Direct air cooling unit control method and system based on MOPSO, electronic equipment and storage medium

Through the MOPSO-based control method, a non-parametric dynamic response model for the fan group inlet flow of the air-cooled unit is established, and the adjustment instructions are optimized, which solves the adjustment deviation and partition control accuracy problems in the operation of the air-cooled unit, and improves its economic and stability.

CN119022679APending Publication Date: 2024-11-26CPI GUANGXI NUCLEAR POWER CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411130477.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The driving cost and performance optimization methods of the prior art hollow cooling units have problems such as large adjustment deviations, serious oscillations and poor accuracy of partition control strategies, which leads to improved operational economics and stability.

Method used

Using the MOPSO-based control method, by obtaining the historical operation data of the air-cooled unit, constructing partition samples and modeling samples, establishing a non-parametric dynamic response model for the inlet flow of the fan group, obtaining multiple optimization solutions, determining the target optimization and adjustment instructions, and realizing the region Control and speed optimization adjustment.

Benefits of technology

It improves the stability and economy of the air-cooled unit, reduces adjustment costs, reduces program jumps and adjustment deviations, and enhances the accuracy of partition control of the fan group.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119022679A_ABST
    Figure CN119022679A_ABST
Patent Text Reader

Abstract

The invention provides an MOPSO-based direct air cooling unit control method and system, electronic equipment and a storage medium, and the method comprises the steps: obtaining the operation data of an air cooling unit, and forming a partition sample and a modeling sample; performing data preprocessing on the partition samples and the modeling samples; partitioning the partitioned sample, and determining a target partitioning scheme through the contour coefficient; identifying the modeling sample, and establishing an air cooling unit fan group inlet flow nonparametric dynamic response model; and the fan group inlet flow and the fan rotating speed are obtained according to the target partition scheme, multiple target optimization solutions are obtained based on the fan group inlet flow, the fan rotating speed and an air cooling unit fan group inlet flow nonparametric dynamic response model, and the multiple target optimization solutions are calculated and optimized to determine a target optimization adjustment instruction. In this way, regionalized accurate control over the air cooling unit is achieved, the fan rotating speed adjusting process is optimized, and therefore the economical efficiency and stability of operation of the air cooling unit are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of state monitoring and operation optimization of direct air-cooling units, and in particular, relates to a control method, system, electronic equipment and storage medium for direct air-cooling units based on MOPSO. Background Art

[0002] As the situation of industrial water use in water-scarce areas becomes increasingly severe, direct air-cooling units have been developed on a large scale and rapidly popularized. For example, air-cooling units use high-power low-speed axial flow fans on air-cooling arrays to drive forced air flow, and explore optimized energy-saving technologies based on the cost and performance of the cold-end drive of the unit, which has important economic significance and social value in promoting the overall clean and efficient transformation of coal-fired power units.

[0003] The method for optimizing the driving cost and performance of the air-cooling unit in the prior art is to adjust the deviation between the measured value and the set value of the inlet flow of the fan group, and adopt the control method of PID plus deviation dead zone, so that the turbine back pressure value is always maintained within the allowable range of the set value, and then continuously adjust the fan operating speed according to its deviation.

[0004] However, this method has obvious disadvantages. On the one hand, when the deviation is large, the program steps are executed. However, during the debugging, it is found that the flow rate has large overshoot and fluctuations during the adjustment process, which easily leads to program skipping, resulting in large adjustment deviation and oscillation, thereby increasing the adjustment cost. On the other hand, due to the complex nonlinear relationship between the fan speed and the condenser inlet flow, it is easy to make the partition control strategy of a certain row or column as an area lack accuracy, thereby reducing stability. In summary, the optimization method in the prior art makes the economy and stability of the air-cooling unit operation need to be improved. Summary of the invention

[0005] To solve the above problems, the present disclosure provides a MOPSO direct air-cooling unit control method, system, electronic device and storage medium, which not only provides an accurate zoning control scheme and improves stability, but also has speed optimization adjustment instructions to reduce the adjustment cost and realize economical and stable operation of the air-cooling unit.

[0006] The first aspect of the present disclosure provides a control method for a direct air-cooling unit based on MOPSO, comprising: obtaining operating data of the air-cooling unit in a historical normal working state, and respectively forming partition samples and modeling samples; performing data preprocessing on the partition samples and the modeling samples; partitioning the partition samples for data preprocessing, and determining a target partitioning scheme through a contour coefficient; identifying the modeling samples for data preprocessing, and establishing a non-parametric dynamic response model for the inlet flow of a fan group of an air-cooling unit; obtaining the inlet flow of a fan group and a fan speed according to the target partitioning scheme, obtaining a plurality of target optimization solutions based on the inlet flow of the fan group, the fan speed, and the non-parametric dynamic response model for the inlet flow of the fan group of the air-cooling unit, and calculating and optimizing the plurality of target optimization solutions to determine a target optimization adjustment instruction.

[0007] With such a setting, by establishing partition samples and modeling samples, it is convenient to form a partition scheme for the fan group of the air-cooling unit, and establish a non-parametric dynamic response model of the inlet flow of the fan group; on the one hand, it is convenient to realize regional control of the fan group on the basis of the nonlinear relationship between the fan speed and the inlet flow, so as to improve the accuracy of the partition control strategy in which certain rows or columns of the fan group are one area, thereby improving the stability; on the other hand, according to the inlet flow of the fan group, the fan speed and the non-parametric dynamic response model of the inlet flow of the fan group, multiple target optimization solutions can be obtained, so as to determine the target optimization adjustment instructions from multiple target optimization solutions, and adjust the inlet flow of the fan group according to the target optimization adjustment instructions, reduce the situation of program jumping, reduce the adjustment deviation, and thus reduce the fan speed adjustment cost, so that the above two aspects can be combined to improve the economy and stability of the operation of the air-cooling unit.

[0008] In some embodiments, the operating data of the air-cooling unit under historical normal working conditions is obtained to form partition samples and modeling samples, respectively, including: the partition samples include the fan inlet flow data of the air-cooling unit under constant speed and natural wind disturbance; the modeling samples include the continuous fan speeds, natural wind speeds and fan inlet flow data of the air-cooling unit under dual random disturbances of natural wind and speed instructions.

[0009] This arrangement makes it possible to obtain the inlet flow data of each fan of the air-cooling unit under constant speed and natural wind disturbance, and at the same time obtain the speed of each fan, natural wind speed and fan inlet flow data of the air-cooling unit under the dual random disturbance of natural wind and speed command, thereby facilitating the acquisition of the fan inlet flow data of the air-cooling unit under constant speed and speed command respectively.

[0010] In some embodiments, the operating data covers the dynamic changes of all parameters of the air-cooling unit under all operating conditions, and the interval time of the operating data set is at most 5 seconds.

[0011] This arrangement shortens the time interval for obtaining the running data set, reduces duplication between running data, and improves the efficiency of obtaining the running data.

[0012] In some embodiments, the data preprocessing of the partition samples and the modeling samples includes: normalizing the running data of the partition samples and the modeling samples.

[0013] This arrangement allows the operating data to be normalized into a range interval, thereby facilitating subsequent calculations based on operating data with different range intervals.

[0014] In some embodiments, partitioning the partition samples of the data preprocessing and determining the target partition scheme by the silhouette coefficient comprises: arranging the partition samples into a K-means input matrix X K-means ; Based on the input matrix X K-means The number of partition samples is set to 2, 3, ..., 6 respectively, and the wind turbine group is classified by K-means algorithm, and each partition scheme is recorded; the silhouette coefficients of the partition schemes are calculated and compared, and the partition scheme with the largest silhouette coefficient is selected as the target partition scheme.

[0015] With this setting, the input matrix X is formed by partitioning the normalized partition samples K-means, So that in the matrix X K-means The number of partition samples is calculated by K-means algorithm to classify the wind turbine group and form various partition schemes, so as to compare the partition scheme with the largest silhouette coefficient as the target partition scheme, that is, the optimal partition scheme.

[0016] In some embodiments, the modeling samples of the data preprocessing are identified, and a non-parametric dynamic response model of the inlet flow of the fan group of the air-cooling unit is established, including: constructing a non-parametric dynamic response model structure of the inlet flow of the fan group of the air-cooling unit; initializing the training parameters of the non-parametric dynamic response model; and training and hyperparameter debugging the non-parametric dynamic response model.

[0017] With this setting, the training parameters of the established non-parametric dynamic response model will be initialized first to optimize the algorithm, model evaluation indicators, and initial learning rate; the non-parametric dynamic response model training and hyperparameter debugging will be performed to make the non-parametric dynamic response model training more perfect by increasing training batches and training rounds.

[0018] In some embodiments, the inlet flow rate and fan speed of the fan group are obtained according to the target partitioning scheme, multiple target optimization solutions are obtained based on the inlet flow rate of the fan group, the fan speed and the non-parametric dynamic response model of the inlet flow rate of the air cooling unit fan group, and the multiple target optimization solutions are calculated and optimized to determine the target optimization adjustment instruction, including: initializing parameters of the target partitioning scheme; obtaining the inlet flow rate and fan speed of the fan group according to the initialized target partitioning scheme, and calculating multiple target optimization solutions based on the inlet air flow rate and fan speed of the fan group. Take 5s as a speed adjustment cycle, calculate the target speed in the multiple target optimization solutions, and compare the target speed with the speed under the last speed instruction. If the speed adjustment amount between the target speed and the speed under the last speed instruction is greater than the preset threshold, the speed instruction is the target optimization adjustment instruction, otherwise the last speed instruction is used.

[0019] This arrangement makes it easier to compare the calculated multiple target speeds with the speed under the last speed instruction. During the comparison process, the target optimization adjustment instruction is determined by the speed adjustment amount between the two speeds being greater than the preset threshold, thereby facilitating the speed optimization adjustment of the air-cooling unit fan through the target optimization adjustment instruction.

[0020] In some embodiments, the initialization parameters of the target partitioning scheme include: determining the decision variable dimension, that is, the number of control speeds, according to the target partitioning scheme; and determining the speed range of the optimization problem according to the historical speed data range of the modeling samples.

[0021] This arrangement makes it easy to obtain the number and range of control speeds of the target partition scheme, which helps to select the speed range in the target partition scheme to calculate the target speed.

[0022] In some embodiments, the fan group inlet flow and fan speed are obtained according to the initialized target partitioning scheme, and multiple target optimization solutions are calculated based on the fan group inlet air flow and fan speed, including: calculating the fan energy consumption and fan regulation mechanical electrical loss based on the fan group inlet air flow and fan speed; calculating the input speed based on the fan group inlet air flow, fan energy consumption, and fan regulation mechanical electrical loss.

[0023] With such settings, the air flow rate at the fan group inlet, the fan energy consumption, and the fan regulation mechanical and electrical losses can be used as optimization targets. The fan group inlet is taken as the maximum value optimization problem and the inverse calculation is used, so that the input speeds of the fan group inlet air flow rate, fan energy consumption, and fan regulation mechanical and electrical losses can be calculated separately.

[0024] In some embodiments, 5s is used as a speed adjustment cycle, the target speeds in the multiple target optimization solutions are calculated, and the target speed is compared with the speed under the previous speed instruction. If the speed adjustment amount between the target speed and the speed under the previous speed instruction is greater than a preset threshold, the speed instruction is the target optimization adjustment instruction, otherwise the previous speed instruction is used, including: calculating the fuzzy membership function of the input speeds of the fan group inlet air flow, fan energy consumption, and fan adjustment mechanical and electrical loss, and weighted summing them to obtain the target speed, and taking the decision variable of the target speed corresponding to the minimum result as the target optimization adjustment instruction.

[0025] Such setting makes it easy to use the decision variable corresponding to the obtained target speed as the adjustment instruction of the output speed, that is, the target optimization adjustment instruction, so as to facilitate the adjustment of the inlet flow of the fan group, reduce the program jumping situation, reduce the adjustment deviation, and thus reduce the fan speed adjustment cost.

[0026] The second aspect of the present disclosure provides a control system for a direct air-cooling unit based on MOPSO, the system comprising: a sample unit, for acquiring operating data of the air-cooling unit in a historical normal working state, and forming partition samples and modeling samples respectively; a preprocessing unit, for performing data preprocessing on the partition samples and the modeling samples; a partitioning unit, for partitioning the partition samples for data preprocessing, and determining a target partitioning scheme through a contour coefficient; a model building unit, for identifying the modeling samples for data preprocessing, and establishing a non-parametric dynamic response model for the inlet flow of a fan group of the air-cooling unit; a calculation unit, for acquiring the inlet flow and fan speed of the fan group according to the target partitioning scheme, acquiring multiple target optimization solutions based on the inlet flow, fan speed and the non-parametric dynamic response model for the inlet flow of the fan group of the air-cooling unit, and calculating and optimizing the multiple target optimization solutions to determine target optimization adjustment instructions.

[0027] Such a setting makes it easy to determine the target partitioning scheme through partitioning samples, establish a non-parametric dynamic response model of the fan group inlet flow through modeling samples, and then obtain the fan group inlet flow and fan speed based on the target partitioning scheme, and then obtain multiple target optimization solutions in the non-parametric dynamic response model of the fan group inlet flow, so as to find the best among multiple target optimization solutions, thereby facilitating the optimization of each control cycle to solve the target optimization adjustment instructions, so as to achieve economical and safe operation of the direct air cooling unit.

[0028] A third aspect of the present disclosure provides an electronic device, comprising: a memory for storing a computer program; and a processor for implementing the method described in the first aspect when executing the computer program.

[0029] The technical effects brought about by any possible implementation of the third aspect can refer to the technical effects brought about by the above-mentioned first aspect, and will not be repeated here.

[0030] A fourth aspect of the present disclosure provides a computer storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in the first aspect is implemented.

[0031] The technical effects brought about by any possible implementation of the fourth aspect can refer to the technical effects brought about by the above-mentioned first aspect, and will not be repeated here.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] 1. Considering the strong coupling characteristics between fans and the complex influence of ambient wind on the air cooling island flow field, the fan speed and the condenser inlet flow rate present a complex nonlinear relationship. The accuracy of the zoning control strategy with certain rows or columns as a region is improved. By analyzing the inlet flow data of each fan and extracting data sample information, the regionalized, detailed and accurate control of the fan is achieved.

[0034] 2. This paper proposes a fan group inlet flow prediction model, which can accurately describe the nonlinear and dynamic characteristics of the controlled object through air-cooling island flow field data modeling and verification, and provides a reference for establishing a thermal control system object model.

[0035] 3. The present disclosure takes into account the back pressure of the unit, the cooling ventilation volume or the energy consumption of the motor, reduces the frequent adjustment of the control system, that is, the equipment loss of the speed gear and the frequency conversion electrical device caused by the frequent action of the controller, takes into account the output cost of the control system, and improves the life of the system and the stability of operation.

[0036] 4. The MOPSO algorithm is used in combination with the weighted fuzzy membership function method to select the optimal speed control command. The inlet flow of the air fan group, the energy consumption of the axial flow fan, and the mechanical and electrical losses are comprehensively considered. Combined with the zoning control of the axial flow fan, the fan energy consumption can be reduced and the response speed can be improved. Under random continuous environmental wind disturbances, the axial flow fan performs well in working stability, providing a reference with engineering application value for the energy-saving operation of direct air-cooled units.

[0037] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or be understood by implementing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 It is a flow chart of a control method of a direct air cooling unit based on MOPSO provided by an embodiment of the present disclosure;

[0040] Figure 2 is a flow chart of the method of S300 provided in an embodiment of the present disclosure;

[0041] Figure 3 is a flow chart of a method for determining a target partitioning scheme using K-Means provided in an embodiment of the present disclosure;

[0042] Figure 4 It is a K-Means clustering partitioning scheme diagram provided by an embodiment of the present disclosure;

[0043] Figure 5 is a comparison diagram of silhouette coefficient values ​​of various partitioning schemes provided in an embodiment of the present disclosure;

[0044] Figure 6 is a flow chart of the method of S400 provided in an embodiment of the present disclosure;

[0045] Figure 7 It is a flow chart of establishing the LSTM non-parametric response model provided by the embodiment of the present disclosure;

[0046] Figure 8 It is a network structure diagram of the LSTM parameter-based dynamic response model provided by the embodiment of the present disclosure;

[0047] Fig. 9 is a flow chart of the method of S500 provided in an embodiment of the present disclosure;

[0048] Fig.10 is a flow chart of the MOPSO optimization control strategy provided by an embodiment of the present disclosure;

[0049] Fig.11 It is a comparison diagram of the predicted value and the actual value of the LSTM model provided by the embodiment of the present disclosure;

[0050] Fig.12 is a flow curve diagram of the fan group inlet provided by the embodiment of the present disclosure;

[0051] Fig.13 is a curve diagram of fan speeds in each zone provided by an embodiment of the present disclosure;

[0052] Fig.14It is a block diagram of a control system of a direct air cooling unit based on MOPSO provided in an embodiment of the present disclosure;

[0053] Fig.15 It is a structural block diagram of an electronic device provided by the present disclosure. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0055] The following is a brief introduction to some concepts that may be involved in the embodiments of the present disclosure.

[0056] MOPSO (Multi-Objective Partic le Swarm Optimization) is a parallel global optimization algorithm that handles the problem of solving multiple objective functions simultaneously. It is suitable for complex optimization scenarios where it is difficult to find a single global optimal solution.

[0057] K-means clustering algorithm: An unsupervised machine learning algorithm used to divide a dataset into a preset number of categories (K clusters).

[0058] LSTM (Long Short-Term Memory) neural network: is a special recurrent neural network (Recurrent Neural Network), long short-term memory, which introduces three key gating units: input gate (InputGate), forget gate (Forget Gate) and output gate (Output Gate).

[0059] At present, air-cooled units use high-power low-speed axial flow fans on air-cooled arrays to drive forced air flow, cool the exhaust steam in the finned tubes to establish the cold end of the generator set, and the total power of the fan group motors reaches 5000-10000MW, accounting for about 1.3% to 1.6% of the power generation of thermal power units. In order to optimize the driving cost and performance of air-cooled units, the existing technology usually adopts the deviation adjustment between the measured value and the set value of the fan group inlet flow, and adopts the control method of PID plus deviation dead zone, so that the turbine back pressure value is always maintained within the range allowed by the set value, and then the fan operating speed is continuously adjusted according to its deviation.

[0060] However, the above-mentioned adjustment method will have large deviations and will easily cause oscillations during adjustment, which will increase the adjustment cost. In addition, since the fan speed and the condenser inlet flow rate present a complex nonlinear relationship, it is easy for the zoning control strategy that uses certain rows or columns as one area to lack accuracy, resulting in the need to improve the economy and stability of the air-cooling unit operation.

[0061] To this end, the present disclosure provides a control method for a direct air cooling unit based on MOPSO, such as Figure 1 As shown, Figure 1 The present invention provides a flow chart of a method for controlling a direct air cooling unit based on MOPSO, the method comprising:

[0062] S100, obtaining operation data of the air cooling unit in a historical normal working state, and forming partition samples and modeling samples respectively.

[0063] In some embodiments, operating data is obtained through an air-cooling island in a thermal power plant, wherein partitioned samples are the inlet flow data of each fan under a constant speed of the overall fan of the air-cooling unit and a natural wind disturbance, and modeling samples are the continuous speed data of each fan, natural wind speed and fan inlet flow data of the air-cooling unit under dual random disturbances of natural wind and speed command. The partitioned samples and modeling samples of the operating data cover the dynamic changes of all parameters of the air-cooling unit under all operating conditions, and the data set interval is at most 5s.

[0064] For example, taking the 1×8 air-cooling unit of a 660MW unit as the object, computational fluid dynamics numerical simulation was performed to obtain: the expected speed of each fan is 40rpm, and the variance is 21.33; the expected natural wind speed is 3m / s, and the variance is 0.25, the speed of each fan, the natural wind speed, and the inlet flow of the fan group are used as modeling samples; the overall speed of the fan is 80rpm, and the inlet flow of each fan under the condition of a natural wind speed of 0-4-8m / s is used as a partition sample.

[0065] S200: performing data preprocessing on partition samples and modeling samples.

[0066] In this embodiment, the running data, that is, the partition samples and the modeling samples, are normalized, and the formula is as follows:

[0067] X std =(XX min ) / (X max -X min ) (1)

[0068] X scaled =X std *(max-min)+min (2)

[0069] In formula (1) and (2), the data sequence of a certain measurement point parameter in the operating data is recorded as Xi , then the column data is recorded as X i =[X i (t1),X i (t2)......X i (t n )], where X is an element in the column, X min is the minimum value of the column data, X max is the maximum value of the column data, X std is the intermediate variable of the normalization process, (max, min) is the interval of the normalization range set by the normalization process, X scaled The data are normalized.

[0070] S300, partitioning the partition samples of the data preprocessing, and determining the target partitioning scheme through the silhouette coefficient.

[0071] In this embodiment, the partition samples are analyzed and partitioned by using the K-means clustering algorithm, and then the target partitioning scheme is determined by the silhouette coefficient.

[0072] S400, identifying the modeling samples of the data preprocessing, and establishing a non-parametric dynamic response model of the inlet flow of the fan group of the air-cooling unit.

[0073] In this embodiment, an LSTM neural network is used to identify modeling samples, and then a non-parametric dynamic response model of the fan group inlet flow is established.

[0074] S500, obtaining the fan group inlet flow and fan speed according to the target partitioning scheme, obtaining multiple target optimization solutions based on the fan group inlet flow, fan speed and the non-parametric dynamic response model of the air-cooling unit fan group inlet flow, calculating and optimizing the multiple target optimization solutions to determine the target optimization adjustment instructions.

[0075] In this embodiment, according to the target zoning scheme, the performance indicators such as the fan inlet air flow, motor energy consumption and mechanical and electrical losses are comprehensively considered to obtain the fan group inlet flow and fan speed, and implement regional control of the fan group to improve the accuracy of the zoning control strategy in which certain rows or columns of the fan group are one area; then, based on the fan group inlet flow, fan speed and the non-parametric dynamic response model of the air-cooled unit fan group inlet flow, multiple target optimization solutions are obtained, and a method for optimizing multiple target optimization solutions based on the MOPSO algorithm is proposed. Finally, the target optimization adjustment instruction of the best target optimization solution is determined by the fuzzy membership function method, so that the fan group inlet flow can be adjusted through the target optimization adjustment instruction, the program jump situation can be reduced, the adjustment deviation can be reduced, and the fan speed adjustment cost can be reduced, which is helpful to improve the economy and stability of the air-cooled unit operation.

[0076] In some embodiments, Figure 2 and Figure 3 As shown, Figure 2 is a flow chart of the method S300 provided in an embodiment of the present disclosure, Figure 3 is a flow chart of a method for determining a target partitioning scheme by K-Means provided in an embodiment of the present disclosure, and the specific steps of S300 include:

[0077] S301, organize the partition samples into K-means input matrix X K-means .

[0078] In this embodiment, the partition sample length is n, the number of wind turbines is m, and it is recorded as X i =[X i (t1),X i (t2)......X i (t n )], i∈1,…,m, and the preprocessed partition samples are used to form an m*n matrix XK-means.

[0079] Step S302: Set the number of partitions k to 2, 3, ..., 6 respectively, classify the wind turbine group by K-means algorithm, and record each partition scheme.

[0080] In this embodiment, if Figure 4 As shown, Figure 4 is a K-Means clustering partitioning scheme diagram provided by an embodiment of the present disclosure, so the Euclidean distance is selected in the K-Means algorithm to represent the distance between the classified sample and the cluster center, as shown in formula (3):

[0081]

[0082] In the formula, c is the cluster center, j∈1:k; k∈2,…,6, j refers to the traversal from 1 to k, and k is the number of different cluster centers, so as to facilitate the classification of wind turbine groups by calculating formula (3) and record each partition scheme.

[0083] Step S303: Calculate and compare the silhouette coefficients of the partitioning schemes, and select the partitioning scheme with the largest silhouette coefficient as the target partitioning scheme.

[0084] In this embodiment, if Figure 5 As shown, Figure 5 is a comparison diagram of the silhouette coefficient values ​​of various partitioning schemes provided in the embodiments of the present disclosure. The silhouette coefficient can be calculated by formula (4), (5), (6):

[0085]

[0086]

[0087] Where s is the silhouette coefficient of a single sample, a(i) is the average distance between i and other samples in the class to which it belongs, b(i) is the minimum value of the average distance between i and samples of other classes. The calculation method of b(i) is similar to that of formula (5). Replace a(i) with b(i) in formula (5). It is necessary to traverse other classes to obtain multiple values ​​{b1(i), ..., bk(i)}. Find k minimum values ​​from formula (6), which represent the k partitioning schemes, and select the minimum value as the final result.

[0088] Finally, the SC values ​​of different partitioning schemes are calculated. SC is the mean of the silhouette coefficients of each sample as the silhouette coefficient of the cluster as a whole. The maximum SC value is taken as the target partitioning scheme, which is also the optimal partitioning scheme.

[0089] In some embodiments, Figure 6 and Figure 7 As shown, Figure 6 is a flow chart of the method S400 provided in an embodiment of the present disclosure, Figure 7 4 is a flowchart of establishing an LSTM non-parametric response model provided by an embodiment of the present disclosure, and the specific steps of S400 include:

[0090] Step S401: construct a non-parametric dynamic response model structure of the inlet flow of the air cooling unit fan group.

[0091] In this embodiment, an LSTM non-parametric dynamic response model structure of the inlet flow of the fan group of the air cooling unit is constructed, including the design of the LSTM network structure and the selection of the activation function.

[0092] For example, let the modeling sample length be n, the number of fans be m, and the fan speed be U i =[U i (t1),U i (t2),…,U i (t n )], i∈1,…,m; the ambient wind speed is V=[V(t1),V(t2),…,V(t n )], the two constitute the LSTM network (m+1)*n dimensional input matrix XLSTM; the total flow rate at the fan inlet constitutes the output matrix YLSTM=[Y(t1),Y(t2),…,Y(t n )];

[0093] For example, Figure 8 As shown, Figure 8It is a network structure block diagram of the LSTM sub-parameter dynamic response model provided by an embodiment of the present disclosure. The XLSTM is input into the LSTM in chronological order as the input of each time step. First, it passes through an LSTM layer. The cell state activation function of the LSTM input layer selects the tanh function, and the activation function of the forget gate, input gate and output gate selects the Sigmoid function; then, a dropout layer is added, and the random inactivation rate of the dropout layer is 0.9; then, a Re lu layer is added; finally, a fully connected layer is added to finally obtain the predicted output of the LSTM non-parametric dynamic model, and the output dimension of the fully connected layer is 1.

[0094] Step S402: Initialize the training parameters of the non-parametric dynamic response model.

[0095] In this embodiment, the training parameters of the LSTM non-parametric dynamic response model are initialized, including the optimization algorithm, the model evaluation index, the initial learning rate, the regularization factor, the learning rate reduction cycle, the learning rate reduction coefficient, and the gradient descent threshold.

[0096] For example, the optimization algorithm is Adam algorithm, and the algorithm parameters are set as follows: initial learning rate lr = 0.001, L2 regularization is 0.001, learning rate reduction cycle is 80, learning rate reduction coefficient is 0.2, and gradient descent threshold is 1. The training index of the LSTM non-parametric dynamic response model is expressed as the root mean square error function RMSE, as shown in formula (7):

[0097]

[0098] Among them, y pred is the predicted value, y real is the actual value.

[0099] Step S403: training the non-parametric dynamic response model and debugging the hyperparameters.

[0100] In this embodiment, the LSTM non-parametric dynamic response model training and hyperparameter debugging include training batches and training rounds. For example, the network hyperparameters are set to: training batches are 64, and training rounds are 150.

[0101] In some embodiments, Fig. 9 and Fig.10 As shown, Fig. 9 is a flow chart of the method S500 provided in an embodiment of the present disclosure, Fig.10 is a flow chart of the MOPSO optimization control strategy provided by the embodiment of the present disclosure, and the specific steps of S500 include:

[0102] Step S501: Initialize parameters of the target partition scheme.

[0103] In this embodiment, the initialization parameters include: determining the decision variable dimension, that is, the number and range of the control speed according to the target partitioning scheme; the MOPSO algorithm population size, the maximum number of iterations, the archive size, the inertia weight coefficient, the inertia weight decay rate, the individual learning factor, the overall learning factor, the number of grids in each dimension (the number of grids for each optimization target), the expansion rate, the mutation probability, the best selection factor, and the best deletion factor.

[0104] Exemplarily, the speed range of the optimization problem is determined according to the historical speed data range of the modeling sample, and is generally taken as [0.1*UR, 1.2*UR] (UR is the rated speed); the MOPSO algorithm population size is 40, the maximum number of iterations is 150, the archive size is 50, the inertia weight coefficient is 0.7, the inertia weight decay rate is 0.98, the individual learning factor is 1.7, the overall learning factor is 1.8, the number of grids in each dimension (the number of grids for each optimization target) is 7, the expansion rate is 0.1, the mutation probability is 0.5, the selection factor is 2, and the deletion factor is 2.

[0105] Step S502: Obtain the fan group inlet flow and fan speed according to the initialized target partitioning scheme, and calculate multiple target optimization solutions based on the fan group inlet air flow and fan speed.

[0106] In this embodiment, the fan energy consumption and the mechanical and electrical loss of the fan adjustment are calculated based on the fan group inlet air flow and fan speed. For example, 5s is used as a sampling period to collect the real-time natural wind speed, and the theoretical fan group inlet flow and fan speed in the previous speed adjustment period are considered. The MOPSO algorithm is designed based on the LSTM model prediction, and the fan group inlet air flow, fan energy consumption and fan adjustment mechanical and electrical loss are calculated by comprehensive consideration. The resulting set of Perato solutions provides multiple target optimization solutions.

[0107] For example, the three optimization objectives in each target optimization solution are: the air flow rate at the inlet of the fan group, the fan energy consumption, and the mechanical and electrical loss of the fan regulation. The inlet of the fan group is a maximum optimization problem, and the inverse calculation is adopted for it. Then, the input speed is calculated based on the air flow rate at the inlet of the fan group, the fan energy consumption, and the mechanical and electrical loss of the fan regulation. For example, the input speed is calculated by formulas (8)-(10):

[0108] Y(t)=Γ(U(t)) (8)

[0109]

[0110] Where t is time, s; U is the speed of the fan group, Γ is the LSTM model, Y is the inlet flow of the fan group, UR and PR are the rated speed and rated power of the fan respectively, and P is the fan power; L is the relative change rate of the fan speed, which indicates the mechanical loss of the fan speed regulation during operation.

[0111] Step S503, taking 5s as a speed adjustment cycle, calculate the target speeds in multiple target optimization solutions, and compare the target speed with the speed under the previous speed instruction. If the speed adjustment amount between the target speed and the speed under the previous speed instruction is greater than the preset threshold, the speed instruction is the target optimization adjustment instruction, otherwise the previous speed instruction is used.

[0112] In this embodiment, within a speed adjustment cycle of 5s, the target speed in the Perato solution is calculated by a weighted fuzzy membership function and compared with the previous speed instruction. If the speed adjustment amount is greater than the preset threshold, the speed instruction is the output speed adjustment instruction, that is, the target optimization adjustment instruction, otherwise the previous speed instruction is used.

[0113] For example, Figure 11-13 , Fig.11 is a comparison chart of the predicted value and the actual value of the LSTM model provided in the embodiment of the present disclosure, Fig.12 is a flow curve diagram of the fan group inlet provided by the embodiment of the present disclosure, Fig.13 It is a curve diagram of the fan speed of each partition provided by the embodiment of the present disclosure. First, the fuzzy membership function is calculated for the three optimization objectives corresponding to Perato, such as formula (11); the fuzzy membership function values ​​of the objective function are weighted and summed respectively, that is, the fuzzy membership function weighted summation is performed from the three aspects of the air flow at the inlet of the fan group, the fan energy consumption, and the mechanical and electrical loss of the fan regulation to obtain the target speed, and then the values ​​of the target speeds are compared, and the decision variable corresponding to the smallest value of the weighted summation result is taken as the output speed adjustment instruction, that is, the target optimization adjustment instruction, where the weights of the three optimization objectives are 0.7, 0.25, and 0.05; the absolute speed change threshold is 5rpm. It should be noted that the mechanical and electrical loss of the fan regulation is taken into account, that is, the equipment loss problem of the speed change gear and the frequency conversion electrical device is taken into account during the speed regulation process, which is helpful to achieve economical and stable operation of the air cooling unit. The formula is as follows:

[0114]

[0115] In the above formula, is the Nobjth objective function The membership function value of the i-th solution, Nobj is 3, i is the number of Perato solution sets, that is, the number of archives is 50. The target optimization adjustment instruction can be obtained through the above formula, so as to adjust the inlet flow of the fan group according to the target optimization adjustment instruction, thereby reducing the fan speed adjustment cost.

[0116] Based on the above method, the embodiment of the present disclosure also provides a MOPSO-based direct air cooling unit control system 2000 corresponding to the above method, such as Fig.14 As shown, Fig.14 This is a block diagram of a control system for a MOPSO direct air cooling unit provided in an embodiment of the present disclosure. The control system includes:

[0117] The sample unit 210 is used to obtain the operation data of the air cooling unit in the historical normal working state, and respectively constitute the partition sample and the modeling sample;

[0118] A preprocessing unit 220, used for performing data preprocessing on partition samples and modeling samples;

[0119] A partitioning unit 230, used to partition the partition samples of the data preprocessing and determine a target partitioning scheme through a silhouette coefficient;

[0120] The model building unit 240 is used to identify the modeling samples of the data preprocessing and establish a non-parametric dynamic response model of the inlet flow of the fan group of the air cooling unit;

[0121] The calculation unit 250 is used to obtain the fan group inlet flow and fan speed according to the target partitioning scheme, obtain multiple target optimization solutions based on the fan group inlet flow, fan speed and the non-parametric dynamic response model of the air-cooling unit fan group inlet flow, calculate and optimize the multiple target optimization solutions to determine the target optimization adjustment instructions.

[0122] In this embodiment, the target partitioning scheme is determined by partitioning samples, and a non-parametric dynamic response model of the fan group inlet flow is established by modeling samples. Then, the fan group inlet flow and fan speed are obtained based on the target partitioning scheme, and then multiple target optimization solutions are obtained in the non-parametric dynamic response model of the fan group inlet flow, so as to find the best solution among multiple target optimization solutions, thereby facilitating the optimization of each control cycle to solve the target optimization adjustment instructions, so as to realize the economical and safe operation of the direct air cooling unit.

[0123] It should be noted that, regarding the system in the above embodiment, the specific manner in which each unit module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0124] Based on the same inventive concept as the above disclosed content, the present disclosure also provides an electronic device 3000. Fig.15 As shown, Fig.15 1 is a structural block diagram of an electronic device 3000 provided by the present disclosure. The electronic device 3000 of the embodiment of the present disclosure includes at least one processor 310 and at least one memory 320 that are electrically connected. The memory 320 is electrically connected to the processor 310, wherein the memory 320 stores instructions that can be executed by the at least one processor 310, and the instructions are executed by the at least one processor 310 so that the at least one processor 310 can execute the above method.

[0125] It should be noted that the electrical connection between the above-mentioned units does not necessarily mean the connection between lines. An indirect connection method can be applied to the embodiments of the present disclosure as long as the purpose of the present disclosure is achieved.

[0126] Based on the same inventive concept, the present disclosure also provides a computer storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described above is performed. The storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0127] Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A control method for a direct air cooling unit based on MOPSO, characterized in that: The method comprises: Obtain the operating data of the air-cooling unit in the historical normal working state to form partition samples and modeling samples respectively; Performing data preprocessing on the partition samples and the modeling samples; Partitioning the partition samples of the data preprocessing, and determining a target partitioning scheme through a silhouette coefficient; Identify the modeling samples of data preprocessing and establish a non-parametric dynamic response model of the inlet flow of the fan group of the air cooling unit; The fan group inlet flow and fan speed are obtained according to the target partitioning scheme, and multiple target optimization solutions are obtained based on the fan group inlet flow, fan speed and the non-parametric dynamic response model of the air-cooling unit fan group inlet flow. The multiple target optimization solutions are calculated and optimized to determine the target optimization adjustment instructions.

2. The method according to claim 1, characterized in that: The operation data of the air cooling unit in the historical normal working state is obtained to form a partition sample and a modeling sample respectively, including: The partition samples include the inlet flow data of each fan under the constant speed of the air cooling unit and the natural wind disturbance; The modeling samples include continuous data of the fan speeds, natural wind speed and fan inlet flow rate of the air-cooling unit under dual random disturbances of natural wind and speed command.

3. The method according to claim 2, characterized in that The operation data covers the dynamic changes of all parameters of the air-cooling unit under all working conditions, and the interval time of the operation data set is at most 5s.

4. The method according to claim 1, characterized in that The data preprocessing of the partition samples and the modeling samples comprises: The operation data of the partition samples and the modeling samples are normalized.

5. The method according to claim 1, characterized in that Partitioning the partition samples of the data preprocessing and determining the target partition scheme by the silhouette coefficient comprises: Arrange the partition samples into the K-means input matrix X K-means ; Based on the input matrix X K-means The number of partition samples is set to 2, 3, ..., 6, and the wind turbine group is classified by K-means algorithm, and each partition scheme is recorded; The silhouette coefficients of the partition schemes are calculated and compared, and the partition scheme with the largest silhouette coefficient is selected as the target partition scheme.

6. The method according to claim 1, characterized in that The method of identifying the modeling samples of the data preprocessing and establishing a non-parametric dynamic response model of the inlet flow of the fan group of the air cooling unit includes: Construct the non-parametric dynamic response model structure of the inlet flow of the fan group of the air-cooling unit; Initializing the training parameters of the non-parametric dynamic response model; The non-parametric dynamic response model is trained and hyperparameters are debugged.

7. The method according to claim 1, characterized in that The step of obtaining the fan group inlet flow and fan speed according to the target partitioning scheme, obtaining multiple target optimization solutions based on the fan group inlet flow, fan speed and the non-parametric dynamic response model of the fan group inlet flow of the air cooling unit, and calculating and optimizing the multiple target optimization solutions to determine the target optimization adjustment instructions includes: Initializing parameters of the target partition scheme; The fan group inlet flow and fan speed are obtained according to the initialized target partitioning scheme, and multiple target optimization solutions are calculated based on the fan group inlet air flow and fan speed. Taking 5s as a speed adjustment cycle, the target speeds in the multiple target optimization solutions are calculated, and the target speed is compared with the speed under the previous speed instruction. If the speed adjustment amount between the target speed and the speed under the previous speed instruction is greater than the preset threshold, the speed instruction is the target optimization adjustment instruction, otherwise the previous speed instruction is used.

8. The method according to claim 7, characterized in that The initialization parameters of the target partition scheme include: Determine the decision variable dimension, i.e., the number of control variable speeds, according to the target partitioning scheme; The speed range of the optimization problem is determined according to the historical speed data range of the modeling samples.

9. The method according to claim 7, characterized in that: The step of obtaining the fan group inlet flow rate and fan speed according to the initialized target partitioning scheme, and calculating multiple target optimization solutions based on the fan group inlet air flow rate and fan speed includes: Calculate the fan energy consumption and the fan adjustment mechanical and electrical losses based on the fan group inlet air flow and fan speed; The input rotation speed is calculated based on the inlet air flow of the fan group, the fan energy consumption, and the fan adjustment mechanical and electrical losses.

10. The method according to claim 7, characterized in that The method of taking 5s as a speed adjustment cycle, calculating the target speeds in the multiple target optimization solutions, and comparing the target speed with the speed under the last speed instruction, if the speed adjustment amount between the target speed and the speed under the last speed instruction is greater than a preset threshold, the speed instruction is the target optimization adjustment instruction, otherwise the last speed instruction is used, including: The fuzzy membership functions of the input speeds of the fan group inlet air flow, fan energy consumption, and fan adjustment mechanical and electrical losses are calculated and weighted summed to obtain the target speed, and the decision variable of the target speed corresponding to the minimum result is taken as the target optimization adjustment instruction.

11. A control system for a direct air cooling unit based on MOPSO, characterized in that: The system comprises: The sample unit is used to obtain the operation data of the air-cooling unit in the historical normal working state, and constitutes a partition sample and a modeling sample respectively; A preprocessing unit, used for performing data preprocessing on the partition samples and the modeling samples; A partitioning unit, used for partitioning the partition samples of the data preprocessing and determining a target partitioning scheme through a silhouette coefficient; A model building unit, used to identify the modeling samples of data preprocessing and establish a non-parametric dynamic response model of the inlet flow of the fan group of the air cooling unit; A calculation unit is used to obtain the fan group inlet flow and fan speed according to the target partitioning scheme, obtain multiple target optimization solutions based on the fan group inlet flow, fan speed and the non-parametric dynamic response model of the air-cooling unit fan group inlet flow, calculate and optimize the multiple target optimization solutions to determine the target optimization adjustment instructions.

12. An electronic device, characterized in that: The electronic device comprises: Memory for storing computer programs; A processor, configured to implement the MOPSO-based direct air cooling unit control method as claimed in any one of claims 1 to 10 when executing the computer program.

13. A computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the MOPSO-based direct air cooling unit control method according to any one of claims 1 to 10.