Calibration method and device for multi-point flowmeter based on RBF neural network

Abstract

This application proposes a method and device for calibrating a multi-point flowmeter based on an RBF neural network. The method includes: using the primary fan frequency, the forced draft fan frequency, the induced draft fan frequency, and the unit load as the input of the RBF neural network, and using the flue gas volume correction coefficient as the output of the RBF neural network, to obtain an input and output data set covering multiple operating conditions of the thermal power unit; using the input and output data sets as training data sets to train a preset RBF neural network and adjust the parameters of the RBF neural network; obtaining real-time input and real-time flue gas flow measured by the multi-point flowmeter, inputting the real-time input into the trained RBF neural network to calculate the real-time flue gas volume correction coefficient; and multiplying the real-time flue gas flow by the real-time flue gas volume correction coefficient to obtain a corrected flue gas flow measurement value. This method improves the accuracy of the measurement results of the multi-point flowmeter by calibrating the original measurement results of the multi-point flowmeter.

Description

Technical Field

[0001] The present application relates to the technical field of flue gas monitoring, and in particular to a calibration method and device for a multi-point flow meter based on an RBF neural network. Background Art

[0002] Currently, thermal power generation remains one of the primary forms of power generation. The large number of thermal power plants results in significant pollutant emissions. To achieve energy conservation and emission reduction, pollutant emissions must be controlled. Online monitoring of various flue gas pollutants is an effective means of controlling these emissions. Flue gas pollutant emissions are generally calculated by multiplying pollutant concentration by flue gas flow rate. While flue gas pollutant concentration measurement technology is highly mature, flue gas flow rate monitoring still requires improvement.

[0003] In related technologies, multi-point flowmeters are widely used for flue gas flow measurement, and many thermal power plants have installed these. A multi-point flowmeter measures flow velocity at multiple points of equal cross-section across a large duct. Specifically, a multi-point flowmeter organically assembles multiple measuring points of equal cross-section, connecting the positive pressure side to the positive pressure side, and the negative pressure side to the negative pressure side. A common pressure lead pipe extends from each side, connecting to the positive and negative terminals of a differential pressure transmitter, respectively. The average flow velocity across the cross-section is measured, and the flue gas flow rate is then calculated.

[0004] However, in medium- and large-scale thermal power plants, the flue cross-sections are mostly rectangular, with some having circular cross-sections, and are generally arranged compactly. The flue system also contains numerous bends, reducers, manifolds, and throttling dampers. The flow field within the flue exhibits significant turbulence, with numerous separations and vortices. The flow uniformity within the flue is poor and varies with operating conditions. Therefore, the measurement results of the multi-point flow meter described above are significantly different from the actual situation, resulting in low measurement accuracy. Summary of the Invention

[0005] The present application aims to solve one of the technical problems in the related art at least to a certain extent.

[0006] To this end, the first purpose of this application is to propose a correction method for a multi-point flow meter based on an RBF neural network. This method only requires adding a correction link to the original multi-point flow meter equipment, and calculates the correction coefficient based on the RBF neural network for correction to obtain more accurate measurement results.

[0007] The second purpose of this application is to propose a calibration device for a multi-point flow meter based on an RBF neural network.

[0008] A third object of the present application is to provide a non-transitory computer-readable storage medium.

[0009] To achieve the above objectives, a first embodiment of the present application is to propose a calibration method for a multi-point flow meter based on an RBF neural network, the method comprising the following steps:

[0010] The primary fan frequency, the forced draft fan frequency, the induced draft fan frequency, and the unit load are used as inputs of the RBF neural network, and the flue gas volume correction coefficient is used as the output of the RBF neural network to obtain an input and output data set covering multiple operating conditions of the thermal power unit;

[0011] Using the input and output data sets as training data sets, training the preset RBF neural network, and adjusting the parameters of the RBF neural network;

[0012] Acquire real-time input and real-time flue gas flow measured by a multi-point flow meter, and input the real-time input into a trained RBF neural network to calculate a real-time flue gas flow correction coefficient;

[0013] The real-time flue gas flow rate is multiplied by the real-time flue gas volume correction coefficient to obtain a corrected flue gas flow rate measurement value.

[0014] Optionally, in one embodiment of the present application, an input and output data set covering multiple operating conditions of a thermal power unit is obtained, including: under the same operating condition, simultaneously measuring the flue gas flow rate by manual measurement and multi-point flow meter measurement; obtaining the measurement results of the manual measurement and the multi-point flow meter measurement under the current operating condition by calculating the average value within the measurement time; taking the measurement result of the manual measurement as the true value, calculating the ratio of the measurement result of the manual measurement to the measurement result of the multi-point flow meter measurement, taking the ratio as the output of the RBF neural network under the current operating condition, and obtaining the input of the RBF neural network under the current operating condition from the distributed control system DCS of the thermal power unit; switching to different operating conditions, and repeating the flue gas flow measurement comparison test and reading the input under each operating condition.

[0015] Optionally, in one embodiment of the present application, the RBF neural network includes an input layer, a hidden layer, and an output layer, and the output of i nodes in the hidden layer is expressed by the following formula:

[0016]

[0017] Among them, i is any node in the hidden layer, u i is the output of the i-th node in the hidden layer, is the input sample vector, The center vector of the Gaussian function, σ i is the normalization constant.

[0018] Optionally, in one embodiment of the present application, the output of the output layer is expressed by the following formula:

[0019]

[0020] Among them, y is the output of the RBF neural network, w i is the weight coefficient from the hidden layer to the output layer.

[0021] Optionally, in one embodiment of the present application, training the preset RBF neural network includes:

[0022] Optimize the objective function expressed in the following formula:

[0023]

[0024] Where N is the number of samples, t p is the expected value of output, y p is the actual value of the output;

[0025] The w is expressed by the following formula i The value of is learned:

[0026]

[0027] Where η is the learning rate, 0<η<1.

[0028] Optionally, in one embodiment of the present application, obtaining the real-time flue gas flow measured by the multi-point flow meter includes: establishing a communication connection with the flue gas online monitoring system CEMS of the thermal power unit, and reading the real-time flue gas flow measured by the multi-point flow meter from the online monitoring system CEMS.

[0029] To achieve the above objectives, the second embodiment of the present application further proposes a calibration device for a multi-point flow meter based on an RBF neural network, comprising the following modules:

[0030] an acquisition module, configured to use the primary fan frequency, the forced draft fan frequency, the induced draft fan frequency, and the unit load as inputs of an RBF neural network, and use the flue gas volume correction coefficient as the output of the RBF neural network, to acquire an input and output data set covering multiple operating conditions of the thermal power unit;

[0031] A training module, configured to train the preset RBF neural network using the input and output data sets as training data sets, and adjust the parameters of the RBF neural network;

[0032] A first calculation module is used to obtain a real-time input quantity and a real-time flue gas flow rate measured by a multi-point flow meter, and input the real-time input quantity into a trained RBF neural network to calculate a real-time flue gas flow correction coefficient;

[0033] The second calculation module is used to multiply the real-time flue gas flow rate by the real-time flue gas volume correction coefficient to obtain a corrected flue gas flow rate measurement value.

[0034] Optionally, in one embodiment of the present application, the acquisition module is specifically used to: under the same operating condition, simultaneously measure the flue gas flow rate by manual measurement and multi-point flow meter measurement; obtain the measurement results of the manual measurement and the multi-point flow meter measurement under the current operating condition by calculating the average value within the measurement time; take the measurement result of the manual measurement as the true value, calculate the ratio of the measurement result of the manual measurement to the measurement result of the multi-point flow meter measurement, take the ratio as the output of the RBF neural network under the current operating condition, and obtain the input of the RBF neural network under the current operating condition from the distributed control system DCS of the thermal power unit; switch to different operating conditions, and repeat the flue gas flow measurement comparison test and read the input under each operating condition.

[0035] Optionally, in one embodiment of the present application, the first calculation module is specifically used to: establish a communication connection with the flue gas online monitoring system CEMS of the thermal power unit, and read the real-time flue gas flow measured by the multi-point flow meter from the online monitoring system CEMS.

[0036] In order to implement the above embodiments, the third aspect of the present application also proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the computer program is executed by the processor, the correction method of the multi-point flow meter based on the RBF neural network in the above embodiments is implemented.

[0037] The technical solutions provided by the embodiments of this application provide at least the following beneficial effects: This application upgrades existing multi-point flowmeters by adding a calibration step. A trained RBF neural network is used to calculate the flowmeter's calibration coefficient under current operating conditions. The detected flue gas flow rate is then corrected using the calibration coefficient calculated in real time, effectively improving the measurement accuracy of the multi-point flowmeter and enhancing the accuracy and reliability of the multi-point flowmeter's measurement results. Furthermore, this application, based on the existing thermal power unit system, reduces the workload for retrofitting and upgrading, simplifies the retrofit solution, reduces costs, and facilitates implementation.

[0038] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0040] Figure 1 A flowchart of a calibration method for a multi-point flow meter based on an RBF neural network proposed in an embodiment of the present application;

[0041] Figure 2 A flowchart of a method for obtaining input and output data sets proposed in an embodiment of the present application;

[0042] Figure 3 This is the structural intent of an RBF neural network model proposed in the embodiment of this application;

[0043] Figure 4 A schematic structural diagram of a multi-point flowmeter calibration device based on an RBF neural network proposed in an embodiment of the present application;

[0044] Figure 5 This is a schematic structural diagram of a specific RBF neural network-based multi-point flowmeter calibration device proposed in an embodiment of the present application. DETAILED DESCRIPTION

[0045] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0046] A method and apparatus for calibrating a multi-point flowmeter based on an RBF neural network according to an embodiment of the present invention will be described in detail below with reference to the accompanying drawings.

[0047] Figure 1 This is a flow chart of a calibration method for a multi-point flow meter based on an RBF neural network proposed in an embodiment of the present application, as shown in FIG. Figure 1 As shown, the method includes the following steps:

[0048] In step S101, the primary fan frequency, the forced draft fan frequency, the induced draft fan frequency and the unit load are used as inputs of the RBF neural network, and the flue gas volume correction coefficient is used as the output of the RBF neural network to obtain an input and output data set covering multiple operating conditions of the thermal power unit.

[0049] A radial basis function neural network (RBF) is a three-layer feedforward network that implements functions such as function approximation and classification. This application uses an RBF neural network model to calculate a flue gas volume correction coefficient to calibrate the flue gas flow measured by a multi-point flow meter. By leveraging the RBF neural network's simple structure, fast learning speed, and excellent approximation performance, the calculation ensures the accuracy of the output flue gas volume correction coefficient and reduces the calibration workload.

[0050] Among them, the input and output data set covering multiple operating conditions of the thermal power unit means that under each operating condition of the thermal power unit, the RBF neural network has corresponding input and output quantities, and this data set includes the input and output quantities under most operating conditions when the thermal power unit is operating normally.

[0051] It should be noted that since this application uses an RBF neural network to correct the flue gas flow rate, and the flue gas flow rate is primarily related to factors such as unit load, primary fan air volume, supply fan air volume, and induced draft fan air volume, the aforementioned parameters are used as inputs to the prediction model for correction. However, since the actual measurement of primary fan air volume, supply fan air volume, and induced draft fan air volume often results in large errors and frequent fluctuations, this application selects the primary fan frequency, supply fan frequency, induced draft fan frequency, and unit load as the inputs to the RBF neural network model. The output of the model is the flue gas volume correction coefficient calculated based on the inputs.

[0052] The primary fan, forced draft fan, and induced draft fan are all fans that control air volume during the operation of a thermal power unit. The primary fan is the fan that provides primary air at a certain pressure and flow rate in the system. The frequency of each fan is a variation of the corresponding drive motor frequency, and each fan frequency is positively correlated with the fan speed.

[0053] Specifically, when flue gas flow correction is performed through an RBF neural network, the parameters of each node in the RBF neural network model need to be trained and obtained through relevant learning algorithms based on a large amount of training data. Therefore, the present application first obtains an input and output data set including data under multiple working conditions, and the input and output data set includes multiple groups of corresponding input quantities and output quantities.

[0054] In one embodiment of the present application, in order to more clearly illustrate the specific implementation process of obtaining input and output data sets of the present application, a method for obtaining input and output data sets under multiple operating conditions of a unit proposed in this embodiment is exemplified below. Figure 2 This is a flow chart of a method for obtaining an input and output data set proposed in an embodiment of the present application, such as Figure 2 As shown, the method includes the following steps:

[0055] Step S201 : Under the same working conditions, the flue gas flow is measured simultaneously by manual measurement and multi-point flow meter measurement.

[0056] Specifically, the input quantity under a specific operating condition can be directly read from the unit system. However, to determine the output quantity under a specific operating condition, a comparison test of manual flue gas flow measurement and online measurement using a multi-point flow meter is first required. That is, the flue gas flow rate is measured simultaneously through manual measurement and online measurement using a multi-point flow meter under the same stable operating conditions. In specific implementation, multi-point flow meter measurement can adopt the measurement method used in related technologies, measuring the average value of multiple measuring points at equal cross-sections. Manual measurement can be performed at each measuring point in turn using a manual measurement device.

[0057] Step S202 , obtaining the measurement results of manual measurement and multi-point flow meter measurement under the current working conditions by calculating the average value within the measurement time.

[0058] Specifically, because manual measurement is slower than multi-point flow meter measurement and the methods for obtaining measurement results are also different, to ensure consistency in the basis for subsequent comparisons, the results of manual measurement and multi-point flow meter online measurement at each operating point in this application are averaged over the measurement time. That is, for each measurement method, the measured values ​​obtained within the measurement time are added together and divided by the measurement time.

[0059] In step S203, the manual measurement result is taken as the true value, the ratio of the manual measurement result to the measurement result of the multi-point flow meter is calculated, the ratio is taken as the output of the RBF neural network under the current working condition, and the input of the RBF neural network under the current working condition is obtained from the distributed control system DCS of the thermal power unit.

[0060] Specifically, the manually measured flue gas flow rate is used as the true value, that is, the actual flue gas flow rate under the current operating conditions. The manually measured result is then divided by the measurement result of the multi-point flow meter, and the two are compared to obtain the output of the RBF neural network model under that operating point, namely the flue gas volume correction coefficient. It can be understood that the actual flow rate measured by the multi-point flow meter under that operating condition is multiplied by the flue gas volume correction coefficient to obtain the actual flue gas flow output by the unit under that operating condition, so that the corrected multi-point flow meter measurement value is equal to the accurate actual value. This flue gas volume correction coefficient is the target flue gas volume correction coefficient that the RBF neural network should output under that operating condition, and this output can be used for subsequent model training.

[0061] The four input quantities—primary fan frequency, forced draft fan frequency, induced draft fan frequency, and unit load—can be directly read from the thermal power unit system based on current conditions, with the input quantities determined based on the values ​​actually recorded by the system. For example, they can be read from the thermal power unit's pre-installed distributed control system (DCS), with the input quantities determined based on the values ​​collected and determined by the DCS. Since DCS technology is relatively mature and most thermal power units have DCS systems, reading input quantities directly from the DCS ensures the accuracy of the acquired input quantities. Furthermore, obtaining input quantities through the thermal power unit's existing system reduces modification work and calibration costs.

[0062] In one embodiment of the present application, when reading input quantities, the four input quantities described above may fluctuate in actual applications. To ensure the accuracy of the acquired input quantities, in this embodiment, when reading the input quantities, if any input quantity is determined to have slight fluctuations, an average value within that time period may be calculated as the value of the input quantity read. The time period for calculating the average value may be the same as the measurement time period used to determine the measurement result in step S202.

[0063] Thus, the present application obtains the corresponding input and output quantities under the current working conditions.

[0064] Step S204 , switching to different working conditions, and repeating the flue gas flow measurement comparison test and reading the input quantity under each working condition.

[0065] Specifically, because a variety of operating conditions exist during the actual operation of a thermal power plant, the flue gas flow correction coefficient is different under each operating condition. Therefore, the input and output quantities under different operating conditions are collected sequentially. That is, when the thermal power plant switches to another operating condition, the collection process of steps S201 to S203 is repeated, and the flue gas flow measurement comparison test is performed under each operating condition to obtain the output quantity under the current operating condition and read the input quantity under that operating condition. This continues until the input and output data set contains the input and output quantities under most operating conditions of the unit's normal operation. It should be noted that the order in which the input and output quantities under each operating condition are obtained is not restricted here, and they can be obtained simultaneously.

[0066] Step S102: Using the input and output data sets as training data sets, training a preset RBF neural network and adjusting the parameters of the RBF neural network.

[0067] Specifically, the parameters of the RBF neural network model require multiple sets of corresponding input and output quantities to be calculated through a learning algorithm. Therefore, the data in the input and output data sets are used as training data to train the pre-built RBF neural network.

[0068] In one embodiment of the present application, the structure of the pre-built RBF neural network model is as follows Figure 3 As shown, the model consists of three layers, of which the first layer, the input layer, is used to input data. The number of nodes in this layer is 4, which is equal to the number of the above-mentioned input quantities. A nonlinear mapping is implemented from the input layer to the hidden layer. The function of the hidden layer is to map vectors from low dimensions to high dimensions. For example, the number of nodes in the hidden layer in this embodiment is 5, which realizes the mapping of the low dimension 4 to the high dimension 5, thereby transforming the low-dimensional linearly inseparable situation to the high-dimensional linearly separable situation. The transformation from the hidden layer space to the output layer space is linear, and the output layer is used to output the calculation results of the network model.

[0069] As a possible implementation method, this application uses the Gaussian activation function as the radial basis function. In other embodiments, other functions can be used according to actual needs, such as an abnormal sigmoid function and a quasi-quadratic function, etc., which are not limited here. In this embodiment, when the Gaussian function is used as the radial basis function, the output of the i nodes in the hidden layer can be expressed by the following formula:

[0070]

[0071] Among them, i is any node in the hidden layer, u i is the output of the i-th node in the hidden layer, is the input sample vector, The center vector of the Gaussian function, σ i is the normalization constant.

[0072] And, the output of the output layer can be expressed by the following formula:

[0073]

[0074] Among them, y is the output of the RBF neural network, w i is the weight coefficient from the hidden layer to the output layer.

[0075] In the RBF neural network model expressed by the above two formulas, σ i and w i is an unknown number. Among them, and σ i The value of σ can be determined based on experience. i Affects the mapping range of the network, and a moderate value is appropriate. The Gaussian basis function should be within the valid input mapping range. i The value of needs to be obtained through a learning algorithm, that is, the value of the parameter is adjusted through training.

[0076] When training a specific model, as a possible implementation method, first construct a performance indicator function, that is, an objective function. In this example, the least squares loss function can be used. Then, the objective function expressed by the following formula is optimized:

[0077]

[0078] Where N is the number of samples, t p is the expected value of output, y p is the actual value of the output. When optimizing, the objective function can be minimized by methods such as stochastic gradient descent (SGD).

[0079] Furthermore, according to the above formula, w can be obtained by the following formula i The value of is learned:

[0080]

[0081] Where η is the learning rate, 0<η<1, and the meanings of the parameters repeated in the above formula are not repeated here.

[0082] In this way, the weighted coefficients from each node in the hidden layer to the output layer can be learned in sequence to complete the training of the RBF neural network. The implementation method of each step in the training process can refer to the RBF neural network training method in related technologies, which will not be repeated here.

[0083] Step S103: obtaining the real-time input quantity and the real-time flue gas flow measured by the multi-point flow meter, and inputting the real-time input quantity into the trained RBF neural network to calculate the real-time flue gas flow correction coefficient.

[0084] Specifically, the above steps of obtaining a data set and performing model training can be performed in advance before the actual flow correction stage. During the real-time flow correction stage, since the flue gas volume correction coefficient changes dynamically with the change of the unit operating parameters, the present application obtains the real-time values ​​of the above four input quantities and the real-time flue gas flow measured by the multi-point flow meter, and inputs the real-time input quantities into the trained RBF neural network to calculate the real-time flue gas volume correction coefficient under the current operating conditions.

[0085] In one embodiment of the present application, when acquiring real-time input quantities, the data can also be obtained in real time from the existing DCS control system of the thermal power plant, in the manner described above for acquiring input quantities in the input and output data sets. When acquiring real-time flue gas flow measured by a multi-point flowmeter, a communication connection can be established with the thermal power plant's continuous emission monitoring system (CEMS) to read the real-time flue gas flow measured by the multi-point flowmeter from the CEMS. Specifically, the present application does not directly acquire data measured by the multi-point flowmeter. This is because multi-point flowmeters generally only output differential pressure signals. In actual operation, the process of calculating flue gas flow from these signals along with parameters such as flue area, flue gas temperature, flue gas pressure, and flue gas humidity is typically implemented in the programmable logic controller (PLC) of the thermal power plant's CMES. Therefore, the present application establishes communication with the CEMS to read the real-time values ​​of the multi-point flowmeter from the CEMS. Thus, the existing system of the thermal power plant is reused to acquire the real-time values ​​measured by the multi-point flowmeter. The data acquisition method of the present embodiment reduces the workload for system upgrades and reduces system complexity for easier maintenance.

[0086] Step S104: multiply the real-time flue gas flow rate by the real-time flue gas volume correction coefficient to obtain a corrected flue gas flow rate measurement value.

[0087] Specifically, based on the trained flue gas volume correction RBF neural network model and the collected input data, the flue gas volume correction coefficient of the current working condition is calculated in real time, and then the real-time flue gas flow measured by the collected multi-point flow meter is multiplied by the real-time flue gas volume correction coefficient to finally obtain the corrected flue gas flow.

[0088] In summary, the RBF neural network-based multi-point flowmeter calibration method of the present embodiment upgrades the existing multi-point flowmeter by adding a calibration step. The trained RBF neural network calculates the flowmeter's calibration coefficient under current operating conditions, and the detected flue gas flow rate is corrected using the real-time calculated calibration coefficient. This effectively improves the measurement accuracy of the multi-point flowmeter, enhancing the accuracy and reliability of the multi-point flowmeter's measurement results. Furthermore, this method, based on the existing thermal power unit system, reduces the workload for retrofitting and upgrading, simplifies the retrofit solution, reduces the cost, and facilitates implementation.

[0089] In order to implement the above embodiment, the present application also proposes a calibration device for a multi-point flow meter based on an RBF neural network. Figure 4 This is a schematic diagram of the structure of a calibration device for a multi-point flowmeter based on an RBF neural network proposed in an embodiment of the present application, as shown in FIG. Figure 4As shown, the apparatus includes an acquisition module 100 , a training module 200 , a first calculation module 300 and a second calculation module 400 .

[0090] Among them, the acquisition module 100 is used to use the primary fan frequency, the forced draft fan frequency, the induced draft fan frequency and the unit load as the input of the RBF neural network, and the flue gas volume correction coefficient as the output of the RBF neural network to obtain the input and output data sets covering multiple operating conditions of the thermal power unit.

[0091] The training module 200 is used to train the preset RBF neural network using the input and output data sets as training data sets, and adjust the parameters of the RBF neural network;

[0092] The first calculation module 300 is used to obtain the real-time input quantity and the real-time flue gas flow measured by the multi-point flow meter, and input the real-time input quantity into the trained RBF neural network to calculate the real-time flue gas flow correction coefficient.

[0093] The second calculation module 400 is configured to multiply the real-time flue gas flow rate by the real-time flue gas volume correction coefficient to obtain a corrected flue gas flow rate measurement value.

[0094] Optionally, in one embodiment of the present application, the acquisition module 100 is specifically used to: under the same operating condition, simultaneously measure the flue gas flow rate by manual measurement and multi-point flow meter measurement; obtain the measurement results of the manual measurement and the multi-point flow meter measurement under the current operating condition by calculating the average value within the measurement time; use the measurement result of the manual measurement as the true value, calculate the ratio of the measurement result of the manual measurement to the measurement result of the multi-point flow meter measurement, use the ratio as the output of the RBF neural network under the current operating condition, and obtain the input of the RBF neural network under the current operating condition from the distributed control system DCS of the thermal power unit; switch to different operating conditions, and repeat the flue gas flow measurement comparison test and read the input under each operating condition.

[0095] Optionally, in one embodiment of the present application, the RBF neural network constructed by the device includes an input layer, a hidden layer, and an output layer, and the output of i nodes in the hidden layer is expressed by the following formula:

[0096]

[0097] Among them, i is any node in the hidden layer, u i is the output of the i-th node in the hidden layer, is the input sample vector, The center vector of the Gaussian function, σ i is the normalization constant.

[0098] Optionally, in one embodiment of the present application, the output of the output layer is expressed by the following formula:

[0099]

[0100] Among them, y is the output of the RBF neural network, w i is the weight coefficient from the hidden layer to the output layer.

[0101] Optionally, in one embodiment of the present application, the training module 200 is specifically configured to:

[0102] Optimize the objective function expressed in the following formula:

[0103]

[0104] Where N is the number of samples, t p is the expected value of output, y p is the actual value of the output;

[0105] The following formula is used to calculate w i The value of is learned:

[0106]

[0107] Where η is the learning rate, 0<η<1.

[0108] Optionally, in one embodiment of the present application, the first calculation module 300 is specifically used to establish a communication connection with the online flue gas monitoring system CEMS of the thermal power unit, and read the real-time flue gas flow measured by the multi-point flow meter from the online monitoring system CEMS.

[0109] Based on the above embodiment, in order to more clearly illustrate the specific implementation process of the calibration method of the multi-point flowmeter based on the RBF neural network of the present application, the following is an example of a calibration process completed by a specific RBF neural network-based multi-point flowmeter calibration device proposed in one embodiment of the present application:

[0110] Figure 5 This is a schematic diagram of a specific RBF neural network-based multi-point flowmeter calibration device proposed in an embodiment of the present application. Figure 5 As shown, the device 10 includes: a digital signal processing chip (DSP) 11 and two communication modules, namely a first communication module 12 and a second communication module 13 .

[0111] Among them, the DSP chip 11 realizes fast neural network calculation, and can pre-train the flue gas volume correction RBF neural network model in the chip and store the model parameters, as well as perform correction calculations on the flue gas volume based on the current real-time collected data. The first communication module 12 communicates with the external unit DCS, and reads the four data of the primary fan frequency, the supply fan frequency, the induced draft fan frequency and the unit load in real time. The four data read can be used for pre-model training and to obtain real-time input in the correction stage. The first communication module 13 communicates with the external flue gas online monitoring system (CEMS), reads the real-time value measured by the multi-point flow meter determined by the PLC of the flue gas CEMS in combination with the environmental data acquisition instrument, and outputs the corrected value calculated by DSP11.

[0112] It should be noted that, because the communication data volume of the two communication modules is relatively small and the communication rate requirement is also relatively low, this application adopts the more commonly used MODBUS-RTU communication protocol, and the physical layer is the RS485 serial interface. The ModbusRTU protocol is an open serial protocol that has been widely used in various industrial monitoring equipment. This protocol uses the RS-485 serial interface for communication and can be supported by the DCS system and CEMS system of the thermal power unit of this application. Therefore, it is convenient to integrate Modbus-compatible devices into the programs of the various existing systems of the thermal power unit. Therefore, the device of this embodiment utilizes the original system of the thermal power unit for data transmission, which can further reduce the workload of the thermal power unit system modification and upgrade when performing flow correction, facilitating implementation.

[0113] Thus, the RBF neural network-based multi-point flowmeter calibration device of this embodiment, with a DSP chip as its core, includes two communication modules and is equipped with an RBF neural network algorithm. It collects real-time input data and the flue gas flow measured by the multi-point flowmeter, calculates a dynamic correction coefficient based on the obtained RBF neural network model, multiplies the coefficient by the flue gas flow measured by the multi-point flowmeter, and finally outputs the corrected flue gas flow.

[0114] It should be noted that the above explanation of the embodiment of the calibration method of the multi-point flowmeter based on the RBF neural network is also applicable to the device of this embodiment, and will not be repeated here.

[0115] In summary, the RBF neural network-based multi-point flowmeter calibration device of the present embodiment upgrades the existing multi-point flowmeter by adding a calibration step. The calibration coefficient for the flowmeter under current operating conditions is calculated using a trained RBF neural network. The detected flue gas flow rate is then corrected using the real-time calculated calibration coefficient, effectively improving the measurement accuracy of the multi-point flowmeter and enhancing the accuracy and reliability of the multi-point flowmeter's measurement results. Furthermore, this device, based on the existing thermal power unit system, reduces the workload for retrofitting and upgrading, simplifies the retrofit solution, reduces the cost, and facilitates implementation.

[0116] In order to implement the above embodiments, the present application also proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a correction method for a multi-point flow meter based on an RBF neural network as described in any of the above embodiments is implemented.

[0117] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0118] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0119] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0120] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.

[0121] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0122] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0123] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0124] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A calibration method for a multi-point flow meter based on an RBF neural network, characterized in that: The following steps are involved: The primary fan frequency, the forced draft fan frequency, the induced draft fan frequency, and the unit load are used as inputs of the RBF neural network, and the flue gas volume correction coefficient is used as the output of the RBF neural network to obtain input and output data sets covering multiple operating conditions of the thermal power unit, wherein the obtaining of the input and output data sets covering multiple operating conditions of the thermal power unit includes: under the same operating condition, simultaneously measuring the flue gas flow rate by manual measurement and multi-point flow meter measurement; obtaining the measurement results of the manual measurement and the multi-point flow meter measurement under the current operating condition by calculating the average value within the measurement time; using the measurement result of the manual measurement as the true value, calculating the ratio of the measurement result of the manual measurement to the measurement result of the multi-point flow meter measurement, and using the ratio as the output of the RBF neural network under the current operating condition, and obtaining the input of the RBF neural network under the current operating condition from the distributed control system DCS of the thermal power unit; switching to different operating conditions, and repeating the flue gas flow measurement comparison test and reading the input under each operating condition; Using the input and output data sets as training data sets, training the preset RBF neural network, and adjusting the parameters of the RBF neural network; Acquire real-time input and real-time flue gas flow measured by a multi-point flow meter, and input the real-time input into a trained RBF neural network to calculate a real-time flue gas flow correction coefficient; The real-time flue gas flow rate is multiplied by the real-time flue gas volume correction coefficient to obtain a corrected flue gas flow rate measurement value.

2. The calibration method according to claim 1, wherein: The RBF neural network includes an input layer, a hidden layer and an output layer. The following formula is used to express the hidden layer: i Output of each node: in, i is any node in the hidden layer, is the hidden layer i The output of the node, is the input sample vector, The center vector of the Gaussian function, is the normalization constant.

3. The calibration method according to claim 2, wherein: The output of the output layer is expressed by the following formula: in, y is the output of the RBF neural network, is the weight coefficient from the hidden layer to the output layer.

4. The calibration method according to claim 3, wherein: Training the preset RBF neural network includes: Optimize the objective function expressed in the following formula: in, N is the sample size, is the expected value of the output, is the actual value of the output; The following formula is used to The value of is learned: in, is the learning rate, 0< <1.

5. The calibration method according to claim 1, wherein: The method of obtaining the real-time flue gas flow rate measured by the multi-point flow meter includes: A communication connection is established with the flue gas online monitoring system CEMS of the thermal power unit, and the real-time flue gas flow measured by the multi-point flow meter is read from the online monitoring system CEMS.

6. A calibration device for a multi-point flow meter based on an RBF neural network, characterized in that: Includes the following modules: an acquisition module, configured to use the primary fan frequency, the forced draft fan frequency, the induced draft fan frequency, and the unit load as inputs of an RBF neural network, and use the flue gas volume correction coefficient as the output of the RBF neural network, to acquire an input and output data set covering multiple operating conditions of the thermal power unit; A training module, configured to train the preset RBF neural network using the input and output data sets as training data sets, and adjust the parameters of the RBF neural network; A first calculation module is used to obtain a real-time input quantity and a real-time flue gas flow rate measured by a multi-point flow meter, and input the real-time input quantity into a trained RBF neural network to calculate a real-time flue gas flow correction coefficient; a second calculation module, configured to multiply the real-time flue gas flow rate by the real-time flue gas volume correction coefficient to obtain a corrected flue gas flow rate measurement value; Among them, the acquisition module is specifically used to: under the same operating condition, simultaneously measure the flue gas flow rate by manual measurement and multi-point flow meter measurement; obtain the measurement results of the manual measurement and the multi-point flow meter measurement under the current operating condition by calculating the average value within the measurement time; use the measurement result of the manual measurement as the true value, calculate the ratio of the measurement result of the manual measurement to the measurement result of the multi-point flow meter measurement, use the ratio as the output of the RBF neural network under the current operating condition, and obtain the input of the RBF neural network under the current operating condition from the distributed control system DCS of the thermal power unit; switch to different operating conditions, and repeat the flue gas flow measurement comparison test and read the input under each operating condition.

7. The calibration device according to claim 6, characterized in that The first calculation module is specifically configured to: A communication connection is established with the flue gas online monitoring system CEMS of the thermal power unit, and the real-time flue gas flow measured by the multi-point flow meter is read from the online monitoring system CEMS.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the correction method of a multi-point flow meter based on an RBF neural network as claimed in any one of claims 1 to 5 is implemented.

Images