350MW thermal power generating unit cold end system optimization method based on big data

Abstract

The invention discloses a 350MW thermal power generating unit cold end system optimization method based on big data, and relates to the field of electric power. The method comprises the following steps: S1, data collection: selecting historical data during the operation period of a unit as modeling data; s2, data preprocessing: preprocessing historical data; s3, constructing a model: constructing a BP neural network model based on the preprocessed data, and optimizing a weight and a threshold value of the BP model through a genetic algorithm and a particle swarm algorithm; s4, calculating the optimal vacuum degree: calculating the optimal vacuum degree under different working conditions according to the optimized BP neural network model; and S5, cold end system control: sending the optimal vacuum degree and a circulating water pump frequency conversion instruction to a DCS (Distributed Control System) through an MODBUS communication protocol to realize optimal control of the cold end system. The method has the beneficial effects that the cold end system of the 350MW thermal power generating unit is optimized by adopting big data analysis and an intelligent algorithm, so that the comprehensive effects of improving the operation efficiency, reducing the energy consumption, saving the cost and reducing the emission are realized.

Description

Technical Field

[0001] The present invention belongs to the field of electric power, relates to the field of thermal power generating units, and specifically relates to a cold end system optimization method of a 350MW thermal power unit based on big data. Background Art

[0002] Circulating water pumps play a crucial role in the cold-end system of thermal power generators. Their primary function is to supply cooling water to the condenser, which condenses the steam discharged from the turbine. During this process, the steam condenses and its volume shrinks, creating a vacuum. The circulating water pump regulates the vacuum level in the condenser by controlling the cooling water flow rate. The vacuum level directly affects the turbine's efficiency: a higher vacuum reduces turbine backpressure, thereby improving thermal efficiency.

[0003] As a key auxiliary equipment with the highest energy consumption in thermal power plants, circulating water pumps consume approximately 1%-1.5% of generated electricity. With the current trend of power system reform, power plants moving towards market access, and grid-connected power through competitive bidding, reducing circulating water pump energy consumption and increasing condenser vacuum have become a conflicting issue. Therefore, optimizing the optimal vacuum and minimizing circulating water pump power for different operating conditions is crucial. Summary of the Invention

[0004] This invention addresses the technical problem of optimizing the cold-end system of a 350MW thermal power unit to improve operational efficiency, reduce energy consumption, and save costs. It provides a big data-based cold-end system optimization method for a 350MW thermal power unit. By employing big data analysis and intelligent algorithms to optimize the cold-end system of a 350MW thermal power unit, the combined effects of improved operational efficiency, reduced energy consumption, cost savings, and reduced emissions are achieved.

[0005] The technical solution adopted by the present invention is to provide a method for optimizing the cold end system of a 350MW thermal power unit based on big data, comprising the following steps:

[0006] S1. Data collection: The historical data of unit load, circulating water inlet temperature, condenser vacuum, circulating water pump power, circulating water pump current, circulating water pump voltage, main steam pressure, main steam temperature, reheat steam temperature, main steam flow, and heat network extraction valve oil motor stroke during unit operation are selected as modeling data;

[0007] S2. Data preprocessing: Preprocess historical data, including data calculation, steady-state condition judgment, outlier processing and data normalization;

[0008] S3. Model building: Build a BP neural network model based on the preprocessed data, and optimize the weights and thresholds of the BP model through genetic algorithm and particle swarm algorithm;

[0009] S4. Calculate the optimal vacuum degree: Calculate the optimal vacuum degree under different working conditions based on the optimized BP neural network model and generate a frequency conversion instruction for the circulating water pump;

[0010] S5. Cold end system control: The optimal vacuum degree and circulating water pump frequency conversion instructions are sent to the DCS system via the MODBUS communication protocol to achieve optimal control of the cold end system.

[0011] Further optimize this technical solution, the data calculation includes the calculation of the circulating water pump power and the calculation of the condenser vacuum; the circulating water pump power P xb After determining the power factor k, voltage U, and current I of the circulating water pump, use the formula Calculation is performed; the condenser vacuum calculation is to select the average of the three measuring points of the condenser vacuum during the operation of the unit as the modeling data of the condenser vacuum.

[0012] To further optimize this technical solution, the steady-state operating condition is judged by whether the difference between the maximum and minimum values ​​of the five variables, namely, unit load, main steam pressure, main steam temperature, reheat steam temperature, and main steam flow, meets the preset threshold within 5 minutes when the unit load exceeds 150MW. If it meets the threshold, it is determined that the unit operating state during this period is stable. If it does not meet the threshold, the operating condition data is eliminated. When the oil motor stroke of the heat network extraction valve is greater than 2%, it is determined to be a heating condition and the operating condition data is eliminated.

[0013] To further optimize this technical solution, the outlier processing is checked and corrected using the quartile method. A set of data is divided into 4 groups in order of size. The data dividing the 4 groups are called Q1, Q2, and Q3. One-quarter of all the data is less than Q1, one-quarter of the data is greater than Q3, and the median of the data is Q2. The difference between Q1 and Q3 is called the interquartile range IQR. The threshold of the data outlier is determined by the interval IQR. The lower threshold D1 = Q1-1.5IQR, and the upper threshold Du = Q3+1.5IQR. All data outside the range of the lower threshold D1 to the upper threshold Du are regarded as outliers. If the value is lower than the lower threshold, the lower threshold D1 is used to replace the value. If the value is greater than the upper threshold Du, the upper threshold Du is used to replace the value for data correction.

[0014] To further optimize this technical solution, data normalization processing is to convert the original data x of each parameter index into normalized data x*, and the calculation method is Then the normalized data y output by the neural network is converted into dimensionless denormalized data y*, which is calculated as

[0015] y * =y×(x max -x min )+xmin .

[0016] To further optimize this technical solution, the number of input layer nodes of the BP neural network model is 3, namely unit power, circulating water pump power, and circulating water inlet temperature; the number of output layer nodes is 1, which is condenser vacuum; the number of hidden layer nodes is determined to be 5 by trial and error, and the training parameters include the number of iterations is 100, the learning rate is 0.1, and the target error is 0.0001.

[0017] To further optimize this technical solution, the genetic algorithm is a 3-layer neural network with 3 nodes in the input layer, 1 node in the output layer, and 5 nodes in the hidden layer. The population size is set to 50, the crossover probability is 0.2, the mutation probability is 0.1, and the iteration is 100 times.

[0018] To further optimize this technical solution, the particle swarm algorithm sets the particle population size to 50, the number of iterations to 100, and the upper and lower limits of speed to ±1, and the upper and lower limits of displacement to ±5.

[0019] To further optimize this technical solution, the optimal vacuum calculation is to establish a model function P = f (F, W, T), and calculate the partial derivatives of the group power F and the circulating water pump power W with respect to the condenser vacuum P, which are recorded as F' and W' respectively. When F′ and W′ are equal, the corresponding vacuum value is the optimal vacuum degree under the current working conditions.

[0020] To further optimize this technical solution, the cold-end system control also includes judging the quality of the received data and the optimal vacuum instruction. Once the data and instruction quality are abnormal, the cold-end system remote control is immediately cut off and switched to the circulating water pump variable frequency control system in the original DCS. The operating status of the system is detected by the heartbeat signal. If the heartbeat signal is not detected, the system is immediately switched to the original DCS system.

[0021] The beneficial effects of the present invention are as follows: by adopting the BP neural network to establish the unit load, circulating water inlet temperature-vacuum model, and using the particle swarm algorithm and genetic algorithm to optimize the weights and thresholds, the best modeling scheme is selected by comparing the optimization effects, thereby improving the efficiency of the unit cold end system and achieving the comprehensive effects of improving operating efficiency, reducing energy consumption, saving costs and reducing emissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A control schematic diagram for optimizing the cold end system according to an embodiment of the present invention;

[0023] Figure 2 This is a steady-state operating condition judgment threshold diagram according to an embodiment of the present invention;

[0024] Figure 3is a graph showing the relationship between the mean square error of the error and the number of hidden layer nodes according to an embodiment of the present invention;

[0025] Figure 4 BP neural network optimal network weight and threshold diagram according to an embodiment of the present invention;

[0026] Figure 5 This is a BP neural network prediction output diagram according to an embodiment of the present invention;

[0027] Figure 6 This is a BP neural network prediction error diagram according to an embodiment of the present invention;

[0028] Figure 7 This is a graph showing the optimal fitness change of the genetic algorithm according to an embodiment of the present invention;

[0029] Figure 8 The genetic algorithm of the embodiment of the present invention optimizes the neural network weights and threshold values;

[0030] Figure 9 This is a genetic algorithm prediction output graph according to an embodiment of the present invention;

[0031] Figure 10 A genetic algorithm prediction error graph according to an embodiment of the present invention;

[0032] Figure 11 This is a graph showing the optimal fitness change of the particle swarm algorithm according to an embodiment of the present invention;

[0033] Figure 12 The particle swarm algorithm of the embodiment of the present invention optimizes the neural network weights and threshold values;

[0034] Figure 13 This is a prediction output graph of the particle swarm algorithm according to an embodiment of the present invention;

[0035] Figure 14 This is a prediction error graph of the particle swarm algorithm according to an embodiment of the present invention;

[0036] Figure 15 Schematic diagram of the optimal vacuum value of a single pump working condition according to an embodiment of the present invention;

[0037] Figure 16 Schematic diagram of the optimal vacuum value of the dual-pump working condition according to an embodiment of the present invention;

[0038] Figure 17 Optimal vacuum curve diagram for single pump working condition according to an embodiment of the present invention;

[0039] Figure 18 This is an optimal vacuum curve diagram for dual pump working conditions according to an embodiment of the present invention;

[0040] Figure 19 A schematic diagram of vacuum setting values ​​for a single pump operating condition according to an embodiment of the present invention;

[0041] Figure 20 Schematic diagram of vacuum setting values ​​for dual-pump working conditions according to an embodiment of the present invention;

[0042] Figure 21 1 is a comparison chart of the optimal vacuum setting value and the current vacuum setting value according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] like Figure 1 As shown in FIG, a method for optimizing the cold end system of a 350MW thermal power unit based on big data includes the following steps:

[0045] S1. Data collection: The historical data of unit load, circulating water inlet temperature, condenser vacuum, circulating water pump power, circulating water pump current, circulating water pump voltage, main steam pressure, main steam temperature, reheat steam temperature, main steam flow, and heat network extraction valve oil motor stroke during unit operation are selected as modeling data;

[0046] S2. Data preprocessing: Preprocess historical data, including data calculation, steady-state condition judgment, outlier processing and data normalization;

[0047] S3. Model building: Build a BP neural network model based on the preprocessed data, and optimize the weights and thresholds of the BP model through genetic algorithm and particle swarm algorithm;

[0048] S4. Calculate the optimal vacuum degree: Calculate the optimal vacuum degree under different working conditions based on the optimized BP neural network model and generate a frequency conversion instruction for the circulating water pump;

[0049] S5. Cold end system control: The optimal vacuum degree and circulating water pump frequency conversion instructions are sent to the DCS system via the MODBUS communication protocol to achieve optimal control of the cold end system.

[0050] During data collection, historical data such as unit load, condenser vacuum, circulating water inlet temperature, circulating water pump power, unit load, main steam pressure, main steam temperature, reheat steam temperature, main steam flow, and heat network extraction valve oil motor stroke during unit operation are collected from the DCS system as modeling data. Due to the uncertainty of historical data, some abnormal data will interfere with the neural network, which will affect the training effect of the neural network and reduce the accuracy of the model. Therefore, effective data preprocessing is required;

[0051] Data preprocessing includes data calculation, steady-state condition judgment, outlier processing and data normalization;

[0052] Data calculation: Calculate the power of the circulating water pump using the formula: Where, P xb is the circulating water pump power (in kW), k is the power factor, U is the circulating water pump voltage (in kV), and I is the circulating water pump current (in A). At the same time, the average of the three measurement points of the unit condenser vacuum is taken as the modeling data. The formula is, where P is the condenser vacuum (in kPa)

[0053] The mean of the two inlet pipe temperatures of the circulating water inlet temperature is taken as the modeling data;

[0054] Steady-state operating condition judgment: The data extracted by DCS includes operating conditions such as shutdown, deep peak regulation, and load increase and decrease. When the unit is in non-design operating conditions or dynamic operating conditions, its thermodynamic parameters and system characteristics will change significantly. Therefore, it is necessary to extract the stable operating status of the unit from the historical operation database (when the load exceeds 150MW, if within 5 minutes, the difference between the maximum and minimum values ​​of the five variables of unit load, main steam pressure, main steam temperature, reheat steam temperature, and main steam flow rate meets the preset threshold, see the threshold Figure 2 ), if it meets the threshold, the unit operation status during this period is considered stable; if it does not meet the threshold, the operating condition data is eliminated;

[0055] On the other hand, in winter, the circulating water inlet temperature is very low, requiring only a low circulating water flow rate to maintain a high condenser vacuum. During this period, the circulating water pump frequency conversion is permanently at its lowest frequency. Therefore, optimizing the unit's cold-end system during the heating period is of little production significance. Therefore, during data preprocessing, data from this period was excluded, and only analysis was performed for pure condensing conditions. (When the oil motor stroke of the heating network extraction valve exceeds 2%, this is considered heating operation and the data is excluded.)

[0056] Outlier processing: Outliers refer to a small number of data in a sample that are significantly different from other monitored values. The data extracted from steady-state conditions are checked and corrected using the "quartile method". By using the quartile method, a set of data is divided into 4 groups in order of size, and the proportion of each group is one-fourth of the whole. The data dividing the 4 groups are called Q1, Q2, and Q3. One-fourth of all the data is less than Q1, one-fourth of the data is greater than Q3, and the median of the data is Q2. The difference between Q1 and Q3 is called the interquartile range IQR, and the formula is IQR=Q3-Q1. The threshold of the data outlier can be determined by the interval IQR. The lower threshold D1=Q1-1.5IQR, and the upper threshold Du=Q3+1.5IQR. Data exceeding the threshold will be treated as an outlier. The steady-state condenser vacuum, circulating water pump power, and circulating water inlet temperature data are checked and corrected for outliers. If the data is an outlier, if the outlier is lower than D1, D1 is used to replace the value. If the outlier is greater than Du, Du is used to replace the value. In this way, the data is corrected.

[0057] Data normalization: During the modeling process, due to the differences in the dimensions and dimensional units of various parameter indicators, it will have a negative impact on data analysis. To eliminate the dimension difference, where x max is the maximum value of the original data, x min is the minimum value of the original data, and x * After the data is normalized, the output data of the neural network is also dimensionless data, and its range is between [0,1]. Therefore, it is necessary to use the denormalization method to convert the output data into dimensional data. The denormalization formula is y * =y×(x max -x min )+x min , where y is the neural network output sequence data, y * By strictly regulating the dimensions of the input data after denormalization, the error caused by the difference in dimensions can be reduced, thereby improving the accuracy of the prediction model and significantly enhancing its convergence ability and precision.

[0058] After data preprocessing, a relationship model between circulating water pump power and condenser vacuum is constructed based on the preprocessed data. The model adopts BP neural network, with 3 input layer nodes (corresponding to unit power, circulating water pump power, and circulating water inlet temperature) and 1 output layer node (corresponding to condenser vacuum). Since the mapping relationship between input and output is not very complicated, it is set as a single hidden layer, that is, a three-layer neural network. The number of hidden layer nodes is calculated by the empirical formula Calculation, where m is the number of hidden layer nodes, n is the number of BP neural network input layer nodes, where a is between 1 and 10, the number of hidden layer nodes is determined by trial and error, and the approximate range of the number of hidden layer nodes is estimated based on the empirical method. Then, the effects of neural network systems with different numbers of hidden layer nodes in the same sample training set on network training and learning generalization are compared, that is, the number of hidden layer nodes corresponding to the minimum model error is selected, the number of iterations is set to 100, the learning rate is 0.1, and the target is 0.0001. After testing, according to Figure 3 , it can be seen that with the increase of the number of hidden layer nodes, the mean square error of the error first decreases and then increases again. When the number of hidden layer nodes is too small, the neural network cannot fit the intrinsic relationship between the output and the output. When the number of hidden layer nodes is too large, it will also cause overfitting of the noise in the training data, resulting in "overfitting" phenomenon (the manifestation of overfitting is that the error of the modeling training data is very small, but the error of the test data is larger). When the number of hidden layer nodes is 5, the root mean square error is the smallest, and the fitting effect for the data is the best. According to Figure 4 When the number of hidden layer nodes is 5, the trained neural network weights and thresholds are obtained, and the test data are predicted using the neural network. The condenser vacuum prediction results and prediction errors are shown in the figure below. Figure 5 、 Figure 6 As shown in the figure, in the condenser vacuum prediction results, the prediction simulation fitting effect of the BP neural network model is relatively good. From the prediction error, it can be seen that the condenser vacuum prediction error value of the BP neural network is between -0.15 and 0.15 kPa, and the change value is not large and relatively stable. However, there are also individual points where the prediction error exceeds 0.2 kPa. The mean square error between the neural network prediction output and the expected output is 0.0079.

[0059] In order to improve the accuracy of the model, a genetic algorithm was used to optimize the modeling. The genetic algorithm set the population size to 50, the crossover probability to 0.2, the mutation probability to 0.1, and iterated 100 times. The mean of the variance of the error between the predicted output and the expected output was used as the individual fitness. Since the smaller the individual fitness, the better the individual, the reciprocal of the fitness value was calculated before individual selection to ensure the probability of individuals with smaller fitness being selected. MATLAB programming training was used to realize the establishment and prediction of the model, and the prediction results of the condenser vacuum using the GA (genetic algorithm)-BP circulating water pump power-vacuum model were obtained. Figure 7It can be seen that the fitness tends to be stable (about 0.03424) after the network has just iterated 24 times, and remains unchanged until it iterates 47 times. After multiple rounds of selection competition, the gene types in the population tend to be stable, and no better individuals can be produced through crossover operations. After that, the best fitness has decreased several times. This is because the mutation operation generates better genes. If the mutation probability is set too small, the gene diversity in the population will be poor. After multiple rounds of competition, there are too few gene types, and no matter how crossover is performed, no better genes will be produced. If the mutation probability is too large, it may cause damage to the genes. Although the breadth of the search has increased, it is difficult to converge to the local optimal solution. Taking the optimal fitness value of 0.03361 as an example, the network initial weight and threshold of the corresponding optimal individual can be found in Figure 8 , assign the initial weight and threshold of the optimal individual to the model, and use the training samples to train the condenser vacuum prediction results and prediction errors as shown in the following figure: Figure 9 、 Figure 10 As shown in the figure, it can be concluded that the change trend of the predicted value of the condenser vacuum by the GA-BP circulating water pump power-vacuum model is basically consistent with the trend of the true value. It can be seen that its fitting effect is relatively good, indicating that the condenser vacuum prediction result of the GA-BP circulating water pump power-vacuum model is relatively accurate. According to the error change trend of the condenser vacuum prediction by the GA-BP circulating water pump power-vacuum model in the figure, it can be seen that the error values ​​of the condenser vacuum prediction by the model are mostly between -0.1 and 0.1 kPa, with a small and stable change range. The mean square error of the error between the neural network prediction output and the expected output is 0.0070.

[0060] Similar to the genetic algorithm, the particle swarm algorithm sets the particle population size to 50 and the number of iterations to 100. In order to prevent the velocity component and particle displacement from being too large, the upper and lower limits of the velocity are set to ±1, and the upper and lower limits of the displacement are set to ±5. The mean variance of the error between the predicted output and the expected output is used as the fitness S of the individual particle. Since the smaller the individual fitness, the better the individual, the position with the smallest particle fitness is recorded as the individual optimal solution in each iteration. At the same time, the position with the lowest fitness among all particles is recorded as the global optimal solution to ensure that the particle velocity points to the direction with the lowest fitness. MATLAB programming training is used to realize the establishment and prediction of the model, and the prediction results of the PSO (particle swarm algorithm)-BP circulating water pump power-vacuum model for condenser vacuum are obtained. See Figure 11 , we can see that the global optimal solution fitness tends to be stable (about 0.04481) after the network starts to iterate 52 times, and remains basically unchanged until the termination of iteration 100. From the results, we can see that the neural network optimized by particle swarm optimization can jump out of the local minimum point, has a strong generalization ability, and can obtain better results. Taking the optimal fitness value of 0.04481 as an example, the initial network weight and threshold of the corresponding optimal individual can be found in Figure 12 Assign the initial weights and thresholds of the optimal individual to the model. After training with the training samples, the predicted results and prediction errors of the condenser vacuum are as follows Figure 13 、 Figure 14 As shown, it can be seen that the changing trend of the predicted values of the condenser vacuum by the PSO-BP circulating water pump power-vacuum model is basically the same as that of the true values, with a good fitting effect, indicating that the predicted results of the condenser vacuum by the PSO-BP circulating water pump power-vacuum model are relatively accurate. According to the changing trend of the prediction errors of the condenser vacuum by the PSO-BP circulating water pump power-vacuum model in the figure, it can be seen that most of the prediction error values of the model for the condenser vacuum are between -0.1 and 0.1 kPa, but there are many points with large error values, and the change range is not stable enough. The mean square error between the neural network prediction output and the expected output is 0.0074;

[0061] In addition, considering the differences in data distribution under different working conditions, by dividing different working conditions, the function is piecewise fitted, and typical load data are selected, namely 175, 210, 245, 280, 315, 350 MW. ±35 MW near each load is used as the data for that load. Under each load, the working conditions are divided according to the circulating water inlet temperature. For the single-pump working condition, the circulating water inlet temperatures of 10 °C, 15 °C, 20 °C, and 25 °C are selected; for the double-pump working condition, the circulating water temperatures of 18 °C, 20 °C, 25 °C, 30 °C, and 35 °C are selected. ±5 °C near each inlet temperature is used as the data for that temperature. The circulating water pump power-vacuum relationship models are established respectively by using the BP neural network algorithm, the neural network optimized by the particle swarm algorithm, and the neural network optimized by the genetic algorithm, and the corresponding optimal vacuum values can be obtained;

[0062] The function formula of the optimal vacuum model is P = f(F, W, T). Through the functional relationship between the unit power (F), the circulating water inlet temperature (T), the circulating water pump power (W), and the condenser vacuum (P), the partial derivatives of the unit power and the circulating water pump power with respect to the vacuum are obtained. The formula is to judge the optimal vacuum. If F′ > W′, the current vacuum degree is lower than the optimal vacuum. Increasing the circulating water flow can improve the net power of the unit. On the contrary, if F′ < W′, the current vacuum degree is higher than the optimal vacuum. Increasing the circulating water flow will instead cause a decrease in the net power of the unit. Therefore, the vacuum when F′ = W′ is the optimal vacuum under the current working condition. In practical applications, F′ and W′ are not easy to calculate. The condenser vacuum increment ΔP1 corresponding to the increase of the circulating water pump power by ΔW at each working condition point, and the condenser vacuum increment ΔP2 required for the increase of the load by ΔF can be used instead. The vacuum when ΔP1 = ΔP2 is the optimal vacuum under the current working condition;

[0063] Using the BP neural network algorithm to establish the circulating water pump power-condenser vacuum relationship model under different loads and ambient temperatures can obtain the corresponding optimal vacuum value, such as Figure 15 、 Figure 16 As shown, and by plotting the relationship between the optimal vacuum and ambient temperature under different loads, as Figure 17 、 Figure 18 As shown, it can be concluded that when the unit load is low, only a small amount of circulating water flow is needed to achieve a higher vacuum degree. At this time, the optimal vacuum degree is higher. As the unit load increases, the circulating water pump power required to increase the circulation ratio m is higher, so the optimal vacuum degree is reduced. When the circulating water inlet temperature is low, the condenser vacuum quickly increases to about -97.5kPa. At this time, the unit back pressure is about 3.8kPa.a, which is better than the design back pressure of 4.9kPa.a. At this time, the condenser vacuum degree has been increased to a very high level. The contribution of increasing the circulating water flow rate to the unit load can be ignored. Therefore, when the circulating water inlet temperature is low, the optimal vacuum degree is higher. In summer, the circulating water inlet temperature is high, and can reach up to 35℃. At this time, the optimal vacuum degree can be reduced to about -92.5kPa.

[0064] The current vacuum setting value of the unit is as follows Figure 19 、 Figure 20 As shown, it is set based on the turbine manufacturer's instructions and operating experience, and the comparison between the optimal vacuum setting value and the current vacuum setting value is as follows: Figure 21 As shown in the figure, it can be found that when the circulating water inlet temperature is lower than 20℃ and a single circulating water pump is running, the current vacuum setting value is relatively conservative. The vacuum and the net output of the unit can be increased by increasing the circulating water pump flow rate;

[0065] When the circulating water inlet temperature is around 18°C ​​and two circulating water pumps are running, the current vacuum setting value is high and the circulating water flow is excessive. This is because the circulating water inlet temperature is low and the circulating water flow required by the unit is small. At this time, starting two circulating water pumps will cause the flow to be too high, and the unit's economic efficiency is poor. At this time, the net output of the unit should be increased by lowering the vacuum setting value or stopping one circulating water pump.

[0066] When the circulating water inlet temperature is around 35°C and the two circulating water pumps are running, the current vacuum setting value is high and the circulating water flow is excessive. This is because the circulating water inlet temperature is too high. At this time, the circulating water flow required to increase the vacuum is too large, and the increase in circulating water pump power is greater than the increase in unit load. Higher unit economic benefits can be obtained by lowering the vacuum setting value.

[0067] Based on the optimal vacuum degree predicted by the model, a circulating water pump frequency conversion instruction is generated. The optimal vacuum setting value and the circulating water pump frequency conversion instruction are sent to the DCS system via the MODBUS communication protocol to achieve optimized control of the cold end system. The optimized control scheme adopts an "external" control solution, using an industrial computer as the optimization control device. Data is exchanged with the DCS system via RS-485 serial communication and MODBUS protocol to ensure the reliability and security of data transmission;

[0068] In order to ensure the reliability and security of the optimization plan, the quality of the received data is judged. For example, if there are abnormal values ​​such as over-limit and over-rate, in normal operation, the remote control input permission signal output is '1'. Once any data abnormality is found, all output control instructions are immediately maintained, and the remote control input permission signal is immediately output to '0'. After receiving the signal, the DCS immediately cuts off the remote control of the cold end system and switches to the circulating water pump variable frequency control system in the original DCS. The cold end optimization system continuously sends heartbeat waves to the DCS to represent the operating status of the system. When the industrial computer crashes or the power is cut off, the DCS cannot detect the heartbeat wave and immediately cuts off the remote control of the cold end system and switches to the circulating water pump variable frequency control system in the original DCS. Water pump variable frequency automatic control system. At the same time, after the DCS side receives the optimal vacuum instruction of the cold end optimization system, the DCS conducts a quality judgment on the control instruction issued by the cold end optimization. If the instruction quality is abnormal, the remote control of the cold end system will be immediately cut off and switched to the circulating water pump variable frequency control system in the original DCS. The safety of the control system will be ensured under any circumstances. The original automatic cut-off logic of the circulating water pump variable frequency automatic control remains unchanged. When there is a large deviation between the vacuum set value and the actual value, a large deviation between the frequency conversion instruction and the feedback, the circulating water pump inlet temperature quality is poor, the condenser vacuum quality is poor, the unit power quality is poor, the inverter frequency feedback quality is poor, etc., the circulating water pump variable frequency control will be automatically switched to manual control, and the output frequency conversion instruction will remain unchanged at the current value.

Claims

1. A method for optimizing the cold-end system of a 350MW thermal power unit based on big data, characterized by: The steps include: S1. Data collection: The historical data of unit load, circulating water inlet temperature, condenser vacuum, circulating water pump power, circulating water pump current, circulating water pump voltage, main steam pressure, main steam temperature, reheat steam temperature, main steam flow, and heat network extraction valve oil motor stroke during unit operation are selected as modeling data; S2. Data preprocessing: Preprocess historical data, including data calculation, steady-state condition judgment, outlier processing and data normalization; S3. Model building: Build a BP neural network model based on the preprocessed data, and optimize the weights and thresholds of the BP model through genetic algorithm and particle swarm algorithm; S4. Calculate the optimal vacuum degree: Calculate the optimal vacuum degree under different working conditions based on the optimized BP neural network model and generate a frequency conversion instruction for the circulating water pump; S5. Cold end system control: Send the optimal vacuum degree and circulating water pump frequency conversion instructions to the DCS system through the MODBUS communication protocol to achieve optimized control of the cold end system.

2. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The data calculation of the data preprocessing in S2 includes using the formula Calculate the circulating water pump power and condenser vacuum; where k is the circulating water pump power factor, U is the circulating water pump voltage, I is the circulating water pump current, and the modeling data of the condenser vacuum is the average of the three measurement points of the condenser vacuum during the operation of the unit.

3. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The steady-state operating condition judgment of the data preprocessing in S2 is when the unit load exceeds 150MW. Within 5 minutes, the difference between the maximum and minimum values ​​of the five variables of unit load, main steam pressure, main steam temperature, reheat steam temperature, and main steam flow rate meets the preset threshold. If it meets the threshold, it is determined that the unit operating state during this period is stable. If it does not meet the threshold, the operating condition data is discarded. In addition, when the stroke of the oil motor of the heat network extraction valve is greater than 2%, it is determined to be a heating condition and the operating condition data is discarded.

4. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The outlier processing of the data preprocessing in S2 is checked and corrected by the quartile method. A set of data is divided into 4 groups in order of size. The data dividing the 4 groups are called Q1, Q2, and Q3. One-quarter of all the data are less than Q1, one-quarter of the data are greater than Q3, and the median of the data is Q2. The difference between Q1 and Q3 is called the interquartile range IQR. The threshold of the data outlier is determined by the interval IQR. The lower threshold D1 = Q1-1.5IQR, and the upper threshold Du = Q3+1.5IQR. All data outside the range from the lower threshold D1 to the upper threshold Du are regarded as outliers. If the value is lower than the lower threshold D1, the lower threshold D1 is used to replace the value. If the value is greater than the upper threshold Du, the upper threshold Du is used to replace the value for data correction.

5. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The data normalization process of the data preprocessing in S2 is to convert the original data x of each parameter index into normalized data x*, which is calculated as follows: Then the normalized data y output by the neural network is converted into dimensionless denormalized data y*, which is calculated as y * =y×(x max -x min )+x min , where x max is the maximum value of the original data, x min is the minimum value of the original data.

6. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The BP neural network model constructed in the S3 has 3 input layer nodes, namely, unit power, circulating water pump power, and circulating water inlet temperature; the output layer has 1 node, which is the condenser vacuum; the number of hidden layer nodes is determined to be 5 by trial and error, and the training parameters include 100 iterations, a learning rate of 0.1, and a target error of 0.0001.

7. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The genetic algorithm used to construct the model in S3 is a three-layer neural network with three nodes in the input layer, one node in the output layer, and five nodes in the hidden layer. The population size is set to 50, the crossover probability is set to 0.2, the mutation probability is set to 0.1, and the iteration is repeated 100 times.

8. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The particle swarm algorithm used to construct the model in S3 sets the particle swarm size to 50, the number of iterations to 100, and the upper and lower limits of velocity to ±1, and the upper and lower limits of displacement to ±5.

9. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The model function for calculating the optimal vacuum degree in S4 is P=f(F, W, T), where F is the unit power, W is the circulating water pump power, P is the condenser vacuum, and T is the condenser inlet water temperature. The optimal vacuum is determined by calculating the partial derivatives F′ and W′ of the unit power F and the circulating water pump power W with respect to the vacuum. When F′ and W′ are equal, the corresponding vacuum value is the optimal vacuum degree under the current working conditions.

10. The method for optimizing the cold end system of a 350MW thermal power unit based on big data according to claim 1, characterized in that: The cold end system control of S5 also includes judging the quality of the received data and the optimal vacuum instruction. Once the data and instruction quality are found to be abnormal, the cold end system remote control is immediately cut off and switched to the circulating water pump variable frequency control system in the original DCS; And the operating status of the system is detected through the heartbeat signal. Once the heartbeat signal is not detected, it will immediately switch to the original DCS system.

Images