A dual-frequency calibrated ultrasonic flow velocity detection method

Abstract

This invention relates to the field of ultrasonic flow velocity detection technology for flues, specifically a dual-frequency calibrated ultrasonic flow velocity detection method. In this invention, a bounded Levenberg-Marquardt algorithm is used to synchronously iteratively solve for sound velocity, axial flow velocity, first channel delay, and second channel delay. A Hidden Markov Model is used to map the dual-frequency flow velocity difference, dual-frequency sound velocity difference, amplitude ratio, upstream / downstream time difference offset, and flow velocity sign consistency to five operating conditions: normal flow, high attenuation, enhanced water vapor, enhanced dust, and local backflow. A gradient boosting decision tree is used to split and calculate the first frequency confidence weight, second frequency confidence weight, flue gas axial flow velocity, four propagation times, and dual-channel delay. The flow velocity calibration deviation is accumulated tree by tree, which can maintain output continuity under conditions of increased dust concentration, enhanced condensation water vapor, and alternating local backflow, and improve the consistency of the dual-frequency calibration value in response to changes in the actual axial flow velocity of the flue.

Description

Technical Field

[0001] This invention relates to the field of ultrasonic flow velocity testing technology for flues, and more particularly to a dual-frequency calibrated ultrasonic flow velocity testing method. Background Technology

[0002] The field of ultrasonic flow velocity testing technology for flues aims to solve the problems of flow velocity measurement errors caused by high temperature, dust, water vapor, corrosive gases, uneven cross-sectional flow velocity distribution, local backflow, and sound wave attenuation in flues. It enables the testing device to output stable flow velocity values ​​under conditions of flue gas composition changes and temperature fluctuations, and reduces the risk of wear, blockage, or corrosion caused by mechanical probes contacting flue gas.

[0003] A dual-frequency calibrated ultrasonic velocity detection method aims to improve the stability and reliability of flue gas velocity detection results under conditions of high temperature, dust, water vapor, corrosive gases, and uneven flow field distribution in flue gas ducts. This method obtains the downstream propagation time and upstream propagation time using a first ultrasonic frequency and a second ultrasonic frequency, respectively. Based on the calculated velocity, sound velocity, received amplitude, or propagation time offset corresponding to the two frequencies, a calibration coefficient is constructed to correct the systematic errors generated by single-frequency ultrasonic velocimetry.

[0004] Existing dual-frequency calibrated ultrasonic velocity measurement typically uses the downstream and upstream propagation times corresponding to the first and second ultrasonic frequencies as core inputs. Calibration coefficients are then constructed based on the calculated velocity, sound velocity, received amplitude, and propagation time offset corresponding to the two frequencies. In actual operation, frequency selection often relies on preset frequency points or fixed sweep ranges. When the dust particle size distribution, water vapor content, and corrosive gas concentration within the flue change, the peak frequency band of the received echo will shift. Using a fixed frequency point in velocity measurement will reduce the effective echo amplitude. Identification of the first wave arrival time is easily affected by subsequent reflected waves. Furthermore, existing calibration procedures often use sound velocity... Changes in axial velocity and electroacoustic channel delay are processed separately. When temperature fluctuations cause changes in sound velocity, the channel delay error may be converted into a propagation time difference, which may then be mistaken for a change in flow velocity. For example, when the flue temperature rises stepwise and the receiving link delay drifts synchronously, simply correcting based on the time difference between forward and reverse flow will result in an artificially high flow velocity. Moreover, current operating condition judgments often use threshold comparison to directly switch calibration coefficients. When local backflow, dust pulses, and short-term enhancements of water vapor occur alternately in adjacent sampling periods, the calibration coefficients are prone to frequent jumps, resulting in sawtooth fluctuations in the output flow velocity, which affects the calculation of flue gas emissions, fan control, and judgment of flue operation status. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a dual-frequency calibrated ultrasonic flow velocity detection method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a dual-frequency calibrated ultrasonic flow velocity detection method, comprising the following steps:

[0007] S1: Based on the working configuration parameters of the flue ultrasonic flow velocity detection device, first make transducer A emit a downstream sound beam to transducer B, read the echo mean value frequency by frequency and compare it with the historical peak value, replace the peak value and switch to the upstream sound path, latch the frequency boundary frequency by frequency to obtain the bidirectional peak frequency boundary table.

[0008] S2: Based on the bidirectional peak frequency boundary table, write the downstream peak frequency into the first drive register area and the upstream peak frequency into the second drive register area. Calculate the sound path according to the flue inner diameter and installation angle, and compare the arrival time and amplitude of the first wave one by one to obtain the dual-frequency propagation constraint group.

[0009] S3: Based on the dual-frequency propagation constraint group, the sound speed, axial flow velocity, first channel delay and second channel delay are substituted into the forward and reverse flow time equations. The Levenburg-Marquardt algorithm is used to calculate the propagation time residual and delay deviation round by round. The out-of-bounds values ​​are replaced with the sound speed, flow velocity and delay boundary values ​​to obtain the sound speed, flow velocity and delay decoupling amount.

[0010] S4: Based on the sound velocity and flow velocity delay decoupling amount and dual-frequency propagation constraint group, a hidden Markov model is used to match the dual-frequency flow velocity difference, dual-frequency sound velocity difference, amplitude ratio, and upstream and downstream time difference offset item by item with the flue working condition. The dual-frequency reliable weight and calibration coefficient are selected according to the state results to obtain the working condition path weight coefficient group.

[0011] S5: Based on the working condition path weight coefficient group, sound velocity flow rate delay decoupling amount and dual frequency propagation constraint group, a gradient boosting decision tree is used to multiply the first frequency flow velocity and the second frequency flow velocity by the credible weights respectively, and then superimpose the upstream and downstream time difference compensation, sound velocity offset compensation and amplitude deviation compensation to obtain the flue flow velocity calibration value.

[0012] As a further aspect of the present invention, the operating configuration parameters of the flue ultrasonic flow velocity detection device include: inlet end closing command, exhaust end closing command, transducer A transmitting port identifier, transducer B receiving port identifier, transducer B transmitting port identifier, transducer A receiving port identifier, sweep frequency lower limit, sweep frequency upper limit, downstream sweep frequency interval, upstream neighborhood sweep frequency interval, single frequency point echo reading count, echo reading interval, excitation voltage, pulse width, burst period, receiving gain, echo search window start point, echo search window length, downstream historical peak initial amplitude, upstream historical peak initial amplitude, downstream frequency boundary extension width, and frequency boundary latching field. The bidirectional peak frequency boundary table includes downstream peak frequency, upstream peak frequency, downstream scan lower limit, downstream scan upper limit, and upstream scan lower limit. The upper limit of the counter-current scan and the peak echo amplitude are defined. The dual-frequency propagation constraint group includes the first frequency downstream propagation time, the first frequency upstream propagation time, the second frequency downstream propagation time, the second frequency upstream propagation time, the first frequency echo amplitude, and the second frequency echo amplitude. The sound velocity and flow velocity delay decoupling amount includes the flue gas sound velocity, the flue gas axial velocity, the first frequency channel delay, the second frequency channel delay, the first delay deviation, and the second delay deviation. The operating condition path weight coefficient group includes the flue operating condition state, the first frequency reliable weight, the second frequency reliable weight, the propagation time compensation coefficient, the sound velocity offset compensation coefficient, the amplitude attenuation compensation coefficient, and the flow velocity output damping coefficient. The flue flow velocity calibration value specifically includes the flue axial flow velocity value, the time compensation amount, the sound velocity compensation amount, the amplitude compensation amount, and the damped flow velocity output value.

[0013] As a further aspect of the present invention, the specific steps for generating the bidirectional peak frequency boundary table are as follows:

[0014] Based on the working configuration parameters of the flue ultrasonic flow velocity detection device, the flue inlet and outlet are closed, the transmitter port of transducer A, the receiver port of transducer B, the lower limit of frequency sweep, the upper limit of frequency sweep, and the frequency sweep interval are set, and the average amplitude of five echoes is read frequency by frequency. The current average amplitude is compared with the historical peak value of the downstream flow to obtain the downstream peak frequency table.

[0015] Based on the downstream peak frequency table, the transmitting port of transducer B and the receiving port of transducer A are switched, and the average amplitude of five echoes is read frequency by frequency along the neighborhood of the downstream frequency boundary. The current average amplitude is compared with the historical peak of the reverse flow, and the downstream peak frequency, the reverse peak frequency and the frequency boundary are merged to obtain the bidirectional peak frequency boundary table.

[0016] As a further aspect of the present invention, the specific steps for generating the dual-frequency propagation constraint group are as follows:

[0017] Based on the bidirectional peak frequency boundary table, the downstream peak frequency is written into the first drive register area, and the upstream peak frequency is written into the second drive register area. The downstream sound path and the upstream sound path are calculated according to the flue inner diameter, transducer installation angle and sound beam oblique distance to obtain the dual-frequency sound path register table.

[0018] Based on the dual-frequency sound path register, the first driving register frequency is called to trigger the downstream sound beam, and the second driving register frequency is called to trigger the upstream sound beam. The arrival time of the first wave, the echo amplitude, and the sound path number are read. The propagation time, echo amplitude, and sound path number are then grouped together to obtain the dual-frequency propagation constraint group.

[0019] As a further aspect of the present invention, the specific steps for generating the sound velocity-flow velocity delay decoupling amount are as follows:

[0020] Based on the dual-frequency propagation constraint group, the first frequency downstream propagation time, the first frequency upstream propagation time, the second frequency downstream propagation time, and the second frequency upstream propagation time are read. The flue inner diameter, transducer installation angle, and dual-frequency sound path are retrieved. The sound velocity, axial flow velocity, first channel delay, and second channel delay are written into the four downstream and upstream time equations respectively to establish a propagation time equation group.

[0021] Based on the propagation time equations, the Levenburg-Marquardt algorithm is used to calculate the time residuals of the first frequency downstream, the first frequency upstream, the second frequency downstream, and the second frequency upstream in turn. The delay of the first channel is subtracted from the delay of the first reference, and the delay of the second channel is subtracted from the delay of the second reference. The time residuals and delay deviations are accumulated according to their weights to obtain the residual deviation series.

[0022] Based on the residual deviation sequence, the residual decrease, sound velocity boundary, flow velocity boundary, first delay boundary and second delay boundary of adjacent rounds are compared. When the sound velocity, flow velocity or delay exceeds the boundary, it is rewritten to the corresponding boundary value. When the residual decrease is lower than the threshold or the round reaches the upper limit, the sound velocity, axial flow velocity and dual-channel delay are latched to obtain the sound velocity, flow velocity and delay decoupling amount.

[0023] As a further aspect of the present invention, the bounded Levenberg-Marquardt algorithm first sets the sound velocity, axial flow velocity, first channel delay, and second channel delay as four parameters to be determined. An initial parameter set is generated based on the initial values ​​of the sound velocity, axial flow velocity, first reference delay, and second reference delay. The initial parameter set is then written into the first frequency downstream time equation, the first frequency upstream time equation, the second frequency downstream time equation, and the second frequency upstream time equation. Combined with the flue inner diameter, transducer installation angle, and dual-frequency sound path, four predicted propagation times are calculated. These four predicted propagation times are then... Subtracting the measured propagation time from the following times (first frequency downstream propagation time, first frequency upstream propagation time, second frequency downstream propagation time, and second frequency upstream propagation time) yields four time residuals. Subtracting the first reference delay from the first channel delay yields the first delay deviation, and subtracting the second reference delay from the second channel delay yields the second delay deviation. Using preset time and delay weights, the four time residuals, the first delay deviation, and the second delay deviation are summed by square weights to obtain the current round's residual deviation value. Based on the four time residuals and the first reference delay... The relationship between the delay deviation and the second delay deviation and the changes in the four parameters to be determined is used to generate a sensitivity matrix, and the parameter correction amount is obtained in combination with the damping coefficient. When the residual deviation value decreases after the parameter correction amount is written, the corrected parameter set is used and the damping coefficient is reduced. When the residual deviation value does not decrease after the parameter correction amount is written, the parameter set before correction is retained and the damping coefficient is increased before re-obtaining the parameter correction amount. After each round of parameter correction, the sound velocity is limited to the range from the lower limit to the upper limit of the sound velocity, and the axial flow velocity is limited to the range from the lower limit to the upper limit of the flow velocity. The first channel... The channel delay is limited to the range of the first delay lower limit to the first delay upper limit, and the second channel delay is limited to the range of the second delay lower limit to the second delay upper limit. The difference between the residual deviation value of the current round and the residual deviation value of the previous round is recorded. When any of the following criteria is met: the difference is lower than the preset convergence threshold, the number of iterations reaches the preset upper limit, the sound velocity reaches the boundary, the axial flow velocity reaches the boundary, the first channel delay reaches the boundary, or the second channel delay reaches the boundary, the iteration stops, and the sound velocity, axial flow velocity, first channel delay, and second channel delay are latched to obtain the sound velocity-flow velocity delay decoupling amount.

[0024] As a further aspect of the present invention, the specific steps for generating the working condition path weight coefficient group are as follows:

[0025] Based on the aforementioned sound velocity and flow velocity delay decoupling quantity and dual-frequency propagation constraint group, the flue gas sound velocity, flue gas axial flow velocity, first channel delay, second channel delay, dual-frequency echo amplitude, and four propagation times are called to calculate the first frequency flow velocity, second frequency flow velocity, dual-frequency flow velocity difference, dual-frequency sound velocity difference, amplitude ratio, upstream and downstream time difference offset, and flow velocity sign consistency quantity, thus obtaining the flue gas duct condition discrimination quantity group;

[0026] Based on the flue gas condition discrimination set, a hidden Markov model is used to compare the dual-frequency velocity difference with the velocity difference threshold, the dual-frequency sound velocity difference with the sound velocity difference threshold, the amplitude ratio with the upper and lower limits of the amplitude ratio, the forward and reverse flow time difference offset with the time offset threshold, and the velocity sign consistency quantity is matched with the backflow judgment condition. The normal flow, high attenuation, water vapor enhancement, dust enhancement, or local backflow state is selected to obtain the condition state path table.

[0027] Based on the operating condition path table, the current operating condition and the previous cycle operating condition are read, and the number of consecutive hits and dwell times of the state are compared. If the threshold is not reached, the coefficient of the previous cycle is used. If the threshold is reached, the first frequency confidence weight, the second frequency confidence weight, the propagation time compensation coefficient, the sound speed offset compensation coefficient, the amplitude attenuation compensation coefficient, and the flow velocity output damping coefficient corresponding to the current operating condition are selected to obtain the operating condition path weight coefficient group.

[0028] As a further aspect of the present invention, the Hidden Markov Model first defines five implicit operating conditions: conventional flow, high attenuation, water vapor enhancement, dust enhancement, and local backflow. It then defines two-frequency velocity difference, two-frequency sound velocity difference, amplitude ratio, upstream / downstream time difference offset, and velocity sign consistency as observation groups. Model parameters are established according to an initial state probability table, a state transition probability table, and an observation matching probability table. Within a single discrimination period, the two-frequency velocity difference is segmented into a velocity difference threshold segment, the two-frequency sound velocity difference is segmented into a sound velocity difference threshold segment, the amplitude ratio is segmented into upper and lower limits, the upstream / downstream time difference offset is segmented into a time offset threshold segment, and the velocity sign consistency is segmented into the backflow determination condition. Based on the segmentation, interval, and consistency markers, the observation matching probability table is searched to generate observation matching scores corresponding to the five implicit operating conditions. The scores of the five types of state paths in the previous period are multiplied by the state transition probabilities and then multiplied by the observation matching scores in the current period to obtain the scores of the five types of candidate paths in the current period. The Viterbi decoding algorithm is used to sort the scores of the five types of candidate paths in descending order. The operating state corresponding to the first candidate path in the sorted order is selected as the operating state of the current period. The operating state of the previous period, the operating state of the current period, the state transition node, the number of consecutive hits, and the number of dwells are recorded. When multiple discrimination periods are input consecutively, threshold segment comparison, observation matching, state transition calculation, candidate path sorting, and state node recording are performed cycle by cycle to generate an operating state path table containing the period number, the operating state of the previous period, the operating state of the current period, the state transition node, the number of consecutive hits, the number of dwells, the first frequency confidence weight, the second frequency confidence weight, the propagation time compensation coefficient, the sound speed offset compensation coefficient, the amplitude attenuation compensation coefficient, and the flow velocity output damping coefficient.

[0029] As a further aspect of the present invention, the specific steps for generating the flue gas velocity calibration value are as follows:

[0030] Based on the aforementioned working condition path weight coefficient group, sound velocity flow rate delay decoupling amount, and dual-frequency propagation constraint group, a gradient boosting decision tree is adopted. The first frequency confidence weight, the second frequency confidence weight, flue gas axial velocity, four propagation times, and dual-channel delay are called to calculate the first frequency velocity and the second frequency velocity respectively. The first frequency velocity is multiplied by the first frequency confidence weight, and the second frequency velocity is multiplied by the second frequency confidence weight to obtain the dual-frequency weighted velocity amount.

[0031] Based on the aforementioned dual-frequency weighted flow velocity, the propagation time compensation coefficient, sound velocity offset compensation coefficient, amplitude attenuation compensation coefficient, dual-frequency upstream and downstream time difference offset, dual-frequency sound velocity difference and amplitude ratio deviation are called. The three compensation coefficients are multiplied by the corresponding offsets respectively to calculate the time compensation, sound velocity compensation and amplitude compensation, and obtain the weighted compensated flow velocity group.

[0032] Based on the weighted compensation velocity group, the dual-frequency weighted velocity, time compensation, sound velocity compensation, amplitude compensation, velocity output damping coefficient, and the velocity output value of the previous cycle are called. The dual-frequency weighted velocity is added to the three compensation values ​​to generate the compensation velocity. The compensation velocity and the velocity output value of the previous cycle are weighted according to the velocity output damping coefficient to obtain the flue gas velocity calibration value.

[0033] As a further aspect of the present invention, the gradient boosting decision tree first comprises an input variable group consisting of a first frequency confidence weight, a second frequency confidence weight, flue gas axial velocity, first frequency downstream propagation time, first frequency upstream propagation time, second frequency downstream propagation time, second frequency upstream propagation time, first channel delay, and second channel delay. The difference between the calibrated velocity and the initial velocity output is set as the training residual. In the first round of training, regression tree nodes are generated using the input variable group. Each input variable is compared against its corresponding splitting threshold, and the first-ranked splitting threshold based on the loss reduction is used as the node splitting threshold. A left branch sample group and a right branch sample group are generated. The mean of the leaf node residuals is calculated and used as the leaf node output. The leaf node output is multiplied by the learning rate and then superimposed onto the initial velocity output to obtain the predicted velocity for this round. After each round of training, the calibrated velocity is subtracted from the predicted velocity for this round to generate the next... One round of training residuals is used to build the next regression tree until any one of the following criteria is met: the number of trees reaches a preset number, the residual decrease is lower than a preset threshold, or the number of samples in a single leaf node is lower than a preset lower limit. Multiple regression trees, node splitting thresholds, leaf node outputs, and learning rates are latched. During online calculation, the current input variable group is written into multiple regression trees in sequence. According to the node splitting threshold, the left branch sample interval and right branch sample interval are entered layer by layer. The corresponding leaf node outputs of multiple regression trees are read and accumulated according to the learning rate to obtain the flow velocity calibration deviation. Then, the first frequency flow velocity, the second frequency flow velocity, the first frequency confidence weight, and the second frequency confidence weight are combined to generate a dual-frequency weighted flow velocity. The time compensation, sound velocity compensation, and amplitude compensation are superimposed to generate a compensated flow velocity. Then, according to the flow velocity output damping coefficient, the compensated flow velocity is weighted and fused with the flow velocity output value of the previous cycle to obtain the flue flow velocity calibration value.

[0034] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0035] In this invention, a closed-loop processing chain is formed by bidirectional sound beam propagation around the flue, dual-frequency peak frequency boundary latching, propagation constraint construction, sound velocity and flow velocity delay synchronous decoupling, working condition path discrimination, and weighted compensation output. Starting from the downstream sound beam from transducer A to transducer B, the echo mean and historical peak values ​​are compared frequency by frequency. After the peak value is replaced, the system switches to the reverse flow path, so that the downstream peak frequency boundary and the reverse flow peak frequency boundary are jointly limited by the current dust attenuation, water vapor enhancement, flue gas temperature, and installation path. This reduces the risk of misjudging the arrival time of the first wave after the fixed drive frequency deviates from the effective echo frequency band.

[0036] In this invention, the bounded Levenburg Marquardt algorithm is used to synchronously iteratively solve for the sound speed, axial flow velocity, first channel delay and second channel delay. By using propagation time residual, delay deviation, damping correction and boundary constraints to jointly limit the decoupling direction, it is possible to reduce crosstalk between sound speed drift, receiving link delay drift and axial flow velocity changes caused by temperature fluctuations.

[0037] In this invention, a hidden Markov model is used to map the dual-frequency velocity difference, dual-frequency sound velocity difference, amplitude ratio, upstream and downstream time difference offset, and velocity sign consistency to five types of operating conditions: normal flow, high attenuation, water vapor enhancement, dust enhancement, and local backflow. The Viterbi decoding algorithm is used to generate an operating condition path table based on the state transition probability and observation matching probability, so that the operating condition judgment is not directly dominated by single-cycle abnormal echoes.

[0038] In this invention, a gradient boosting decision tree is used to perform split calculations on the first frequency confidence weight, the second frequency confidence weight, the flue gas axial velocity, the four propagation times, and the dual-channel delay. The velocity calibration deviation is accumulated for each regression tree. Then, the first frequency velocity, the second frequency velocity, the upstream and downstream time difference compensation, the sound velocity offset compensation, the amplitude deviation compensation, and the velocity output value of the previous cycle are fused together. This can maintain the continuity of output under conditions of increased dust concentration, enhanced condensation and water vapor, and alternating local backflow, and improve the consistency of the dual-frequency calibration value in response to changes in the actual axial velocity of the flue. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the main steps of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0041] Example 1

[0042] Please see Figure 1 This invention provides a technical solution: a dual-frequency calibrated ultrasonic flow velocity detection method, comprising the following steps:

[0043] S1: Based on the working configuration parameters of the flue ultrasonic flow velocity detection device, first make transducer A emit a downstream sound beam to transducer B, read the echo mean value frequency by frequency and compare it with the historical peak value, replace the peak value and switch to the upstream sound path, latch the frequency boundary frequency by frequency to obtain the bidirectional peak frequency boundary table.

[0044] S2: Based on the bidirectional peak frequency boundary table, the downstream peak frequency is written into the first drive register area and the upstream peak frequency is written into the second drive register area. The sound path is calculated according to the flue inner diameter and installation angle. The arrival time and amplitude of the first wave are compared one by one to obtain the dual-frequency propagation constraint group.

[0045] S3: Based on the dual-frequency propagation constraint group, the sound speed, axial flow velocity, first channel delay and second channel delay are substituted into the forward and reverse flow time equation. The Levenburg-Marquardt algorithm is used to calculate the propagation time residual and delay deviation round by round. The out-of-bounds values ​​are replaced with the sound speed, flow velocity and delay boundary values ​​to obtain the sound speed, flow velocity and delay decoupling amount.

[0046] S4: Based on the decoupling amount of sound velocity and flow velocity delay and the dual-frequency propagation constraint group, the hidden Markov model is adopted to match the dual-frequency flow velocity difference, dual-frequency sound velocity difference, amplitude ratio, and the offset of the upstream and downstream time difference with the flue working condition. The dual-frequency reliable weight and calibration coefficient are selected according to the state results to obtain the working condition path weight coefficient group.

[0047] S5: Based on the working condition path weight coefficient group, sound velocity and flow velocity delay decoupling amount and dual frequency propagation constraint group, a gradient boosting decision tree is used to multiply the first frequency flow velocity and the second frequency flow velocity by the confidence weight respectively, and then superimpose the upstream and downstream time difference compensation, sound velocity offset compensation and amplitude deviation compensation to obtain the flue flow velocity calibration value.

[0048] The operating configuration parameters of the flue gas ultrasonic flow velocity detection device include: inlet end shut-off command, exhaust end shut-off command, transducer A transmitter port identifier, transducer B receiver port identifier, transducer B transmitter port identifier, transducer A receiver port identifier, sweep frequency lower limit, sweep frequency upper limit, downstream sweep frequency interval, upstream neighborhood sweep frequency interval, single frequency point echo reading count, echo reading interval, excitation voltage, pulse width, burst period, receiver gain, echo search window start point, echo search window length, downstream historical peak initial amplitude, upstream historical peak initial amplitude, downstream frequency boundary extension width, and frequency boundary latching field. The bidirectional peak frequency boundary table includes downstream peak frequency, upstream peak frequency, downstream scan lower limit, downstream scan upper limit, upstream scan lower limit, and upstream scan... The upper limit and peak echo amplitude, the dual-frequency propagation constraint group includes the first frequency downstream propagation time, the first frequency upstream propagation time, the second frequency downstream propagation time, the second frequency upstream propagation time, the first frequency echo amplitude, and the second frequency echo amplitude. The sound velocity and flow velocity delay decoupling amount includes the flue gas sound velocity, the flue gas axial velocity, the first frequency channel delay, the second frequency channel delay, the first delay deviation, and the second delay deviation. The operating condition path weight coefficient group includes the flue operating condition state, the first frequency reliable weight, the second frequency reliable weight, the propagation time compensation coefficient, the sound velocity offset compensation coefficient, the amplitude attenuation compensation coefficient, and the flow velocity output damping coefficient. The flue flow velocity calibration value specifically includes the flue axial flow velocity value, the time compensation amount, the sound velocity compensation amount, the amplitude compensation amount, and the damped flow velocity output value.

[0049] The specific steps for generating a bidirectional peak frequency boundary table are as follows:

[0050] Based on the operating configuration parameters of the flue gas ultrasonic flow velocity detection device, a fixed-step ascending-order frequency sweep enumeration method is used to control the closure of the flue gas inlet and exhaust ends. Closure control is achieved by writing the inlet valve control item to "closed," writing the exhaust valve control item to "closed," setting the valve holding time to 3000ms, continuously reading the inlet valve status three times at 100ms intervals, continuously reading the exhaust valve status three times at 100ms intervals, and locking the flow channel lock item when both the inlet and exhaust valves are closed three times. The valve closure status is then registered. A port register mapping configuration method is used to set the transmitter port of transducer A and the receiver port of transducer B. The ports are configured by writing the transmit port selection to transducer A, the receive port selection to transducer B, the transmit enable to be enabled, the receive enable to be enabled, the receive gain to be 12dB, the pulse width to be 8μs, the burst period to be 8 cycles, and the excitation voltage to be 24V. The port configuration is recorded. A constant-step frequency increment method is used to set the lower sweep limit, upper sweep limit, and sweep interval. The sweep start point is set to 38000Hz, the sweep end point to 42000Hz, and the sweep interval to be 100Hz. The frequency increment starts at 38000Hz and increases by 100Hz each time until it stops at 42000Hz. The process involves: generating 41 downstream frequency sweep points and writing them into a downstream frequency sweep list; using the five-point arithmetic mean echo amplitude calculation method; and reading the average amplitude of five echoes frequency-by-frequency. This is achieved by writing the current frequency in the downstream frequency sweep list into the excitation frequency field, triggering a transmission, setting the echo window start point to 1.20ms, the echo window length to 0.80ms, and the analog-to-digital sampling rate to 1MHz. Five echo peak values ​​are then read consecutively at 20ms intervals. The first, second, third, fourth, and fifth amplitudes are recorded. The sum of the five amplitudes is divided by five to obtain the current average amplitude. The current frequency and the current amplitude are then compared. The average value is written to the downstream frequency sweep buffer, and the current frequency is traversed. The historical peak monotonic replacement comparison method is used to compare the current average amplitude with the downstream historical peak. The comparison is performed by setting the initial value of the downstream historical peak amplitude to 0mV and the initial value of the downstream historical peak frequency to 38000Hz. When the current average amplitude is greater than the downstream historical peak amplitude, the downstream historical peak amplitude is rewritten to the current average amplitude and the downstream historical peak frequency is rewritten to the current frequency. When the current average amplitude is not greater than the downstream historical peak amplitude, the downstream historical peak amplitude and downstream historical peak frequency remain unchanged. The traversal stops at 42000Hz and a downstream peak frequency table is generated.

[0051] Based on the downstream peak frequency table, a port swap register writing method is used to switch the transmit port of transducer B and the receive port of transducer A. The switching is achieved by writing the transmit enable to be off, the receive enable to be off, the transmit port selection to be transducer B, the receive port selection to be transducer A, the transmit enable to be on, and the receive enable to be on. A port switching wait time of 100ms is set. The transmit port selection is read and registered as transducer B, the receive port selection is read and registered as transducer A, the transmit enable is read and registered as on, and the receive enable is read and registered as on. Reverse port latching is performed, and a frequency boundary neighborhood expansion method is used to set a reverse frequency sweep sequence along the downstream frequency boundary neighborhood. The process involves reading the downstream peak frequency field from the downstream peak frequency table, extending the downstream peak frequency 300Hz towards the low-frequency side to obtain the low-frequency boundary, extending the downstream peak frequency 300Hz towards the high-frequency side to obtain the high-frequency boundary, writing the neighborhood sweep interval to 50Hz, increasing by 50Hz each time from the low-frequency boundary to the high-frequency boundary, forming 13 reverse-current neighborhood sweep points, and writing these 13 reverse-current neighborhood sweep points into the reverse-current sweep list. Using the five-point arithmetic mean echo amplitude calculation method, the average echo amplitude is read five times frequency-by-frequency along the downstream frequency boundary neighborhood. Frequency-by-frequency reading involves writing the current frequency in the reverse-current sweep list into the excitation frequency item, performing one transmit trigger, setting the echo window start point to 1.20ms, and setting the echo window length to 0.8ms. 0ms, set analog-to-digital sampling rate to 1MHz, continuously read five echo peak values ​​at 20ms intervals, record the first, second, third, fourth, and fifth reverse current amplitudes, add the five reverse current amplitudes and divide by five to obtain the current average amplitude, write the current frequency and current average amplitude to the reverse current frequency sweep buffer, and perform neighborhood frequency traversal. Use the reverse current historical peak monotonic replacement comparison method to compare the current average amplitude with the reverse current historical peak. The comparison is performed by setting the initial value of the reverse current historical peak amplitude to 0mV and the initial value of the reverse current historical peak frequency to the low-frequency boundary. When the current average amplitude is greater than the reverse current historical peak amplitude, the reverse current historical peak amplitude is rewritten as the current average amplitude. The historical peak frequency is rewritten to the current frequency. When the current average amplitude is not greater than the historical peak amplitude of the countercurrent, the historical peak amplitude and frequency of the countercurrent remain unchanged. The process stops after traversing to the high-frequency boundary. A key-value field merging table construction method is used to merge the downstream peak frequency, the countercurrent peak frequency, and the frequency boundary. The merging is achieved by writing the low-frequency boundary and the high-frequency boundary into the boundary number field, writing the downstream peak frequency into the downstream frequency field, writing the countercurrent peak frequency into the countercurrent frequency field, writing the low-frequency boundary into the boundary lower limit field, writing the high-frequency boundary into the boundary upper limit field, writing the downstream historical peak amplitude into the downstream amplitude field, writing the countercurrent historical peak amplitude into the countercurrent amplitude field, sorting the records in ascending order of the boundary number, and generating a bidirectional peak frequency boundary table.

[0052] The specific steps for generating the dual-frequency propagation constraint set are as follows:

[0053] Based on a bidirectional peak frequency boundary table, a dual-register sequential mapping write method is used to write the downstream and upstream peak frequencies. The register writing process involves reading the downstream frequency field and the upstream frequency field from the bidirectional peak frequency boundary table, setting the starting address of the first drive register to 2100H, the starting address of the second drive register to 2200H, setting the write word length to 32 bits, setting the frequency unit to Hz, setting the write order to low byte first, high byte last, writing the downstream peak frequency to the first drive register, writing the upstream peak frequency to the second drive register, reading the first drive register twice consecutively, reading the second drive register twice consecutively, writing the two read values ​​to the frequency register verification record, and performing frequency register latching. A three-parameter sound path lookup table conversion method is used, based on the flue inner diameter... The transducer installation angle and beam oblique distance are used to calculate the downstream and upstream sound paths. The conversion is achieved by reading the flue inner diameter as 1200mm, the transducer installation angle as 45°, and the beam oblique distance as 1697mm. The initial value of the sound path number is set to P001. The corresponding records for the inner diameter of 1200mm, installation angle of 45°, and oblique distance of 1697mm are matched in the sound path conversion table. The downstream and upstream sound paths are read as 1697mm and 1697mm respectively. The sound path number P001 is written into the sound path number field. The address of the first drive register area is written into the downstream frequency source field. The address of the second drive register area is written into the upstream frequency source field. The downstream sound path is written into the downstream sound path field. The upstream sound path is written into the upstream sound path field. The upstream sound path is written into the upstream sound path field. The upstream sound path is written into the upstream sound path field. The upstream sound path is written into the upstream sound path field. The upstream sound path is registered in ascending order of the sound path number and a dual-frequency sound path register table is generated.

[0054] Based on a dual-frequency sound path register, a dual-frequency directional pulse triggering method is adopted. The first drive register frequency is called to trigger the downstream sound beam. This is achieved by reading the latched frequency from the first drive register, writing the downstream trigger port as transducer A, writing the downstream receiver port as transducer B, setting the downstream excitation voltage to 24V, setting the downstream pulse width to 8μs, setting the downstream burst period to 8 cycles, setting the downstream trigger count to 1, executing the downstream transmission trigger, recording the downstream trigger number as FWD001, and then proceeding... The reverse current acoustic beam emission recording was performed using a dual-frequency directional pulse triggering method. The second drive register frequency was used to trigger the reverse current acoustic beam. This was achieved by reading the latched frequency from the second drive register, writing the reverse current trigger port as transducer B, writing the reverse current receiver port as transducer A, setting the reverse current excitation voltage to 24V, setting the reverse current pulse width to 8μs, setting the reverse current burst period to 8 cycles, setting the reverse current trigger count to 1, executing the reverse current emission trigger, and recording the reverse current trigger number as REV001. The reverse-current acoustic beam emission recording uses a first-wave threshold arrival determination method to read the arrival time, echo amplitude, and path number of the first wave. The reading process involves setting a sampling rate of 1MHz, an echo search start point of 0.50ms, an echo search end point of 5.00ms, and an amplitude threshold of 30mV. Starting from the echo search start point, sampling amplitudes are read in 1μs increments. The first sampling point with an amplitude not less than 30mV is recorded as the arrival time of the first wave. The peak value of this sampling point is read and recorded as the echo amplitude. The path number field in the dual-frequency path register is read and recorded as the path number. A path number key-value aggregation method is used to aggregate the propagation time, echo amplitude, and path number. Aggregation is achieved by using the path number as the primary key, writing the following fields: downstream propagation time, downstream echo amplitude, reverse propagation time, reverse echo amplitude, first drive register frequency, and second drive register frequency. Records are arranged in ascending order by path number, and a dual-frequency propagation constraint group is generated.

[0055] The specific steps for generating the sound velocity-flow-velocity delay decoupling quantity are as follows:

[0056] Based on a dual-frequency propagation constraint group, a four-propagation time field assembly method is used to read the first frequency downstream propagation time, the first frequency upstream propagation time, the second frequency downstream propagation time, and the second frequency upstream propagation time. The reading process involves using the sound path number as an index field, reading the first drive register frequency field, the second drive register frequency field, the downstream propagation time field, the upstream propagation time field, the echo amplitude field, and the sound path number field from the dual-frequency propagation constraint group, setting the time unit to μs, the frequency unit to Hz, and the sound path unit to mm, registering the first drive register frequency as the first frequency, and registering the second drive register frequency as the second frequency. The frequency, the downstream propagation time of the first frequency, the upstream propagation time of the first frequency, the downstream propagation time of the second frequency, and the upstream propagation time of the second frequency are all recorded in the first downstream time field. A geometric parameter fixed-value retrieval method is used to retrieve the flue inner diameter, transducer installation angle, and dual-frequency sound path. The retrieval results show the flue inner diameter as 1200mm, the transducer installation angle as 45°, the downstream sound path of the first frequency as 1697mm, the upstream sound path of the first frequency as 1697mm, and the downstream sound path of the second frequency as 1697mm. The second frequency reverse flow path was read as 1697 mm. The flue inner diameter was entered into the pipe diameter field, the transducer installation angle was entered into the angle field, and the four sound paths were entered into the sound path field. Geometric parameters were registered. Using the four-unknown-quantity time equation writing method, the sound velocity, axial flow velocity, first channel delay, and second channel delay were written into the four forward and reverse flow time equations respectively. The writing was done by setting the unknown quantity order as sound velocity, axial flow velocity, first channel delay, and second channel delay. The initial writing value of sound velocity was set to 346 m / s, the initial writing value of axial flow velocity was set to 0 m / s, the initial writing value of first channel delay was set to 2.0 μs, and the initial writing value of second channel delay was set to 2 μs. 0μs, write the first frequency downstream time term into the first time equation, write the first frequency upstream time term into the second time equation, write the second frequency downstream time term into the third time equation, write the second frequency upstream time term into the fourth time equation, write the first frequency downstream sound path and the first channel delay into the first time equation, write the first frequency upstream sound path and the first channel delay into the second time equation, write the second frequency downstream sound path and the second channel delay into the third time equation, write the second frequency upstream sound path and the second channel delay into the fourth time equation, write the flue inner diameter and the transducer installation angle into the common parameter area of ​​the four time equations, and establish a propagation time equation group;

[0057] Based on the propagation time equations, the Levenberg-Marquardt damped least squares iterative algorithm is used to iteratively load unknowns round by round. The loading is performed by setting a maximum of 30 rounds, an initial damping coefficient of 0.01, a damping amplification factor of 10, a damping reduction factor of 0.1, a sound velocity disturbance of 0.1 m / s, an axial velocity disturbance of 0.01 m / s, a first channel delay disturbance of 0.1 μs, a second channel delay disturbance of 0.1 μs, a time residual unit of μs, and a variable update order of sound velocity, axial velocity, first channel delay, and second channel delay. The variables for the current round are registered, and a four-residual sequential calculation method is used to calculate the first frequency sequentially round by round. The following time residuals are calculated: the first frequency countercurrent time residual, the second frequency downstream time residual, and the second frequency countercurrent time residual. The calculation is performed by reading the calculation times from the first, second, third, and fourth time equations. The difference between the first frequency downstream propagation time and the first time equation calculation time is written into the first frequency downstream time residual; the difference between the first frequency countercurrent propagation time and the second time equation calculation time is written into the first frequency countercurrent time residual; the difference between the second frequency downstream propagation time and the third time equation calculation time is written into the second frequency downstream time residual; and the difference between the second frequency countercurrent propagation time and the fourth time equation calculation time is written into the second frequency countercurrent time residual. Four residuals were registered, and the Levenberg-Marquardt variable increment calculation was performed using the central difference variable perturbation filling method. The calculation involved applying positive and negative perturbations to the sound velocity, the axial velocity, the first channel delay, and the second channel delay, respectively. The changes in the four time residuals were then read, and the increment calculation table was filled in four rows and four columns according to the variable order. The damping coefficient was written into the main diagonal augmented term of the increment calculation table. The four variable increments were solved using the Cholesky decomposition method. The sound velocity increment was written into the sound velocity update term, the axial velocity increment into the velocity update term, and the first channel delay increment into the first delay term. The update item writes the second channel delay increment into the second delay update item and rewrites the current round variable. Using the delay baseline deviation accumulation method, the first channel delay is subtracted from the first baseline delay, and the second channel delay is subtracted from the second baseline delay. The time residual and delay deviation are accumulated according to their weights. The accumulation is achieved by setting the first baseline delay to 2.0 μs, the second baseline delay to 2.0 μs, setting the weight of the first frequency downstream time residual to 1.00, setting the weight of the first frequency upstream time residual to 1.00, setting the weight of the second frequency downstream time residual to 1.00, setting the weight of the second frequency upstream time residual to 1.00, setting the weight of the first channel delay deviation to 0.20, and setting the weight of the second channel delay deviation to 0.20. Take the absolute value of the four time residuals, multiply them by their corresponding weights and accumulate them; take the absolute value of the first channel delay deviation, multiply it by 0.20 and accumulate it; take the absolute value of the second channel delay deviation, multiply it by 0.20 and accumulate it; write the accumulated value of the current round into the residual deviation record, and obtain the residual deviation sequence;

[0058] Based on the residual deviation sequence, a method for determining the residual decrease in adjacent rounds is adopted. This involves comparing the residual decrease in adjacent rounds by reading the current round's residual deviation value, reading the previous round's residual deviation value, subtracting the previous round's residual deviation value from the current round's residual deviation value and recording this difference as the residual decrease. A threshold of 0.001 μs is set for the residual decrease. When the residual decrease is less than 0.001 μs, a stop flag is set to 1; when the residual decrease is not less than 0.001 μs, a stop flag is set to 0. Round status is then recorded. A boundary projection rewriting method is used to compare the sound velocity boundary, flow velocity boundary, and first delay edge. The boundary and the second delay boundary are compared by setting the lower boundary of sound velocity to 300 m / s, the upper boundary of sound velocity to 500 m / s, the lower boundary of flow velocity to -40 m / s, the upper boundary of flow velocity to 40 m / s, the lower boundary of the first delay to 0 μs, the upper boundary of the first delay to 100 μs, the lower boundary of the second delay to 0 μs, and the upper boundary of the second delay to 100 μs. When the sound velocity is less than 300 m / s, the sound velocity is rewritten to 300 m / s; when the sound velocity is greater than 500 m / s, the sound velocity is rewritten to 500 m / s; when the axial flow velocity is less than -40 m / s, the axial flow velocity is rewritten to 500 m / s. The velocity is rewritten as -40 m / s. When the axial velocity is greater than 40 m / s, the axial velocity is rewritten as 40 m / s. When the first channel delay is less than 0 μs, the first channel delay is rewritten as 0 μs. When the first channel delay is greater than 100 μs, the first channel delay is rewritten as 100 μs. When the second channel delay is less than 0 μs, the second channel delay is rewritten as 0 μs. When the second channel delay is greater than 100 μs, the second channel delay is rewritten as 100 μs. Boundary state registration is performed. A round upper limit and variable latching method is adopted. When the residual decrease is lower than the threshold or the round reaches the upper limit, the sound velocity is latched. Axial velocity and dual-channel delay are latched by reading the stop flag, reading the current round, writing a latch instruction to enable when the stop flag is 1, writing a latch instruction to enable when the current round reaches 30 rounds, writing the sound velocity into the sound velocity latch register when the latch instruction is enabled, writing the axial velocity into the flow velocity latch register, writing the first channel delay into the first delay latch register, writing the second channel delay into the second delay latch register, writing the current round into the decoupling round field, writing the residual decrease into the convergence judgment field, and writing the boundary state registration into the out-of-bounds flag field, thus obtaining the sound velocity flow velocity delay decoupling amount.

[0059] The bounded Levenberg-Marquardt algorithm first sets the sound velocity, axial flow velocity, first channel delay, and second channel delay as four parameters to be determined. An initial parameter set is generated based on the initial values ​​of the sound velocity, axial flow velocity, first reference delay, and second reference delay. This initial parameter set is then written into the first frequency downstream time equation, the first frequency upstream time equation, the second frequency downstream time equation, and the second frequency upstream time equation. Combined with the flue inner diameter, transducer installation angle, and dual-frequency sound path, four predicted propagation times are calculated. Finally, the readings are subtracted from the four predicted propagation times. The following time residuals are obtained: the downstream propagation time of the first frequency, the upstream propagation time of the first frequency, the downstream propagation time of the second frequency, and the upstream propagation time of the second frequency. Four time residuals are obtained: the first channel delay is subtracted from the first reference delay to obtain the first delay deviation; the second channel delay is subtracted from the second reference delay to obtain the second delay deviation; the four time residuals, the first delay deviation, and the second delay deviation are summed by square weights according to preset time weights and preset delay weights to obtain the residual deviation value for this round; based on the four time residuals, the first delay deviation, and the second delay deviation... The relationship between the delay deviation and the changes in the four parameters to be determined is used to generate a sensitivity matrix, and the parameter correction amount is obtained in combination with the damping coefficient. When the residual deviation value decreases after the parameter correction amount is written, the corrected parameter set is used and the damping coefficient is reduced. When the residual deviation value does not decrease after the parameter correction amount is written, the original parameter set is retained and the damping coefficient is increased before re-obtaining the parameter correction amount. After each round of parameter correction, the sound velocity is limited to the range from the lower limit to the upper limit of the sound velocity, the axial flow velocity is limited to the range from the lower limit to the upper limit of the flow velocity, and the delay of the first channel is... The second channel delay is limited to the range of the lower limit to the upper limit of the first delay. The difference between the residual deviation value of the current round and the residual deviation value of the previous round is recorded. When any of the following criteria is met: the difference is lower than the preset convergence threshold, the number of iterations reaches the preset upper limit, the sound velocity reaches the boundary, the axial flow velocity reaches the boundary, the first channel delay reaches the boundary, or the second channel delay reaches the boundary, the iteration stops, and the sound velocity, axial flow velocity, first channel delay, and second channel delay are latched to obtain the sound velocity-flow velocity delay decoupling amount.

[0060] The specific steps for generating the working condition path weight coefficient group are as follows:

[0061] Based on the decoupling of sound velocity and flow velocity delay and the dual-frequency propagation constraint group, a dual-frequency propagation field assembly method is adopted to call the following four propagation times: flue gas sound velocity, flue gas axial flow velocity, first channel delay, second channel delay, dual-frequency echo amplitude, and propagation time. The call is performed by reading the sound velocity latch register, flow velocity latch register, first delay latch register, second delay latch register, first frequency downstream echo amplitude field, first frequency upstream echo amplitude field, second frequency downstream echo amplitude field, second frequency upstream echo amplitude field, first frequency downstream propagation time field, first frequency upstream propagation time field, second frequency downstream propagation time field, and second frequency upstream propagation time field. The sound velocity unit is set to m / s, the flow velocity unit is set to m / s, the delay unit is set to μs, and the propagation time unit is set. The amplitude unit is set to mV, and dual-frequency propagation is recorded. The flow velocity at the first frequency and the flow velocity at the second frequency are calculated using the path-time inversion velocity lookup method. The calculation is performed by sending the first frequency downstream propagation time, the first frequency upstream propagation time, the flue gas velocity, the first channel delay, and the first frequency path length into the first frequency inversion lookup area. The velocity step size of the first frequency inversion lookup area is set to 0.01 m / s, and the velocity range of the first frequency inversion lookup area is set to -40 m / s to 40 m / s. The minimum difference between the table entry time difference and the measured time difference is recorded as the first frequency flow velocity. The second frequency downstream propagation time, the second frequency upstream propagation time, the flue gas velocity, the second channel delay, and the second frequency path length are sent into the second frequency inversion lookup area. The velocity step size of the second frequency inversion lookup area is set to 0.0.1 m / s, set the flow velocity range of the second frequency inversion lookup area to -40 m / s to 40 m / s, record the minimum difference between the table entry time difference and the measured time difference as the second frequency flow velocity, and perform dual-frequency flow velocity registration. Use the absolute difference field calculation method to calculate the dual-frequency flow velocity difference and dual-frequency sound velocity difference. Calculate by reading the first frequency flow velocity, second frequency flow velocity, flue gas sound velocity, and dual-frequency sound velocity reference item. Take the absolute value of the difference between the first frequency flow velocity and the second frequency flow velocity and record it as the dual-frequency flow velocity difference. Take the absolute value of the difference between the flue gas sound velocity and the dual-frequency sound velocity reference item and record it as the dual-frequency sound velocity difference. Use the dual amplitude quotient registration method to calculate the amplitude ratio. Calculate by reading the first frequency downstream echo amplitude, the first frequency upstream echo amplitude, the second frequency downstream echo amplitude, and the second frequency upstream echo amplitude. Record the average of the two echo amplitudes of the first frequency as the first amplitude reference. Record the average of the two echo amplitudes of the second frequency as the second amplitude reference. The quotient of the first amplitude reference and the second amplitude reference is recorded as the amplitude ratio. The upstream and downstream time difference offset registration method is used to calculate the upstream and downstream time difference offset. This is achieved by reading the upstream and downstream time difference of the first frequency, the upstream and downstream time difference of the second frequency, the delay of the first channel, and the delay of the second channel. The absolute value of the difference between the two frequency upstream and downstream time differences is taken, and the difference in delay between the two channels is added to the offset registration field and recorded as the upstream and downstream time difference offset. The flow velocity sign consistency determination method is used to calculate the flow velocity sign consistency value. This is achieved by reading the first frequency flow velocity and the second frequency flow velocity. When both the first and second frequency flow velocities are positive, the flow velocity sign consistency value is written as one; when both the first and second frequency flow velocities are negative, the flow velocity sign consistency value is written as one; when the first and second frequency flow velocities have opposite signs, the flow velocity sign consistency value is written as zero; and when either flow velocity is zero, the flow velocity sign consistency value is written as two. This yields a set of flue gas condition discrimination quantities.

[0062] Based on the flue gas operating condition discrimination set, a discrete five-state hidden Markov model is used to determine the operating condition state sequence. The model is set by registering the state set as normal flow, high attenuation, water vapor enhancement, dust enhancement, and local recirculation. The initial state weights are set as follows: normal flow 0.60, high attenuation 0.10, water vapor enhancement 0.10, dust enhancement 0.10, and local recirculation 0.10. The state residence weight is set to 0.80, the adjacent state transition weight is set to 0.05, and cross-class... The state transition weight is set to 0.025, and state parameters are loaded. A threshold segment observation symbol encoding method is used to compare the dual-frequency velocity difference with the velocity difference threshold segment. This comparison is achieved by reading the dual-frequency velocity difference, setting a first velocity difference threshold of 0.20 m / s, a second velocity difference threshold of 0.80 m / s, and registering the velocity difference as a low segment when the dual-frequency velocity difference is less than or equal to 0.20 m / s, and registering the velocity difference as a low segment when the dual-frequency velocity difference is greater than 0.20 m / s and less than or equal to 0.80 m / s. The velocity difference is coded as follows: When the velocity difference between the two frequencies is greater than 0.80 m / s, it is registered as a high-frequency difference. A sound velocity difference threshold comparison method is used, comparing the velocity difference between the two frequencies with the specified thresholds. The comparison is achieved by reading the velocity difference between the two frequencies, setting the first threshold to 1.0 m / s, and the second threshold to 3.0 m / s. When the velocity difference between the two frequencies is less than or equal to 1.0 m / s, it is registered as a low-frequency difference; when the velocity difference between the two frequencies is greater than 1.0 m / s and less than or equal to 3.0 m / s, it is registered as a high-frequency difference. When the sound velocity difference is 3.0 m / s, it is registered as mid-range; when the sound velocity difference between the two frequencies is greater than 3.0 m / s, it is registered as high-range. An amplitude ratio limiting comparison method is used, comparing the amplitude ratio with its upper and lower limits. The comparison is achieved by reading the amplitude ratio, setting the lower limit to 0.65, and the upper limit to 1.35. When the amplitude ratio is less than 0.65, it is registered as low amplitude; when the amplitude ratio is between 0.65 and 1.35, it is registered as mid-amplitude; and when the amplitude ratio is greater than 1...At 35, the amplitude ratio is registered as high amplitude. A time offset threshold comparison method is used, comparing the upstream and downstream time difference offset with the time offset threshold. The comparison involves reading the upstream and downstream time difference offset, setting a first time offset threshold of 5μs, and a second time offset threshold of 15μs. When the upstream and downstream time difference offset is less than or equal to 5μs, the time offset is registered as low; when it is greater than 5μs but less than or equal to 15μs, it is registered as medium; and when it is greater than 15μs, it is registered as high. A return flow condition matching method is used, matching the flow velocity sign consistency value with the return flow determination condition. Matching is done by reading the flow velocity sign consistency value. When the flow velocity sign consistency value is zero and the time offset is high, a local return flow candidate is registered; when the flow velocity sign consistency value is one and the amplitude ratio is low... When the amplitude is high, high attenuation candidates are registered; when the sound velocity difference code is high and the amplitude ratio code is high, water vapor enhancement candidates are registered; when the velocity difference code is high and the amplitude ratio code is low, dust enhancement candidates are registered; when the velocity difference code is low, the sound velocity difference code is low, the amplitude ratio code is medium, the time offset code is low, and the velocity sign consistency is one, conventional flow candidates are registered. The Viterbi maximum path backtracking method is used to select conventional flow, high attenuation, water vapor enhancement, dust enhancement, or local backflow states. The selection process involves reading the current observation symbol code, reading the previous cycle state path, reading the five-state transition weights, reading the five-state observation matching weights, registering the scores of the five candidate paths, selecting the state with the highest score and writing it into the current cycle's operating condition state, writing the previous cycle path number and the current cycle state number into the path record, and obtaining the operating condition state path table.

[0063] Based on the operating condition path table, a state dwell time counting method is adopted to determine the operating condition. The current operating condition and the previous cycle's operating condition are read. This is achieved by retrieving the current cycle's operating condition status field, the previous cycle's operating condition status field, the current cycle's consecutive hit count field, the previous cycle's dwell time count field, and the path number field. When the current operating condition matches the previous cycle's operating condition, the consecutive hit count is incremented by one; when the current operating condition differs from the previous cycle's operating condition, the consecutive hit count is set to one; when the previous cycle's operating condition continues to appear, the dwell time count is incremented by one; when the previous cycle's operating condition is changed, the dwell time count is set to zero. The state count field is then recorded. A dwell time threshold gating method is used to compare the consecutive hit count with the dwell time threshold. The comparison was conducted by setting thresholds for consecutive hits of normal flow (2 times), high attenuation (3 times), water vapor enhancement (3 times), dust enhancement (3 times), local backflow (2 times), and state dwell time (2 times). When the consecutive hit count for a state did not reach the corresponding threshold, the coefficient field of the previous cycle was read; when the dwell time did not reach 2 times, the coefficient field of the previous cycle was read and reused. A five-state coefficient lookup table selection method was used. When the threshold was reached, the following parameters were selected for the current operating condition: first frequency confidence weight, second frequency confidence weight, propagation time compensation coefficient, sound velocity offset compensation coefficient, amplitude attenuation compensation coefficient, and flow velocity output damping coefficient. The parameters are selected by reading the operating condition coefficient table. When the current operating condition is normal flow, the following values ​​are written: first frequency confidence weight 0.50, second frequency confidence weight 0.50, propagation time compensation coefficient 0.00μs, sound velocity offset compensation coefficient 0.00m / s, amplitude attenuation compensation coefficient 1.00, flow velocity output damping coefficient 0.20. When the current operating condition is high attenuation, the following values ​​are written: first frequency confidence weight 0.65, second frequency confidence weight 0.35, propagation time compensation coefficient 1.50μs, sound velocity offset compensation coefficient 0.50m / s, amplitude attenuation compensation coefficient 1.30, flow velocity output damping coefficient 0.35. When the current operating condition is water vapor enhancement, the following values ​​are written: first frequency confidence weight 0.40, second frequency confidence weight 0.60. The propagation time compensation coefficient is 0.80 μs, the sound velocity offset compensation coefficient is 2.00 m / s, the amplitude attenuation compensation coefficient is 0.85, the flow velocity output damping coefficient is 0.30, and when the current operating condition is dust enhancement, the first frequency confidence weight is 0.70, the second frequency confidence weight is 0.30, the propagation time compensation coefficient is 1.20 μs, the sound velocity offset compensation coefficient is 1.00 m / s, the amplitude attenuation compensation coefficient is 1.50, the flow velocity output damping coefficient is 0.40, and when the current operating condition is local backflow, the first frequency confidence weight is 0.50, the second frequency confidence weight is 0.50, the propagation time compensation coefficient is 2.00 μs, the sound velocity offset compensation coefficient is 1.50 m / s, the amplitude attenuation compensation coefficient is 1.20, and the flow velocity output damping coefficient is 0.60. Write the selected coefficients into the current cycle coefficient field, the number of consecutive state hits into the hit count field, and the number of dwell times into the dwell count field, and obtain the working condition path weight coefficient group.

[0064] The Hidden Markov Model (HMM) first defines five implicit operating conditions: conventional flow, high attenuation, enhanced water vapor, enhanced dust, and local recirculation. It then defines observation groups: dual-frequency velocity difference, dual-frequency sound velocity difference, amplitude ratio, upstream / downstream time difference offset, and velocity sign consistency. Model parameters are established according to initial state probability tables, state transition probability tables, and observation matching probability tables. Within a single discrimination period, dual-frequency velocity difference and dual-frequency sound velocity difference are segmented and marked within velocity difference threshold segments. Amplitude ratio is segmented using upper and lower limits. Upstream / downstream time difference offset is segmented within time offset threshold segments. Velocity sign consistency is marked using recirculation criteria. Based on the segmentation, interval, and consistency marks, the observation matching probability table is searched to generate observation matching scores corresponding to the five implicit operating conditions. The previous period's five conditions are then used to determine the matching scores. The state path score is multiplied by the state transition probability and then by the observation matching score of the current period to obtain the scores of the five candidate paths for the current period. The Viterbi decoding algorithm is used to sort the scores of the five candidate paths in descending order. The operating state corresponding to the first candidate path in the sorted order is selected as the operating state of the current period. The operating state of the previous period, the operating state of the current period, the state transition node, the number of consecutive hits, and the number of dwells are recorded. When multiple discrimination periods are input consecutively, threshold segment comparison, observation matching, state transition calculation, candidate path sorting, and state node recording are performed cycle by cycle to generate an operating state path table containing the period number, the operating state of the previous period, the operating state of the current period, the state transition node, the number of consecutive hits, the number of dwells, the first frequency confidence weight, the second frequency confidence weight, the propagation time compensation coefficient, the sound speed offset compensation coefficient, the amplitude attenuation compensation coefficient, and the flow velocity output damping coefficient.

[0065] The specific steps for generating flue gas velocity calibration values ​​are as follows:

[0066] Based on the working condition path weight coefficient group, sound velocity-flow velocity delay decoupling amount, and dual-frequency propagation constraint group, the LightGBM gradient boosting decision tree velocity regression method is adopted to perform calls on the first frequency reliable weight, second frequency reliable weight, flue gas axial velocity, four propagation times, and dual-channel delay. The calls are performed by reading the first frequency reliable weight field, second frequency reliable weight field, sound velocity latch register, velocity latch register, first delay latch register, second delay latch register, first frequency downstream propagation time field, first frequency upstream propagation time field, second frequency downstream propagation time field, and second frequency upstream propagation time field, with a set number of trees. With 80 trees, a maximum of 15 leaf nodes, a maximum tree depth of 4 layers, a learning rate of 0.05, a minimum of 20 leaf node samples, a sample sampling ratio of 1.00, a column sampling ratio of 1.00, and absolute error as the loss term, the input fields were set in the following order: first frequency confidence weight, second frequency confidence weight, flue gas axial velocity, first frequency downstream propagation time, first frequency upstream propagation time, second frequency downstream propagation time, second frequency upstream propagation time, second frequency upstream propagation time, first channel delay, and second channel delay. A dual-frequency velocity split-path regression method was used to calculate the first and second frequency velocities separately. The first frequency confidence weight, flue gas axial velocity, first frequency downstream propagation time, first frequency upstream propagation time, and first channel delay are written into the input buffer of the first frequency tree model. Leaf node output values ​​are read sequentially from 1 to 80 according to the 80 tree numbers. These 80 leaf node output values ​​are accumulated and registered as the first frequency velocity using a learning rate of 0.05. The second frequency confidence weight, flue gas axial velocity, second frequency downstream propagation time, second frequency upstream propagation time, and second channel delay are written into the input buffer of the second frequency tree model. Leaf node output values ​​are read sequentially from 1 to 80 according to the 80 tree numbers. These 80 leaf node output values ​​are accumulated and registered using a learning rate of 0.05. The second frequency flow rate is used, and dual-frequency flow rate registration is performed. The trusted weight multiplication accumulation method is adopted. The first frequency flow rate is multiplied by the first frequency trusted weight, and the second frequency flow rate is multiplied by the second frequency trusted weight. The calculation is performed by reading the first frequency flow rate field, reading the second frequency flow rate field, reading the first frequency trusted weight field, reading the second frequency trusted weight field, writing the product of the first frequency flow rate and the first frequency trusted weight into the first weighted flow rate field, writing the product of the second frequency flow rate and the second frequency trusted weight into the second weighted flow rate field, and accumulating the first weighted flow rate field and the second weighted flow rate field into the dual-frequency weighted flow rate field to obtain the dual-frequency weighted flow rate.

[0067] Based on dual-frequency weighted flow velocity, a three-term compensation coefficient directional mapping method is employed to call the propagation time compensation coefficient, sound velocity offset compensation coefficient, amplitude attenuation compensation coefficient, dual-frequency upstream and downstream time difference offset, dual-frequency sound velocity difference, and amplitude ratio deviation. This is achieved by reading the propagation time compensation coefficient field, sound velocity offset compensation coefficient field, amplitude attenuation compensation coefficient field, upstream and downstream time difference offset field, dual-frequency sound velocity difference field, and amplitude ratio field. The amplitude ratio baseline value is set to 1.00, and the difference between the amplitude ratio and 1.00 is calculated. The absolute value is registered as the amplitude ratio deviation, the time offset unit is set to μs, the sound velocity difference unit is set to m per second, and the amplitude ratio deviation unit is set to dimensionless. Three offsets are registered, and the compensation product registration method is used. The three compensation coefficients are multiplied by their corresponding offsets to calculate the time compensation, sound velocity compensation, and amplitude compensation. The calculation is performed by multiplying the propagation time compensation coefficient by the dual-frequency forward and reverse flow time difference offset and writing it into the time compensation field, and multiplying the sound velocity offset compensation coefficient by the dual-frequency sound velocity difference and writing it into the sound velocity field. The compensation amount field is configured as follows: multiply the amplitude attenuation compensation coefficient by the amplitude ratio deviation and write the result to the amplitude compensation amount field; set the lower limit of the time compensation amount to -3.00 m / s; set the upper limit of the time compensation amount to 3.00 m / s; set the lower limit of the sound velocity compensation amount to -2.00 m / s; set the upper limit of the sound velocity compensation amount to 2.00 m / s; set the lower limit of the amplitude compensation amount to -1.50 m / s; set the upper limit of the amplitude compensation amount to 1.50 m / s; when the time compensation amount is small... When the time compensation is greater than 3.00 m / s, it is rewritten as -3.00 m / s; when the sound velocity compensation is less than -2.00 m / s, it is rewritten as -2.00 m / s; when the sound velocity compensation is greater than 2.00 m / s, it is rewritten as 2.00 m / s; when the amplitude compensation is less than -1.50 m / s, it is rewritten as -1.50 m / s; when the amplitude compensation is greater than 1.50 m / s, it is rewritten as 1.50 m / s, and the weighted compensated velocity group is obtained.

[0068] Based on a weighted compensated velocity group, a damped weighted output latching method is used to retrieve the dual-frequency weighted velocity, time compensation, sound velocity compensation, amplitude compensation, velocity output damping coefficient, and the velocity output value of the previous cycle. This retrieval is achieved by reading the dual-frequency weighted velocity field, time compensation field, sound velocity compensation field, amplitude compensation field, velocity output damping coefficient field, and the velocity output field of the previous cycle. The velocity output unit is set to m / s, the number of decimal places is set to three, the lower limit of the output amplitude is set to -40.000 m / s, and the upper limit of the output amplitude is set to 40.000 m / s. Output parameters are registered. A compensated velocity accumulation method is used, adding the dual-frequency weighted velocity to the three compensation values ​​to generate the compensated velocity. Calculation is performed by writing the dual-frequency weighted velocity to the initial compensated velocity field, adding the time compensation to the initial compensated velocity field, adding the sound velocity compensation to the initial compensated velocity field, and adding the amplitude compensation to the initial compensated velocity field. The process involves inputting the initial compensation velocity field, writing the cumulative value into the compensation velocity field, and using a first-order damping weighted smoothing method to weight the compensation velocity and the previous cycle's velocity output value according to the velocity output damping coefficient. The calculation is performed by reading the velocity output damping coefficient, reading the compensation velocity field, reading the previous cycle's velocity output field, writing the compensation velocity into the current cycle's weight item according to the corresponding proportion of the velocity output damping coefficient, writing the previous cycle's velocity output value into the previous cycle's weight item according to the remaining proportion, accumulating the current cycle's weight item and the previous cycle's weight item into the initial calibration velocity, rewriting the initial calibration velocity to -40.000 m / s when it is less than -40.000 m / s, and rewriting it to 40.000 m / s when it is greater than 40.000 m / s, writing the rewritten value into the velocity output latch register, writing the current cycle number into the calibration cycle field, and writing the velocity output damping coefficient into the damping record field, thus obtaining the flue gas velocity calibration value.

[0069] The gradient boosting decision tree first sets up input variables including the first frequency confidence weight, the second frequency confidence weight, the flue gas axial velocity, the first frequency downstream propagation time, the first frequency upstream propagation time, the second frequency downstream propagation time, the second frequency upstream propagation time, the first channel delay, and the second channel delay. The difference between the calibrated velocity and the initial velocity output is set as the training residual. In the first round of training, regression tree nodes are generated using the input variable set. Each input variable is compared with its corresponding splitting threshold, and the first splitting threshold based on the loss reduction is used as the node splitting threshold, generating left and right branch sample sets. The mean of the leaf node residuals is calculated and used as the leaf node output. The leaf node output is multiplied by the learning rate and then added to the initial velocity output to obtain the predicted velocity for this round. After each round of training, the calibrated velocity is subtracted from the predicted velocity for this round to generate the training residual for the next round. The process continues to build the next regression tree until any one of the following criteria is met: the number of trees reaches a preset number, the residual decrease is lower than a preset threshold, or the number of samples in a single leaf node is lower than a preset lower limit. Multiple regression trees, node splitting thresholds, leaf node outputs, and learning rates are latched. During online calculation, the current input variable set is sequentially written into multiple regression trees. Based on the node splitting threshold, the process enters the left branch sample interval and right branch sample interval layer by layer. The outputs of the corresponding leaf nodes of multiple regression trees are read and accumulated according to the learning rate to obtain the flow velocity calibration deviation. Then, the flow velocity at the first frequency, the flow velocity at the second frequency, the confidence weight of the first frequency, and the confidence weight of the second frequency are combined to generate a dual-frequency weighted flow velocity. The time compensation, sound velocity compensation, and amplitude compensation are then superimposed to generate a compensated flow velocity. Subsequently, the compensated flow velocity is weighted and fused with the flow velocity output value of the previous cycle according to the flow velocity output damping coefficient to obtain the flue flow velocity calibration value.

[0070] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A dual-frequency calibrated ultrasonic flow velocity detection method, characterized in that, Includes the following steps: S1: Based on the working configuration parameters of the flue ultrasonic flow velocity detection device, first make transducer A emit a downstream sound beam to transducer B, read the echo mean value frequency by frequency and compare it with the historical peak value, replace the peak value and switch to the upstream sound path, latch the frequency boundary frequency by frequency to obtain the bidirectional peak frequency boundary table. S2: Based on the bidirectional peak frequency boundary table, write the downstream peak frequency into the first drive register area and the upstream peak frequency into the second drive register area. Calculate the sound path according to the flue inner diameter and installation angle, and compare the arrival time and amplitude of the first wave one by one to obtain the dual-frequency propagation constraint group. S3: Based on the dual-frequency propagation constraint group, the sound speed, axial flow velocity, first channel delay and second channel delay are substituted into the forward and reverse flow time equations. The Levenburg-Marquardt algorithm is used to calculate the propagation time residual and delay deviation round by round. The out-of-bounds values ​​are replaced with the sound speed, flow velocity and delay boundary values ​​to obtain the sound speed, flow velocity and delay decoupling amount. S4: Based on the sound velocity and flow velocity delay decoupling amount and dual-frequency propagation constraint group, a hidden Markov model is used to match the dual-frequency flow velocity difference, dual-frequency sound velocity difference, amplitude ratio, and upstream and downstream time difference offset item by item with the flue working condition. The dual-frequency reliable weight and calibration coefficient are selected according to the state results to obtain the working condition path weight coefficient group. S5: Based on the working condition path weight coefficient group, sound velocity flow rate delay decoupling amount and dual frequency propagation constraint group, a gradient boosting decision tree is used to multiply the first frequency flow velocity and the second frequency flow velocity by the credible weights respectively, and then superimpose the upstream and downstream time difference compensation, sound velocity offset compensation and amplitude deviation compensation to obtain the flue flow velocity calibration value.

2. The ultrasonic flow velocity detection method with dual-frequency calibration according to claim 1, characterized in that, The operating configuration parameters of the flue ultrasonic flow velocity detection device include: inlet end shut-off command, exhaust end shut-off command, transducer A transmitter port identifier, transducer B receiver port identifier, transducer B transmitter port identifier, transducer A receiver port identifier, sweep frequency lower limit, sweep frequency upper limit, downstream sweep frequency interval, upstream neighborhood sweep frequency interval, single frequency point echo reading count, echo reading interval, excitation voltage, pulse width, burst period, receiving gain, echo search window start point, echo search window length, downstream historical peak initial amplitude, upstream historical peak initial amplitude, downstream frequency boundary extension width, and frequency boundary latching field. The bidirectional peak frequency boundary table includes downstream peak frequency, upstream peak frequency, downstream scan lower limit, downstream scan upper limit, upstream scan lower limit, and upstream scan upper limit. The peak echo amplitude, the dual-frequency propagation constraint group includes the first frequency downstream propagation time, the first frequency upstream propagation time, the second frequency downstream propagation time, the second frequency upstream propagation time, the first frequency echo amplitude, and the second frequency echo amplitude; the sound velocity delay decoupling amount includes the flue gas sound velocity, the flue gas axial velocity, the first frequency channel delay, the second frequency channel delay, the first delay deviation, and the second delay deviation; the operating condition path weight coefficient group includes the flue operating condition state, the first frequency reliable weight, the second frequency reliable weight, the propagation time compensation coefficient, the sound velocity offset compensation coefficient, the amplitude attenuation compensation coefficient, and the flow velocity output damping coefficient; the flue flow velocity calibration value specifically includes the flue axial flow velocity value, the time compensation amount, the sound velocity compensation amount, the amplitude compensation amount, and the damped flow velocity output value.

3. The ultrasonic flow velocity detection method with dual-frequency calibration according to claim 1, characterized in that, The specific steps for generating the bidirectional peak frequency boundary table are as follows: Based on the working configuration parameters of the flue ultrasonic flow velocity detection device, the flue inlet and outlet are closed, the transmitter port of transducer A, the receiver port of transducer B, the lower limit of frequency sweep, the upper limit of frequency sweep, and the frequency sweep interval are set, and the average amplitude of five echoes is read frequency by frequency. The current average amplitude is compared with the historical peak value of the downstream flow to obtain the downstream peak frequency table. Based on the downstream peak frequency table, the transmitting port of transducer B and the receiving port of transducer A are switched, and the average amplitude of five echoes is read frequency by frequency along the neighborhood of the downstream frequency boundary. The current average amplitude is compared with the historical peak of the reverse flow, and the downstream peak frequency, the reverse peak frequency and the frequency boundary are merged to obtain the bidirectional peak frequency boundary table.

4. The ultrasonic flow velocity detection method with dual-frequency calibration according to claim 1, characterized in that, The specific steps for generating the dual-frequency propagation constraint group are as follows: Based on the bidirectional peak frequency boundary table, the downstream peak frequency is written into the first drive register area, and the upstream peak frequency is written into the second drive register area. The downstream sound path and the upstream sound path are calculated according to the flue inner diameter, transducer installation angle and sound beam oblique distance to obtain the dual-frequency sound path register table. Based on the dual-frequency sound path register, the first driving register frequency is called to trigger the downstream sound beam, and the second driving register frequency is called to trigger the upstream sound beam. The arrival time of the first wave, the echo amplitude, and the sound path number are read. The propagation time, echo amplitude, and sound path number are then grouped together to obtain the dual-frequency propagation constraint group.

5. The ultrasonic flow velocity detection method with dual-frequency calibration according to claim 1, characterized in that, The specific steps for generating the sound velocity-flow velocity delay decoupling amount are as follows: Based on the dual-frequency propagation constraint group, the first frequency downstream propagation time, the first frequency upstream propagation time, the second frequency downstream propagation time, and the second frequency upstream propagation time are read. The flue inner diameter, transducer installation angle, and dual-frequency sound path are retrieved. The sound velocity, axial flow velocity, first channel delay, and second channel delay are written into the four downstream and upstream time equations respectively to establish a propagation time equation group. Based on the propagation time equations, the Levenburg-Marquardt algorithm is used to calculate the time residuals of the first frequency downstream, the first frequency upstream, the second frequency downstream, and the second frequency upstream in turn. The delay of the first channel is subtracted from the delay of the first reference, and the delay of the second channel is subtracted from the delay of the second reference. The time residuals and delay deviations are accumulated according to their weights to obtain the residual deviation series. Based on the residual deviation sequence, the residual decrease, sound velocity boundary, flow velocity boundary, first delay boundary and second delay boundary of adjacent rounds are compared. When the sound velocity, flow velocity or delay exceeds the boundary, it is rewritten to the corresponding boundary value. When the residual decrease is lower than the threshold or the round reaches the upper limit, the sound velocity, axial flow velocity and dual-channel delay are latched to obtain the sound velocity, flow velocity and delay decoupling amount.

6. The ultrasonic flow velocity detection method with dual-frequency calibration according to claim 5, characterized in that, The bounded Levenberg-Marquardt algorithm first sets the sound velocity, axial flow velocity, first channel delay, and second channel delay as four parameters to be determined. An initial parameter set is generated based on the initial values ​​of the sound velocity, axial flow velocity, first reference delay, and second reference delay. This initial parameter set is then written into the first frequency downstream time equation, the first frequency upstream time equation, the second frequency downstream time equation, and the second frequency upstream time equation. Combined with the flue inner diameter, transducer installation angle, and dual-frequency sound path, four predicted propagation times are calculated. Finally, the first frequency downstream propagation time, the first frequency upstream propagation time, the second frequency downstream propagation time, and the second frequency upstream propagation time are subtracted from the four predicted propagation times to obtain four time residuals. The first delay deviation is obtained by subtracting the first reference delay from the first channel delay, and the second delay deviation is obtained by subtracting the second reference delay from the second channel delay. The four time residuals, the first delay deviation, and the second delay deviation are then summed using a weighted square based on preset time and delay weights to obtain the current round residual deviation value. A sensitivity matrix is ​​generated based on the correspondence between the four time residuals, the first delay deviation, and the second delay deviation and the changes in the four parameters to be determined, and the parameter correction is calculated using the damping coefficient. When the residual deviation value decreases after the parameter correction is written, the corrected parameter set is used and the damping coefficient is reduced. When the residual deviation value does not decrease after the parameter correction is written, the original parameter set is retained, the damping coefficient is increased, and the parameter correction is calculated again. After each round of parameter correction, the sound velocity is limited to the range from the lower limit to the upper limit, the axial flow velocity is limited to the range from the lower limit to the upper limit, the first channel delay is limited to the range from the lower limit to the upper limit, and the second channel delay is limited to the range from the lower limit to the upper limit. The difference between the residual deviation value of this round and the residual deviation value of the previous round is recorded. When any of the following criteria is met: the difference is lower than the preset convergence threshold, the number of iterations reaches the preset upper limit, the sound velocity reaches the boundary, the axial flow velocity reaches the boundary, the first channel delay reaches the boundary, or the second channel delay reaches the boundary, the iteration stops, and the sound velocity, axial flow velocity, first channel delay, and second channel delay are latched to obtain the sound velocity-flow velocity-delay decoupling amount.

7. The ultrasonic flow velocity detection method with dual-frequency calibration according to claim 1, characterized in that, The specific steps for generating the working condition path weight coefficient group are as follows: Based on the aforementioned sound velocity and flow velocity delay decoupling quantity and dual-frequency propagation constraint group, the flue gas sound velocity, flue gas axial flow velocity, first channel delay, second channel delay, dual-frequency echo amplitude, and four propagation times are called to calculate the first frequency flow velocity, second frequency flow velocity, dual-frequency flow velocity difference, dual-frequency sound velocity difference, amplitude ratio, upstream and downstream time difference offset, and flow velocity sign consistency quantity, thus obtaining the flue gas duct condition discrimination quantity group; Based on the flue gas condition discrimination set, a hidden Markov model is used to compare the dual-frequency velocity difference with the velocity difference threshold, the dual-frequency sound velocity difference with the sound velocity difference threshold, the amplitude ratio with the upper and lower limits of the amplitude ratio, the forward and reverse flow time difference offset with the time offset threshold, and the velocity sign consistency quantity is matched with the backflow judgment condition. The normal flow, high attenuation, water vapor enhancement, dust enhancement, or local backflow state is selected to obtain the condition state path table. Based on the operating condition path table, the current operating condition and the previous cycle operating condition are read, and the number of consecutive hits and dwell times of the state are compared. If the threshold is not reached, the coefficient of the previous cycle is used. If the threshold is reached, the first frequency confidence weight, the second frequency confidence weight, the propagation time compensation coefficient, the sound speed offset compensation coefficient, the amplitude attenuation compensation coefficient, and the flow velocity output damping coefficient corresponding to the current operating condition are selected to obtain the operating condition path weight coefficient group.

8. The ultrasonic flow velocity detection method with dual-frequency calibration according to claim 7, characterized in that, The Hidden Markov Model (HMM) first defines five implicit operating conditions: conventional flow, high attenuation, water vapor enhancement, dust enhancement, and local backflow. It then defines observation groups: dual-frequency velocity difference, dual-frequency sound velocity difference, amplitude ratio, upstream / downstream time difference offset, and velocity sign consistency. Model parameters are established according to the initial state probability table, state transition probability table, and observation matching probability table. Within a single discrimination period, the dual-frequency velocity difference and dual-frequency sound velocity difference are segmented and marked within a velocity difference threshold range. The amplitude ratio is segmented using upper and lower limits. The upstream / downstream time difference offset is segmented within a time offset threshold range. The velocity sign consistency is marked using the backflow judgment condition. Based on the segmented, interval, and consistency marks, the observation matching probability table is searched to generate the observation matching scores corresponding to the five implicit operating conditions. The previous period's five... The path scores for each state are multiplied by the state transition probability and then by the observation matching score for the current period to obtain the scores for the five candidate paths for the current period. The Viterbi decoding algorithm is used to sort the scores of the five candidate paths in descending order. The operating state corresponding to the first candidate path in the sorted order is selected as the operating state for the current period. The operating state of the previous period, the operating state of the current period, the state transition node, the number of consecutive hits, and the number of dwells are recorded. When multiple discrimination periods are input consecutively, threshold segment comparison, observation matching, state transition calculation, candidate path sorting, and state node recording are performed cycle by cycle to generate an operating state path table containing the period number, the operating state of the previous period, the operating state of the current period, the state transition node, the number of consecutive hits, the number of dwells, the first frequency confidence weight, the second frequency confidence weight, the propagation time compensation coefficient, the sound speed offset compensation coefficient, the amplitude attenuation compensation coefficient, and the flow velocity output damping coefficient.

9. The ultrasonic flow velocity detection method with dual-frequency calibration according to claim 1, characterized in that, The specific steps for generating the flue gas velocity calibration value are as follows: Based on the aforementioned working condition path weight coefficient group, sound velocity flow rate delay decoupling amount, and dual-frequency propagation constraint group, a gradient boosting decision tree is adopted. The first frequency confidence weight, the second frequency confidence weight, flue gas axial velocity, four propagation times, and dual-channel delay are called to calculate the first frequency velocity and the second frequency velocity respectively. The first frequency velocity is multiplied by the first frequency confidence weight, and the second frequency velocity is multiplied by the second frequency confidence weight to obtain the dual-frequency weighted velocity amount. Based on the aforementioned dual-frequency weighted flow velocity, the propagation time compensation coefficient, sound velocity offset compensation coefficient, amplitude attenuation compensation coefficient, dual-frequency upstream and downstream time difference offset, dual-frequency sound velocity difference and amplitude ratio deviation are called. The three compensation coefficients are multiplied by the corresponding offsets respectively to calculate the time compensation, sound velocity compensation and amplitude compensation, and obtain the weighted compensated flow velocity group. Based on the weighted compensation velocity group, the dual-frequency weighted velocity, time compensation, sound velocity compensation, amplitude compensation, velocity output damping coefficient, and the velocity output value of the previous cycle are called. The dual-frequency weighted velocity is added to the three compensation values ​​to generate the compensation velocity. The compensation velocity and the velocity output value of the previous cycle are weighted according to the velocity output damping coefficient to obtain the flue gas velocity calibration value.

10. The dual-frequency calibrated ultrasonic flow velocity detection method according to claim 9, characterized in that, The gradient boosting decision tree first uses the following input variables: first frequency confidence weight, second frequency confidence weight, flue gas axial velocity, first frequency downstream propagation time, first frequency upstream propagation time, second frequency downstream propagation time, second frequency upstream propagation time, first channel delay, and second channel delay. The difference between the calibrated velocity and the initial velocity output is set as the training residual. In the first training round, regression tree nodes are generated using the input variable set. Each input variable is compared against its corresponding splitting threshold, and the first-ranked splitting threshold based on the loss reduction is used as the node splitting threshold. Left and right branch sample sets are generated. The mean of the leaf node residuals is calculated and used as the leaf node output. The leaf node output is multiplied by the learning rate and then added to the initial velocity output to obtain the predicted velocity for this round. After each training round, the calibrated velocity is subtracted from the predicted velocity to generate the training residual for the next round. The process continues to build the next regression tree until any one of the following criteria is met: the number of trees reaches the preset number, the residual decrease is lower than the preset threshold, or the number of samples in a single leaf node is lower than the preset lower limit. Multiple regression trees, node splitting thresholds, leaf node outputs, and learning rates are latched. During online calculation, the current input variable group is sequentially written into multiple regression trees. Based on the node splitting threshold, the process enters the left branch sample interval and the right branch sample interval layer by layer. The outputs of the corresponding leaf nodes of multiple regression trees are read and accumulated according to the learning rate to obtain the flow velocity calibration deviation. Then, the first frequency flow velocity, the second frequency flow velocity, the first frequency confidence weight, and the second frequency confidence weight are combined to generate a dual-frequency weighted flow velocity. The time compensation, sound velocity compensation, and amplitude compensation are superimposed to generate a compensated flow velocity. Subsequently, according to the flow velocity output damping coefficient, the compensated flow velocity is weighted and fused with the flow velocity output value of the previous cycle to obtain the flue flow velocity calibration value.

Images