Data-driven coal-fired unit circulating water pump operation optimization method

Abstract

The invention discloses a data-driven operation optimization method for a circulating water pump of a coal-fired unit, which comprises the following steps of: performing variable primary selection, data preprocessing and feature selection through a historical database of unit operation, and determining key operation parameters which influence the power consumption of the circulating water pump; working condition division is carried out by adopting an improved GMM algorithm based on an akaike information criterion, and a plurality of combined working conditions are obtained through cross combination; key operation parameters influencing the power consumption of the circulating water pump are weighted through a grey relational analysis method and a CRITIC method, and then weight combination is carried out; a Bayesian classifier is adopted to establish an operation optimization model of the circulating water pump, historical optimal operation data is learned, and based on combination of a historical optimal working condition library and real-time data, the optimal operation mode of the circulating water pump can be accurately determined under complex working conditions. The operation optimization method can be suitable for complex working conditions, dynamically adjusts operation suggestions, is high in matching degree, is more reasonable and objective, is good in accuracy, and reduces energy waste.

Description

Technical Field

[0001] The invention relates to the technical field of water pump operation management, and in particular to a data-driven coal-fired unit circulating water pump operation optimization method. Background Art

[0002] The circulating water pump is one of the auxiliary machines with the largest power consumption in coal-fired power plants, and its power consumption accounts for about 1% to 1.5% of the total power generation of the power plant. The power consumption rate of the circulating water pump is closely related to its operation mode. Therefore, it is an important measure to improve the economic efficiency of power plant operation to study the optimal operation mode of the circulating water pump under certain environmental and turbine load conditions and determine the reasonable number of circulating water pumps in operation.

[0003] The commonly used methods to reduce the energy consumption of circulating water systems are as follows: First, to transform and optimize the cold end of the power plant, such as frequency conversion speed regulation of the circulating water pump, to achieve the purpose of throttling by reducing the speed and flow rate; second, the optimal vacuum method, that is, under the conditions of certain unit load and cooling water temperature, increase the circulating water flow rate so that the difference between the increment of turbine power and the increment of circulating water pump energy consumption reaches the maximum. The corresponding circulating water flow rate is the optimal circulating water flow rate. The above circulating water system optimization methods are all based on mechanism models. Due to the large project investment, difficult to determine boundary conditions and low model accuracy, their application effects have certain limitations.

[0004] At present, although some public technical solutions in the prior art can solve the problem of high energy consumption of circulating water pumps of coal-fired units, there are still some defects. For example: the Chinese patent with announcement number CN107829924B uses big data to control the most energy-saving water pump, but this method mainly maintains the frequency of different water pumps, ignores the impact of key operating factors on water pump energy consumption under different working conditions, and fails to fully consider the energy consumption problem under complex working conditions. The Chinese patent with announcement number CN114576149B mainly relies on the optimization model of historical data. Although self-learning is mentioned, its feedback mechanism is relatively lagging and cannot cope with real-time changes in working conditions. It may not be flexible enough in actual operation. The Chinese patent with announcement number CN111894828A, its overall optimization method still remains in the logical optimization method based on empirical rules, and no more advanced data-driven or intelligent algorithms are proposed to assist the optimization process. The limitations of this method may make it difficult to achieve more significant energy-saving effects when facing more complex operating conditions. Summary of the invention

[0005] The purpose of the present invention is to provide a data-driven coal-fired unit circulating water pump operation optimization method to solve the above defects.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A data-driven coal-fired unit circulating water pump operation optimization method comprises the following steps:

[0008] S1. Analysis of factors affecting power consumption of circulating water pump:

[0009] Based on the historical database of unit operation, variables are initially selected to obtain possible factors affecting the power consumption of the circulating water pump. Then data preprocessing and feature selection are performed to determine the key operating parameters affecting the power consumption of the circulating water pump.

[0010] S2. Working conditions division:

[0011] The sliding window detection technology is used to identify and screen the steady-state working conditions of the generator power, determine the steady-state working conditions of the unit in stable operation in the historical data, and build a steady-state database of the unit; then the working conditions of the generator power and the circulating water inlet temperature are divided respectively, and then they are combined in pairs to obtain the combined working conditions;

[0012] S3. Build the optimal working condition database:

[0013] Two weighting methods based on grey correlation analysis and CRITIC method were used to analyze the weights of the key operating parameters that affect the power consumption of the circulating water pump, and then weighted combination was performed to construct an evaluation function. The evaluation function was used to automatically optimize the operating data of different combined operating conditions and build a database of the optimal operating conditions of the unit.

[0014] S4. Constructing the circulating water pump operation optimization model:

[0015] Based on the optimal operating condition database of the unit, a circulating water pump operation optimization model is established using the Bayesian classifier. The real-time operating data of the randomly selected circulating water pumps are then input into the circulating water pump operation optimization model to output the recommended combination of two circulating water pumps and the recommended circulating pump operating current, thereby achieving the operation optimization of the circulating water pump of the coal-fired unit.

[0016] Preferably, in step S1, possible factors affecting the power consumption of the circulating water pump include: generator power, circulating water inlet temperature, circulating water outlet temperature, low-pressure condenser pressure, exhaust temperature, main feed water flow, main feed water pressure, main steam temperature, main steam pressure, condensate flow, condensate temperature, and circulating pump current.

[0017] Preferably, in step S1, the data preprocessing includes missing value processing, outlier detection, and normalization processing.

[0018] Preferably, in step S1, the feature selection comprises the following specific steps: using the preprocessed data of possible factors affecting the power consumption of the circulating water pump as input data and the circulating pump current as output data, performing feature selection, utilizing MIC mutual information, retaining operating parameters with large correlation whose mutual information number is greater than 0.3, thereby determining the key operating parameters affecting the power consumption of the circulating water pump.

[0019] Preferably, in step S2, the sliding window detection technology is specifically:

[0020] (1) On one-dimensional time series data, select s data in a period of time as a window and slide along the time direction, and calculate the relative standard deviation δ of the s data. During the sliding process, the data in the window and δ are continuously updated. The calculation formula of the relative standard deviation δ is as follows:

[0021]

[0022] In the formula, δ is the relative standard deviation of the dimensionless quantity, is the average value of all data in the sliding window, s represents the number of data points in the window, and x i Represents the value of the i-th data point in the window;

[0023] (2) When the data in a certain window satisfies the condition that the relative standard deviation δ is less than the predefined threshold ε, that is, δ<ε, it is considered that the data fluctuation in the window is small and the system is in a steady state; the predefined threshold ε is the accuracy requirement of the relative standard deviation. When δ<ε is satisfied, the generator power operating condition meets the data steady-state screening requirements.

[0024] Preferably, in step S2, a Gaussian clustering algorithm GMM improved based on Akaike information criterion AIC is used to perform Gaussian clustering and operating condition division on the generator power and circulating water inlet temperature in the steady-state data, respectively, and the cluster number corresponding to the minimum AIC number is taken as the optimal operating condition number to obtain the operating condition division result.

[0025] Preferably, the S3 step includes the following specific steps:

[0026] S31. Grey correlation analysis weight analysis:

[0027] Grey correlation analysis method studies the size of data correlation and measures the degree of correlation between data through correlation degree, thus determining the feature weight;

[0028] (1) Determine the analysis series;

[0029] The key operating parameters that affect the power consumption of the circulating water pump are used as the characteristic sequence of the grey relational analysis, and the circulating water pump current is used as the parent sequence;

[0030] (2) Variables are dimensionless;

[0031] Since the dimensional differences between variables will affect the training results of the model, the data is normalized to the maximum and minimum values ​​to unify the variable range to [0,1]. The specific formula is as follows:

[0032]

[0033] In the formula, x' i is the normalized value, x i is the actual value, x max is the maximum value of the variable, x min is the minimum value of the variable;

[0034] (3) Calculate the correlation coefficient;

[0035] The correlation coefficient formula of grey relational analysis is:

[0036]

[0037] Where k = 1, 2, ..., m; y(k) is the kth sample in the parent sequence; x i (k) is the i-th feature of the k-th sample in the subsequence; n is the number of samples in the parent sequence and the subsequence; m is the number of subsequence features; ρ is the resolution coefficient, 0<ρ<1;

[0038] (4) Calculate the degree of association;

[0039] The formula of correlation degree ri is as follows:

[0040]

[0041] In the formula, ξ i (k) is the correlation coefficient, n is the number of characteristic sequence samples;

[0042] (5) Determine the indicator weights;

[0043] Weight coefficient w i The formula is as follows:

[0044]

[0045] (6) Determine the weight coefficient;

[0046] According to the grey correlation analysis method, the weight coefficients of each index relative to the energy consumption of the circulating water pump are determined as follows: generator power 0.1448, low-pressure condenser pressure 0.1419, circulating water inlet temperature 0.1549, condensate temperature 0.1331, low-pressure cylinder exhaust temperature 0.1419, main steam temperature 0.1381, circulating water outlet temperature 0.1454;

[0047] S32, CRITIC method weight analysis:

[0048] The CRITIC method is to comprehensively measure the objective weight of the indicators based on the comparative strength of the evaluation indicators and the conflict between the indicators; the comparative strength of the indicators is measured by the standard deviation, and the conflict between the indicators is measured by the correlation coefficient;

[0049] (1) Determine the analysis series;

[0050] The key operating parameters that affect the power consumption of the circulating water pump are used as the characteristic sequence for CRITIC method analysis;

[0051] (2) Variables are dimensionless;

[0052] (3) Calculate the standard deviation;

[0053] The standard deviation reflects the amount of information contained in the indicator. The larger the standard deviation, the more information it contains and the higher the weight of the corresponding indicator. The formula for the standard deviation Sj is as follows:

[0054]

[0055] In the formula, S j represents the standard deviation of the jth indicator, x ij represents the observed value of the i-th sample under the j-th index, represents the mean of the jth indicator, and m is the total number of samples;

[0056] (4) Calculate the correlation coefficient;

[0057] The conflict between indicators is measured by the correlation coefficient, which reflects the size of the correlation between indicators. The larger the correlation coefficient, the smaller the conflict and the smaller the weight of the corresponding indicator. The correlation coefficient formula is as follows:

[0058]

[0059]

[0060] In the formula, σ jk represents the correlation coefficient between the jth indicator and the kth indicator, A j is the conflict of the jth indicator, j∈[1,n], k∈[1,n], x ij represents the observed value of the i-th sample on the j-th index, represents the mean value of the jth indicator, represents the mean of the kth index, x ik Represents the observed value of the i-th sample on the k-th indicator;

[0061] (5) Determine the amount of information Qj;

[0062] The amount of information Qj is obtained by the comparison strength and conflict of the indicators. The larger the Qj, the greater the role of the jth indicator in the entire evaluation indicator system. The formula for the amount of information Qj is as follows:

[0063] Qj=SjAj;

[0064] (6) Determine the indicator weight w j ;

[0065] Weight coefficient w j The formula is as follows:

[0066]

[0067] (7) Determine the weight coefficient;

[0068] According to the CRITIC method, the weight coefficient of each indicator relative to the energy consumption of the circulating water pump is determined:

[0069] Generator power 0.3889, low-pressure condenser pressure 0.0863, circulating water inlet temperature 0.2101, condensate temperature 0.0609, low-pressure cylinder exhaust temperature 0.0596, main steam temperature 0.1320, circulating water outlet temperature 0.0623;

[0070] S33, weighted combination:

[0071] The weight coefficient determined by the grey relational analysis method and the weight coefficient determined by the CRITIC method are weighted together to obtain the combined weight. The weighted formula is as follows:

[0072]

[0073] In the formula, wk represents the combined weight coefficient, wi is the weight coefficient obtained by the grey relational analysis method, and wj is the weight coefficient obtained by the CRITIC method;

[0074] According to the weighted formula, the combined evaluation coefficient is:

[0075] Generator power 0.352, low-pressure condenser pressure 0.09, circulating water inlet temperature 0.2668, condensate temperature 0.053, low-pressure cylinder exhaust temperature 0.056, main steam temperature 0.119, circulating water outlet temperature 0.061;

[0076] S34, obtaining evaluation function:

[0077] The evaluation function between each operating parameter and energy consumption based on the evaluation weight is:

[0078] IF=0.2885x1+0.086x2+0.3954x3+0.057x4+0.059x5+0.2x6+0.063x7,

[0079] In the formula, x1 is the circulating water inlet temperature, x2 is the low-pressure condenser pressure, x3 is the generator power, x4 is the condensate temperature, x5 is the low-pressure cylinder exhaust temperature, x6 is the main steam temperature, and x7 is the circulating water outlet temperature;

[0080] S35, Optimizing working conditions:

[0081] The evaluation function is used to automatically optimize the historical operating data of different combined operating conditions, calculate the difference between the evaluation index value and the actual operating value, and take the data with the largest difference in a certain combined clustering condition as the optimal operation record in the historical operation process to construct the optimal operating condition database of the unit's historical operation.

[0082] Preferably, the S4 step comprises the following specific steps:

[0083] S41. Establishing an optimization model:

[0084] Based on the historical optimal operating condition database of the unit, a circulating water pump operation optimization model was established using a Bayesian classifier. The key operating parameters that affect the power consumption of the circulating water pump were used as input parameters of the Bayesian classifier, and the decimal-coded classification labels of the two circulating water pumps were used as output.

[0085] S42, model training and learning:

[0086] The data of the unit's optimal operating condition database is divided into a training set and a test set in a ratio of 8:2. Then, the circulating water pump operation optimization model established by the Bayesian classifier is used to learn the unit's historical optimal operating condition database to obtain the optimal circulating pump combination corresponding to the unit's optimal operating condition database.

[0087] S43. Optimize model output:

[0088] By real-time monitoring of the unit's operating data, when new operating conditions appear in the system, a Bayesian classifier is used to establish a circulating water pump operation optimization model. Based on previously learned knowledge, it can predict and output the recommended combination of the two circulating water pumps and the recommended circulating pump operating current, thereby achieving the operation optimization of the circulating water pump of the coal-fired unit.

[0089] The beneficial effects of the present invention are:

[0090] (1) The present invention provides a data-driven method for optimizing the operation of circulating water pumps of coal-fired units. The method adopts an improved GMM algorithm based on the Akaike information criterion to divide the operating conditions. A plurality of combined operating conditions are obtained through cross-combination, and the operating conditions are accurately divided to achieve full coverage of the unit operation conditions, thereby ensuring the adaptability of circulating water pump optimization under different load conditions and circulating water inlet temperature conditions. Compared with the relatively rough operating condition classification method in the prior art, the precise operating condition division of the present invention can better reflect the actual operating conditions, thereby providing a more accurate optimization strategy.

[0091] (2) The present invention provides a data-driven method for optimizing the operation of a circulating water pump of a coal-fired unit. The method weights the key operating parameters that affect the power consumption of the circulating water pump through the grey correlation analysis method and the CRITIC method, respectively. The weights obtained by the two methods are then combined using a weighted formula, thereby ensuring the rationality and objectivity of the weight calculation.

[0092] (3) The present invention provides a data-driven method for optimizing the operation of circulating water pumps of coal-fired units. The method adopts a Bayesian classifier to learn historical optimal operation data and establishes an operation optimization model for circulating water pumps. The model can quickly output the optimal combination of circulating water pumps and the recommended operating current of circulating water pumps according to real-time input data. Based on the combination of the historical optimal operating condition library and real-time data, the optimization model can adapt to the real-time changing operating conditions and provide the optimal operation suggestions. The present invention can accurately determine the optimal operation mode of the circulating water pump under complex operating conditions. Compared with the prior art, the technology can dynamically adjust the operation suggestions to adapt to the changes in operating conditions, ensuring that the optimal operation strategy can be provided under various operating conditions to match the actual operation of the unit, thereby reducing energy waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 is a flow chart of a method for optimizing the operation of a circulating water pump according to an embodiment of the present invention;

[0094] Figure 2 : is the AIC curve diagram corresponding to the power operating condition division of the generator in the embodiment of the present invention;

[0095] Figure 3 : is a diagram showing the division of the power operating conditions of the generator in an embodiment of the present invention;

[0096] Figure 4 : is the AIC curve corresponding to the circulating water inlet temperature working condition division in the embodiment of the present invention;

[0097] Figure 5 : is a diagram showing the division of circulating water inlet temperature working conditions in an embodiment of the present invention;

[0098] Figure 6 It is an optimal operation data diagram obtained by searching for an optimal operation under a certain combination of working conditions in an embodiment of the present invention;

[0099] Figure 7 is a diagram of the operating optimization model of a circulating water pump constructed in an embodiment of the present invention;

[0100] Figure 8 It is a comparison diagram of circulating water pump current before and after optimization in an embodiment of the present invention. DETAILED DESCRIPTION

[0101] The present invention is further described below in conjunction with the embodiments. It should be noted that these are merely examples and illustrations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they should be deemed to fall within the protection scope of the present invention.

[0102] Embodiment 1:

[0103] Figure 1 It is a schematic diagram of the process of the present invention, such as Figure 1 As shown, a data-driven coal-fired unit circulating water pump operation optimization method includes the following steps:

[0104] S1. Analysis of factors affecting power consumption of circulating water pump.

[0105] Based on the historical database of unit operation, the historical operation data of the unit is obtained. The research data comes from the effective operation data of Unit 1 of a power plant from January to December 2023, a total of 30,603 data, with a time interval of 15 minutes. The unit has two circulating water pumps AB, and both circulating water pump motors are three-phase motors, so the power consumption of the circulating water pump can be represented by its current.

[0106] Based on the historical operation data of the unit and combined with the operation mechanism of the circulating water system, a preliminary selection of variables is carried out to obtain possible factors affecting the power consumption of the circulating water pump, including: generator power, circulating water inlet temperature, circulating water outlet temperature, low-pressure condenser pressure, exhaust temperature, main feed water flow, main feed water pressure, main steam temperature, main steam pressure, condensate flow, condensate temperature, circulating pump current, etc., so as to obtain historical operation data that may affect the power consumption of the circulating water pump.

[0107] Data preprocessing is performed on the historical operating data of the circulating water pump power consumption, including missing value processing, outlier detection, and normalization processing.

[0108] Then perform feature selection on the preprocessed data set. The specific steps are as follows:

[0109] (1) The preprocessed data of possible factors affecting the power consumption of the circulating water pump are used as input data, and the circulating pump current is used as the output parameter for feature selection to perform feature selection.

[0110] (2) In general, the absolute value interval of the correlation coefficient [0, 0.3] indicates that the variables have weak correlation, the absolute value interval [0.3, 0.7] indicates moderate correlation, and the absolute value interval [0.7, 1] indicates strong correlation. MIC is a measure of variable correlation. The size of the MIC value between variables represents the size of their correlation. The larger the MIC, the greater the correlation between the variables. When variables X and Y are independent of each other and have no relevant information, the minimum mutual information is 0.

[0111] By using MIC mutual information, the operating parameters with high correlation and mutual information number greater than 0.3 are retained to determine the key operating parameters affecting the power consumption of the circulating water pump. Seven operating factors, including generator power, circulating water inlet temperature, circulating water outlet temperature, low-pressure condenser pressure, exhaust temperature, main steam temperature, and condensate temperature, are obtained as the key factors affecting the power consumption of the circulating water pump.

[0112] S2. Working conditions division:

[0113] S21. Use sliding window detection technology to identify and screen the steady-state operating conditions of the generator power, determine the steady-state operating conditions of the unit in stable operation in historical data, and build a steady-state database for the unit.

[0114] Among them, the sliding window detection technology is specifically:

[0115] (1) In one-dimensional time series data, select s data in a period of time as a window to slide along the time direction, and calculate the relative standard deviation δ of the s data. During the sliding process, the data in the window and δ are continuously updated. The calculation formula of the relative standard deviation δ is as follows:

[0116]

[0117] In the formula, δ is the relative standard deviation of the dimensionless quantity, is the average value of all data in the sliding window, s represents the number of data points in the window, and x i Represents the value of the i-th data point in the window.

[0118] (2) When the data in a certain window satisfies the condition that the relative standard deviation δ is less than the predefined threshold ε, that is, δ<ε, it is considered that the data fluctuation in the window is small and the system is in a steady state; the predefined threshold ε is the accuracy requirement of the relative standard deviation. In the embodiment of the method of the present invention, ε=0.01 is selected. When δ<ε is satisfied, the generator power operating condition meets the data steady-state screening requirements.

[0119] S22. Gaussian clustering algorithm GMM based on Akaike information criterion AIC is used to perform Gaussian clustering and operating condition division on generator power and circulating water inlet temperature in steady-state data. The cluster number corresponding to the minimum AIC number is the optimal operating condition number. The operating condition division result is obtained. The clustering effect is as follows: Figures 2 to 5 As shown,

[0120] in Figure 3 and Figure 5 They are the cluster distribution results of generator power and circulating water inlet temperature, Figure 2 and Figure 4 They are the AIC values ​​under different operating condition clustering numbers. It can be seen that when the generator power is clustered into 8 types of operating conditions and the circulating water inlet temperature is clustered into 6 types of operating conditions, the AIC number is the smallest. The clustering results of the generator power and the circulating water inlet temperature are combined in pairs, and a total of 48 combined operating conditions are obtained. The results of the combined operating condition division are shown in Table 1.

[0121] Table 1. Combined working condition division results

[0122]

[0123]

[0124] Akaike Information Criterion (AIC) is a criterion for evaluating the complexity of statistical models and measuring the goodness of fit of statistical models. AIC is based on the concept of information entropy. The smaller the AIC, the better the model. Usually, the number of clusters with the smallest AIC is selected as the optimal number of clusters. AIC is used to evaluate the effect of working condition division and determine the optimal number of working conditions. AIC can be expressed as:

[0125] AIC = 2k-2ln(L),

[0126] Where: k is the number of parameters, L is the likelihood function, a small k means a simple model, and a large L means an accurate model, so both simplicity and accuracy are taken into account when evaluating the model.

[0127] S3. Build the optimal working condition database:

[0128] Two weighting methods based on grey correlation analysis and CRITIC method are used to analyze the weights of the key operating parameters that affect the power consumption of the circulating water pump, and then weighted combination is performed to construct an evaluation function. The evaluation function is used to automatically optimize the operating data of different combined operating conditions and build a database of the optimal operating conditions of the unit. The specific steps are as follows:

[0129] S31. Grey correlation analysis weight analysis:

[0130] Grey correlation analysis method studies the size of data correlation (the degree of correlation between the parent sequence and the feature sequence) and measures the degree of correlation between data through the correlation degree (i.e. the size of correlation), thereby determining the feature weight.

[0131] (1) Determine the analysis sequence. The seven key operating parameters that affect the power consumption of the circulating water pump are used as the characteristic sequence of grey relational analysis, and the circulating water pump current is used as the parent sequence.

[0132] (2) Variable non-dimensionalization. Since the dimension difference between variables will affect the training results of the model, the data is normalized to the maximum and minimum values, and the variable range is unified to [0, 1]. The specific formula is as follows:

[0133]

[0134] In the formula, x' i is the normalized value, x i is the actual value, x max is the maximum value of the variable, x min is the minimum value of the variable.

[0135] (3) Calculate the correlation coefficient. The correlation coefficient formula of grey relational analysis is:

[0136]

[0137] Where k = 1, 2, ..., m; y(k) is the kth sample in the parent sequence; x i (k) is the i-th feature of the k-th sample in the subsequence; n is the number of samples in the parent sequence and the subsequence; m is the number of subsequence features; ρ is the resolution coefficient, 0<ρ<1.

[0138] (4) Calculate the correlation degree.

[0139] The formula of correlation degree ri is as follows:

[0140]

[0141] In the formula, q i (k) is the correlation coefficient, and n is the number of feature sequence samples.

[0142] (5) Determine the indicator weights.

[0143] Weight coefficient w i The formula is as follows:

[0144]

[0145] (6) Determine the weight coefficient.

[0146] In the grey correlation analysis method, the automatic calculation process of the weight coefficient is realized by writing Python code. The code sets the relevant formula in the grey correlation analysis method as a built-in function. After the data is prepared, the code can automatically calculate the weight coefficient of each indicator relative to the energy consumption of the circulating water pump, avoiding the tedious process of manual step-by-step calculation.

[0147] According to the grey correlation analysis method, the weight coefficient of each index relative to the energy consumption of the circulating water pump (total current of two circulating water pumps) is determined:

[0148] Generator power 0.1448, low-pressure condenser pressure 0.1419, circulating water inlet temperature 0.1549, condensate temperature 0.1331, low-pressure cylinder exhaust temperature 0.1419, main steam temperature 0.1381, circulating water outlet temperature 0.1454.

[0149] S32, CRITIC method weight analysis:

[0150] The CRITIC method comprehensively measures the objective weight of indicators based on the comparative strength of evaluation indicators and the conflict between indicators. The comparative strength of indicators is measured by standard deviation, and the conflict between indicators is measured by correlation coefficient. This method determines the weight of influencing factors while considering the comparative strength of indicators and the correlation between indicators.

[0151] (1) Determine the analysis sequence. The seven key operating parameters that affect the power consumption of the circulating water pump are used as the characteristic sequence for CRITIC analysis.

[0152] (2) Variable non-dimensionalization: Same as step S31.

[0153] (3) Calculate the standard deviation.

[0154] The standard deviation reflects the amount of information contained in the indicator. The larger the standard deviation, the more information it contains and the higher the weight of the corresponding indicator. The formula for the standard deviation Sj is as follows:

[0155]

[0156] In the formula, S j represents the standard deviation of the jth indicator, x ij represents the observed value of the i-th sample under the j-th index, represents the mean of the jth indicator, and m is the total number of samples.

[0157] (4) Calculate the correlation coefficient.

[0158] The conflict between indicators is measured by the correlation coefficient, which reflects the size of the correlation between indicators. The larger the correlation coefficient, the smaller the conflict and the smaller the weight of the corresponding indicator. The correlation coefficient formula is as follows:

[0159]

[0160]

[0161] In the formula, σ jk represents the correlation coefficient between the jth indicator and the kth indicator, A j is the conflict of the jth indicator, j∈[1,n], k∈[1,n]; x ij represents the observed value of the i-th sample on the j-th index, represents the mean value of the jth indicator, represents the mean of the kth index, x ik Represents the observed value of the i-th sample on the k-th index.

[0162] (5) Determine the amount of information Qj.

[0163] The amount of information Qj is obtained by the comparison strength and conflict of the indicators. The larger the Qj, the greater the role of the jth indicator in the entire evaluation indicator system. The formula for the amount of information Qj is as follows:

[0164] Qj=SjAj.

[0165] (6) Determine the indicator weight w j .

[0166] Weight coefficient w j The formula is as follows:

[0167]

[0168] (7) Determine the weight coefficient.

[0169] According to the CRITIC method, the weight coefficient of each indicator relative to the energy consumption of the circulating water pump is determined:

[0170] Generator power 0.3889, low-pressure condenser pressure 0.0863, circulating water inlet temperature 0.2101, condensate temperature 0.0609, low-pressure cylinder exhaust temperature 0.0596, main steam temperature 0.1320, circulating water outlet temperature 0.0623.

[0171] S33, weighted combination:

[0172] The weight coefficient determined by the grey relational analysis method and the weight coefficient determined by the CRITIC method are weighted together to obtain the combined weight. The weighted formula is as follows:

[0173]

[0174] Where wk represents the combined weight coefficient, wi is the weight coefficient obtained by the grey relational analysis method, and wj is the weight coefficient obtained by the CRITIC method.

[0175] According to the weighted formula, the combined evaluation coefficient is:

[0176] Generator power 0.3954, low-pressure condenser pressure 0.086, circulating water inlet temperature 0.2885, condensate temperature 0.057, low-pressure cylinder exhaust temperature 0.059, main steam temperature 0.2, circulating water outlet temperature 0.063.

[0177] S34, obtaining evaluation function:

[0178] In order to express the comprehensive impact of different operating parameters on the power consumption of the circulating water pump, the calculated combined weight coefficient and the key operating parameters affecting the power consumption of the circulating water pump are weighted and summed to establish an evaluation function, which represents the comprehensive impact of each key operating parameter on the power consumption of the circulating water pump. By calculating the value of IF, the overall impact of these parameters on the energy consumption of the circulating water pump under different working conditions can be quantified, providing a basis for subsequent optimization scheduling.

[0179] The evaluation function between each operating parameter and energy consumption based on the evaluation weight is:

[0180] IF=0.2885x1+0.086x2+0.3954x3+0.057x4+0.059x5+0.2x6+0.063x7,

[0181] Wherein, x1 is the circulating water inlet temperature, x2 is the low-pressure condenser pressure, x3 is the generator power, x4 is the condensate temperature, x5 is the low-pressure cylinder exhaust temperature, x6 is the main steam temperature, and x7 is the circulating water outlet temperature.

[0182] S35, Optimizing working conditions:

[0183] The evaluation function is used to automatically optimize the historical operating data of different combined operating conditions, calculate the difference between the evaluation index value and the actual operating value, and take the data with the largest difference in a certain combined clustering condition as the optimal operation record in the historical operation process to construct the optimal operating condition database of the unit's historical operation.

[0184] The obtained unit operation optimal condition database data is tested, and the operation data under a certain combination of conditions is selected. According to the optimization result, the circulating water pump current in the historical operation data under the same clustering condition is compared to verify whether the circulating pump current is at a low level. The optimization result is shown in the attached figure. Figure 6 shown.

[0185] Since the established evaluation function characterizes the influence of various operating parameters on the power consumption of the circulating pump, and the power consumption of the circulating pump is mainly characterized by the total operating current of the two circulating water pumps, when the operating data with a large difference between the calculated evaluation index value and the actual operating value is taken, the optimal operating data obtained is the operating data with relatively low power consumption of the circulating water pump under the same operating conditions.

[0186] S4. Constructing the circulating water pump operation optimization model:

[0187] Based on the historical optimal operating condition database of the unit, a circulating water pump operation optimization model is established using the Bayesian classifier. The real-time operating data of the randomly selected circulating water pump is then input into the circulating water pump operation optimization model to output the recommended combination of two circulating water pumps and the recommended circulating pump operating current, thereby realizing the operation optimization of the circulating water pump of the coal-fired unit. The specific steps are as follows:

[0188] S41. Establishing an optimization model:

[0189] The Bayesian classifier is a classification algorithm based on the Bayesian probability theory. Based on the Bayesian formula, the algorithm uses the prior probability and conditional probability of the variable to obtain the posterior probability, and uses the Bayesian maximum a posteriori criterion to infer the category to which the variable belongs. The Bayesian classifier is used to determine the optimal circulating water pump combination operation mode for the current operating conditions of the coal-fired unit.

[0190] Based on the historical optimal operating condition database of the unit, a Bayesian classifier was used to establish the circulating water pump operation optimization model. The seven key operating parameters that affect the power consumption of the circulating water pump were used as the input parameters of the Bayesian classifier, and the decimal-coded classification labels of the two circulating water pumps were used as output.

[0191] Take a power plant as an example. The two circulating water pumps in the power plant have a status of 1 when the pump is turned on and a status of 0 when the pump is turned off. There are three combinations of the two circulating water pumps, namely 01, 10, and 11. Convert the binary codes of the two circulating water pumps into decimal labels.

[0192] S42, model training and learning:

[0193] The data of the optimal operating condition database of the unit is divided into a training set and a test set in a ratio of 8:2. The optimization model is run as follows: Figure 7 As shown, the circulating water pump operation optimization model established by the Bayesian classifier is used to study the historical optimal operating condition database of the unit, and the optimal circulating pump combination corresponding to the historical optimal operating condition database of the unit is obtained.

[0194] S43. Optimize model output:

[0195] By real-time monitoring of the unit's operating data, when the system encounters a new operating condition, a Bayesian classifier is used to establish a circulating water pump operation optimization model. Based on previously learned knowledge, it can predict and output the recommended combination of two circulating water pumps and the recommended circulating pump operating current, thereby optimizing the operation of the circulating water pump of the coal-fired unit. In this way, the system can automatically select the best pump combination based on the current operating status, thereby ensuring that the unit is always in the optimal operating state under different operating conditions, improving efficiency and saving energy.

[0196] 500 real-time operating data of circulating water pumps are randomly selected and input into the circulating water pump operation optimization model, which outputs the recommended combination scheme of two circulating water pumps and the recommended circulating pump operating current.

[0197] According to the actual operation of the power plant, under the clustered conditions of the same generator power and circulating water inlet temperature, the lower the circulating water pump current, the more it meets the optimal operation. Due to factors such as actual operating conditions and measurement errors, the actual operating current of the circulating water pump fluctuates, so the total current of the recommended operating mode is the average current of the circulating water pump under the recommended operating state.

[0198] Table 2. Circulating water pump operation optimization data (partial)

[0199]

[0200] The data of circulating water pump operation optimization are shown in Table 2. Through the analysis of 500 circulating water pump operation data after optimization, among the 500 operation data, there are 141 data that realize operation optimization and reduce the power consumption of circulating pumps, and 351 data whose operation optimization results are consistent with the actual operation conditions. Among these 351 actual operation data, 242 data only turn on pump A, 57 data only turn on pump B, and 52 data both pumps are fully turned on. The classification results are consistent with the actual operation, which is the optimal pump opening method corresponding to this working condition, and the classifier is all classified correctly. Due to errors in the classifier, 8 data points were found whose power consumption was higher than the original combination and the operation optimization was not achieved, accounting for 1.6% of the total data volume. Among the 141 data points that reduced the power consumption of the circulating pump, 99 data points had both pumps turned on during actual operation, but only one circulating water pump turned on after optimization. The power consumption of this part of the circulating water pump was reduced by 53% compared with that before optimization. Another 42 data points actually turned on one pump B, while the recommended operation mode turned on pump A. Since the average operating current of pump A was less than that of pump B, this part was reduced by 23% compared with that before optimization.

[0201] Comparison of the actual operating current of the two circulating water pumps before optimization and the recommended operating current of the two circulating water pumps after optimization. Figure 8 shown.

[0202] According to this optimization method, for 500 randomly selected operating conditions, the total power consumption of the two circulating water pumps can be saved to 50% of the actual total power consumption before optimization. That is, the data-driven circulating water pump operation optimization method is feasible and can be further developed in engineering practice.

[0203] The present invention discloses a data-driven method for optimizing the operation of a circulating water pump of a coal-fired unit. The method adopts an improved GMM algorithm based on the Akaike information criterion to divide the operating conditions. A plurality of combined operating conditions are obtained through cross-combination, and the operating conditions are accurately divided to achieve full operating condition coverage of the unit, thereby ensuring the adaptability of the circulating water pump optimization under different load conditions and circulating water inlet temperature conditions. Compared with the relatively rough operating condition classification method in the prior art, the precise operating condition division of the present invention can better reflect the actual operating conditions, thereby providing a more accurate optimization strategy.

[0204] The present invention provides a data-driven method for optimizing the operation of a circulating water pump of a coal-fired unit. The key operating parameters that affect the power consumption of the circulating water pump are weighted by a grey correlation analysis method and a CRITIC method respectively. The weights obtained by the two methods are then combined using a weighted formula to ensure the rationality and objectivity of the weight calculation.

[0205] The present invention discloses a data-driven method for optimizing the operation of a circulating water pump of a coal-fired unit. The method adopts a Bayesian classifier to learn historical optimal operation data and establishes an operation optimization model for the circulating water pump. The model can quickly output the optimal combination of the circulating water pump and the recommended operating current of the circulating pump according to the real-time input data. Based on the combination of the historical optimal operating condition library and the real-time data, the method ensures that the optimization model can adapt to the real-time changing operating conditions and provide the optimal operation suggestions.

[0206] The present invention can accurately determine the optimal operating mode of the circulating water pump under complex working conditions. Compared with the existing technology, this technology can dynamically adjust the operation suggestions to adapt to changes in working conditions, ensuring that the optimal operation strategy can be provided under various operating conditions to match the actual operation of the unit, thereby reducing energy waste.

[0207] The above is an exemplary description of the invention. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as such non-substantial improvements are made by adopting the method concept and technical solution of the present invention, or the concept and technical solution of the invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A data-driven method for optimizing the operation of circulating water pumps of coal-fired units, characterized in that: The following steps are involved: S1. Analysis of factors affecting power consumption of circulating water pump: Based on the historical database of unit operation, variables are initially selected to obtain possible factors affecting the power consumption of the circulating water pump. Then data preprocessing and feature selection are performed to determine the key factors and key operating parameters affecting the power consumption of the circulating water pump. S2. Working conditions division: The sliding window detection technology is used to identify and screen the steady-state working conditions of the generator power, determine the steady-state working conditions of the unit in stable operation in the historical data, and build a steady-state database of the unit; then the working conditions of the generator power and the circulating water inlet temperature are divided respectively, and then they are combined in pairs to obtain the combined working conditions; S3. Build the optimal working condition database: Two weighting methods based on grey correlation analysis and CRITIC method were used to analyze the weights of the key operating parameters that affect the power consumption of the circulating water pump, and then weighted combination was performed to construct an evaluation function. The evaluation function was used to automatically optimize the operating data of different combined operating conditions and build a database of the optimal operating conditions of the unit. S4. Constructing the circulating water pump operation optimization model: Based on the optimal operating condition database of the unit, a circulating water pump operation optimization model is established using the Bayesian classifier. The real-time operating data of the randomly selected circulating water pumps are then input into the circulating water pump operation optimization model to output the recommended combination of two circulating water pumps and the recommended circulating pump operating current, thereby achieving the operation optimization of the circulating water pump of the coal-fired unit.

2. A data-driven coal-fired unit circulating water pump operation optimization method according to claim 1, characterized in that: In step S1, possible factors affecting the power consumption of the circulating water pump include: generator power, circulating water inlet temperature, circulating water outlet temperature, low-pressure condenser pressure, exhaust temperature, main feed water flow, main feed water pressure, main steam temperature, main steam pressure, condensate flow, condensate temperature, and circulating pump current.

3. A data-driven coal-fired unit circulating water pump operation optimization method according to claim 1, characterized in that: In step S1, the data preprocessing includes missing value processing, outlier detection, and normalization processing.

4. A data-driven coal-fired unit circulating water pump operation optimization method according to claim 1, characterized in that: In step S1, the feature selection specifically comprises the following steps: taking the preprocessed data of possible factors affecting the power consumption of the circulating water pump as input data and the circulating pump current as output data, performing feature selection, and utilizing MIC mutual information to retain operating parameters with large correlation whose mutual information number is greater than 0.3, thereby determining the key factors and key operating parameters affecting the power consumption of the circulating water pump.

5. A data-driven coal-fired unit circulating water pump operation optimization method according to claim 1, characterized in that: In step S2, the sliding window detection technology is specifically: (1) On one-dimensional time series data, select s data in a period of time as a window and slide along the time direction, and calculate the relative standard deviation δ of the s data. During the sliding process, the data in the window and δ are continuously updated. The calculation formula of the relative standard deviation δ is as follows: In the formula, δ is the relative standard deviation of the dimensionless quantity, is the average value of all data in the sliding window, s represents the number of data points in the window, and x i Represents the value of the i-th data point in the window; (2) When the data in a certain window satisfies the condition that the relative standard deviation δ is less than the predefined threshold ε, that is, δ<ε, it is considered that the data fluctuation in the window is small and the system is in a steady state; the predefined threshold ε is the accuracy requirement of the relative standard deviation. When δ<ε is satisfied, the generator power operating condition meets the data steady-state screening requirements.

6. A data-driven coal-fired unit circulating water pump operation optimization method according to claim 1, characterized in that: In step S2, the Gaussian clustering algorithm GMM improved based on Akaike information criterion AIC is used to perform Gaussian clustering and operating condition division on the generator power and circulating water inlet temperature in the steady-state data respectively. The cluster number corresponding to the minimum AIC number is taken as the optimal operating condition number, and the operating condition division result is obtained.

7. A data-driven coal-fired unit circulating water pump operation optimization method according to claim 1, characterized in that: The specific steps of step S3 are as follows: S31. Grey correlation analysis weight analysis: Grey correlation analysis method studies the size of data correlation and measures the degree of correlation between data through correlation degree, thus determining the feature weight; (1) Determine the analysis sequence; The key operating parameters that affect the power consumption of the circulating water pump are used as the characteristic sequence of the grey relational analysis, and the circulating water pump current is used as the parent sequence; (2) Variables are dimensionless; Since the dimensional differences between variables will affect the training results of the model, the data is normalized to the maximum and minimum values ​​to unify the variable range to [0,1]. The specific formula is as follows: In the formula, x' i is the normalized value, x i is the actual value, x max is the maximum value of the variable, x min is the minimum value of the variable; (3) Calculate the correlation coefficient; The correlation coefficient formula of grey relational analysis is: Where k = 1, 2, ..., m; y(k) is the kth sample in the parent sequence; x i (k) is the i-th feature of the k-th sample in the subsequence; n is the number of samples in the parent sequence and the subsequence; m is the number of subsequence features; ρ is the resolution coefficient, 0<ρ<1; (4) Calculate the degree of association; The formula of correlation degree ri is as follows: In the formula, q i (k) is the correlation coefficient, n is the number of characteristic sequence samples; (5) Determine the indicator weights; Weight coefficient w i The formula is as follows: (6) Determine the weight coefficient; According to the grey correlation analysis method, the weight coefficients of each index relative to the energy consumption of the circulating water pump are determined as follows: generator power 0.1448, low-pressure condenser pressure 0.1419, circulating water inlet temperature 0.1549, condensate temperature 0.1331, low-pressure cylinder exhaust temperature 0.1419, main steam temperature 0.1381, circulating water outlet temperature 0.1454; S32, CRITIC method weight analysis: The CRITIC method is to comprehensively measure the objective weight of the indicators based on the comparative strength of the evaluation indicators and the conflict between the indicators; the comparative strength of the indicators is measured by the standard deviation, and the conflict between the indicators is measured by the correlation coefficient; (1) Determine the analysis series; The key operating parameters that affect the power consumption of the circulating water pump are used as the characteristic sequence for CRITIC method analysis; (2) Variables are dimensionless; (3) Calculate the standard deviation; The standard deviation reflects the amount of information contained in the indicator. The larger the standard deviation, the more information it contains and the higher the weight of the corresponding indicator. The formula for the standard deviation Sj is as follows: In the formula, S j represents the standard deviation of the jth indicator, x ij represents the observed value of the i-th sample under the j-th index, represents the mean of the jth indicator, and m is the total number of samples; (4) Calculate the correlation coefficient; The conflict between indicators is measured by the correlation coefficient, which reflects the size of the correlation between indicators. The larger the correlation coefficient, the smaller the conflict and the smaller the weight of the corresponding indicator. The correlation coefficient formula is as follows: In the formula, σ jk represents the correlation coefficient between the jth indicator and the kth indicator, A j is the conflict of the jth indicator, j∈[1,n], k∈[1,n], x ij represents the observed value of the i-th sample on the j-th index, represents the mean value of the jth indicator, represents the mean of the kth index, x ik Represents the observed value of the i-th sample on the k-th indicator; (5) Determine the amount of information Qj; The amount of information Qj is obtained by the comparison strength and conflict of the indicators. The larger the Qj, the greater the role of the jth indicator in the entire evaluation indicator system. The formula for the amount of information Qj is as follows: Qj=SjAj; (6) Determine the indicator weight w j ; Weight coefficient w j The formula is as follows: (7) Determine the weight coefficient; According to the CRITIC method, the weight coefficient of each indicator relative to the energy consumption of the circulating water pump is determined: Generator power 0.3889, low-pressure condenser pressure 0.0863, circulating water inlet temperature 0.2101, condensate temperature 0.0609, low-pressure cylinder exhaust temperature 0.0596, main steam temperature 0.1320, circulating water outlet temperature 0.0623; S33, weighted combination: The weight coefficient determined by the grey relational analysis method and the weight coefficient determined by the CRITIC method are weighted together to obtain the combined weight. The weighted formula is as follows: In the formula, wk represents the combined weight coefficient, wi is the weight coefficient obtained by the grey relational analysis method, and wj is the weight coefficient obtained by the CRITIC method; According to the weighted formula, the combined evaluation coefficient is: Generator power 0.3954, low-pressure condenser pressure 0.086, circulating water inlet temperature 0.2885, condensate temperature 0.057, low-pressure cylinder exhaust temperature 0.059, main steam temperature 0.2, circulating water outlet temperature 0.063; S34, obtaining evaluation function: The evaluation function between each operating parameter and energy consumption based on the evaluation weight is: IF=0.2885x1+0.086x2+0.3954x3+0.057x4+0.059x5+0.2x6+0.063x7, In the formula, x1 is the circulating water inlet temperature, x2 is the low-pressure condenser pressure, x3 is the generator power, x4 is the condensate temperature, x5 is the low-pressure cylinder exhaust temperature, x6 is the main steam temperature, and x7 is the circulating water outlet temperature; S35, Optimizing working conditions: The evaluation function is used to automatically optimize the historical operating data of different combined operating conditions, calculate the difference between the evaluation index value and the actual operating value, and take the data with the largest difference in a certain combined clustering condition as the optimal operation record in the historical operation process to construct the optimal operating condition database of the unit's historical operation.

8. A data-driven coal-fired unit circulating water pump operation optimization method according to claim 1, characterized in that: The specific steps of step S4 are as follows: S41. Establishing an optimization model: Based on the data of the historical optimal operating condition database of the unit, a Bayesian classifier is used to establish the circulating water pump operation optimization model. The key operating parameters that affect the power consumption of the circulating water pump are used as the input parameters of the Bayesian classifier, and the decimal-coded classification labels of the two circulating water pumps are used as output. S42, model training and learning: The data of the historical optimal operating condition database of the unit is divided into a training set and a test set in a ratio of 8:

2. The circulating water pump operation optimization model established by the Bayesian classifier is used to learn the historical optimal operating condition database of the unit, and the optimal circulating pump combination corresponding to the optimal operating condition database of the unit is obtained. S43. Optimize model output: By real-time monitoring of the unit's operating data, when new operating conditions appear in the system, a Bayesian classifier is used to establish a circulating water pump operation optimization model. Based on previously learned knowledge, it can predict and output the recommended combination of the two circulating water pumps and the recommended circulating pump operating current, thereby achieving the operation optimization of the circulating water pump of the coal-fired unit.

Images