channel Interaction) is the self-channel interaction network in the FEDformer time series prediction algorithm; the MQA (Multi-Query Attention) is the multi-query attention mechanism algorithm; the X en is the input data sequence of the encoder; F (Fourier Transform) is the Fourier transform algorithm; S en_t is the time domain sequence decomposed by the encoder in the FEDformer time series prediction algorithm; S en is the output of the encoder; S de_1 、S de_2 and S de They are respectively the first seasonal sequence result, the second seasonal sequence result and the final seasonal sequence result obtained by decoder decomposition in the FEDformer time series prediction algorithm; T de is the final trend sequence result obtained by the decoder; T de_1 、T de_2 and T de_3 are respectively the first trend sequence result, the second trend sequence result and the third trend sequence result obtained by the decoder decomposition; X de is the input of the decoder; W s is the training parameter of the dust removal fan operating parameter prediction model; Step 3: Divide all the data in the historical operation data set in order of arrangement, divide the first 70% of the data into a training set, and divide the last 30% of the data into a test set; input the data in the training set and the test set into the dust removal fan operation parameter prediction model for model training; after the training is completed, the input of the dust removal fan operation parameter prediction model is the negative pressure set value of the dust removal fan during the period when operation control is required, and the output is the operating parameter of the dust removal fan during the period when operation control is required; Step 4: Establish the objective function of dust removal fan operation control, as shown in the following formula (2): In formula (2), α, β, and γ are different weight parameters in the objective function, which are empirical values; N is the duration of the period in minutes during which the dust removal fan needs to be controlled; M is the number of dust removal fans in the fan system; P j,i is the negative pressure of the j-th fan in the system at the i-th moment; I (P j,i ,X j,i ) is the P j,i Substitute the fan current value of the j-th fan in the system at the i-th moment obtained from the dust removal fan operation parameter prediction model after training; n (P j,i ,X j,i ) is the P j,i Substitute the fan speed value of the jth fan in the system at the i-th moment obtained from the dust removal fan operation parameter prediction model after training; ΔP set is the pressure difference setting value of the dust removal fan, which is an empirical value; ΔP (P j,i ,X j,i ) is the P j,i Substitute the fan pressure difference value of the j-th fan in the system at the i-th moment obtained from the dust removal fan operation parameter prediction model after training; Step 5: Set the time length T for the dust removal fan to perform one operation control; select the distribution range of the negative pressure setting value in the historical operation data within 30 days from the start of the operation control as the P in the objective function j,i The optimal negative pressure setting value in the operation control period is obtained by optimizing the objective function through the cuckoo search optimization algorithm; the optimal negative pressure setting value is compared with the current negative pressure setting value of the dust removal fan. If the difference exceeds the threshold, the optimal negative pressure setting value is sent to the dust removal fan for operation control; if the difference does not exceed the threshold, the dust removal fan operates normally according to the current negative pressure setting value, and no operation control is performed.

2. The dust removal fan operation control method according to claim 1, characterized in that: The model training steps of the dust removal fan operating parameter prediction model in step 3 are as follows: Step 3.1: Set the input data types of the encoder and decoder; the input data types of the encoder are the dust removal fan operating parameters in the continuous operation data, all the equipment production capacity data, and all the discrete operation data; the input data types of the decoder are the negative pressure set value in the continuous operation data and all the equipment environment data; Step 3.2: Use a sliding window method to perform sliding window division on the data in the training set and the test set; the sliding window length in the sliding window method is three times the time period required for the dust removal fan operating parameter prediction model to perform prediction; input the data belonging to the encoder input data type in the first two-thirds of the data obtained from the sliding window division into the dust removal fan operating parameter prediction model, and input the data belonging to the decoder input data type in the last one-third of the data into the dust removal fan operating parameter prediction model; at this time, the fan current, fan speed, and fan pressure difference of the dust removal fan within the time period required for prediction can be calculated; Step 3.3: According to the data division method described in step 3.2, the training set data is input into the dust removal fan operating parameter prediction model for model training. The training adopts 20 epochs, that is, the complete training set is input into the model 20 times. The model parameters are updated by the back propagation algorithm. The optimizer selected is the adaptive moment estimation algorithm (Adam). The loss function used is: In the above formula, Represents the predicted value of each indicator, y _,i represents the actual value of each indicator, and α, β, and γ are the weight parameters of current loss, speed loss, and pressure difference loss, respectively. Step 3.4: Use the test set divided in step 3.2 to evaluate the effect of the dust removal fan operating parameter prediction model. When the R-squared value of the model's prediction of fan current, fan speed, and fan pressure difference is greater than 0.8, the training of the prediction model is terminated.

3. The dust removal fan operation control method according to claim 1, characterized in that: The specific steps of the cuckoo search optimization algorithm in step 5 to perform optimization calculation on the objective function are as follows: Step 5.1: The objective function of the cuckoo search optimization process is Equation (2) in Step 4. The optimization range is the distribution range of the negative pressure set value in the historical operation data within 30 days when the operation control starts. During initialization, the negative pressure value is obtained by Levy flight as the initial value, and then the objective function value f is obtained. i ; Step 5.2: Randomly select a negative pressure value within the negative pressure optimization range and obtain the objective function value f j , compare f i and f j If the value of f i Less than f j , then f j The negative pressure value at f i The negative pressure value is replaced; Step 5.3: Within the negative pressure optimization range, discard some bad negative pressure values, such as those with a large number of negative pressure values ​​or those with unreasonable values. The new values ​​of the discarded positions are generated by the Levy flight formula. The current negative pressure values ​​are arranged and the negative pressure value that minimizes the objective function is retained. Step 5.4: Repeat steps 5.2 and 5.3, set the maximum number of loops to 50, and stop the optimization when the maximum number of loops is reached.

Images