Early warning method for blockage of pulverized coal pipeline of thermal power plant

Abstract

The invention discloses a thermal power plant pulverized coal pipeline blockage early warning method, relates to the technical field of electric power industry automation and state monitoring, and aims to automatically identify flow characteristic differences of different coal types, dynamically adjust an early warning threshold value and avoid false alarms caused by coal type changes. By adopting a mode of combining a generative adversarial network and computational fluid dynamics simulation, virtual monitoring of key areas such as a bent pipe is realized, and the limitation that monitoring can only depend on a straight pipe section in the past is broken through; the transient abnormality capturing capability is remarkably improved through a high-frequency sampling and cumulative sum algorithm, and blockage precursor features can be found earlier; through linkage of a grading early warning mechanism and disposal suggestions, the system not only can discover problems, but also can provide targeted solutions, the operation and maintenance efficiency is greatly improved, a software algorithm upgrading-based mode is adopted, existing hardware resources are utilized to the maximum extent, and the modification cost and the implementation difficulty are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power industry automation and condition monitoring, and particularly relates to a coal powder pipeline blockage early warning method for thermal power plants. BACKGROUND

[0002] At present, in order to adapt to the change of the coal market and reduce the fuel cost, the thermal power plants generally adopt the operation mode of mixed burning of multiple coal types; different coal types have differences in physical and chemical properties.

[0003] The existing coal powder pipeline blockage early warning system is mostly based on electrostatic sensing technology, which judges the pipeline state by monitoring the charge signals generated by the flow of coal powder; such system alarms according to the preset threshold, and has certain effect under the working condition of single coal type or stable coal quality; however, under the condition of mixed burning of multiple coal types, due to the significant difference in the charging characteristics of different coal types, the signal baseline drifts and the fluctuation characteristics change; the early warning mechanism with fixed threshold is difficult to adapt to this dynamic working condition, and may have early warning deviation.

[0004] In order to solve the above problems, some existing solutions adjust the threshold by artificial experience or increase the coal quality detection link to improve the early warning accuracy; such solutions need to rely on the experience judgment of the operation personnel, or introduce additional coal quality analysis equipment, which has adjustment lag and increases the system complexity in actual application; at the same time, due to the frequent change of the mixed burning ratio and the combination of coal types, the adaptability of such method is still insufficient. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] The present application provides a coal powder pipeline blockage early warning method for thermal power plants, which solves the problem that the existing fixed threshold early warning system has high false alarm rate and lacks adaptability due to the large difference in the characteristics of coal powder caused by mixed burning of multiple coal types.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application embodiment provides a coal powder pipeline blockage early warning method for thermal power plants, which comprises,

[0009] Step S1: collecting charge signals by an electrostatic sensor array arranged in the coal powder conveying pipeline;

[0010] Step S2: collecting normal flow signals of different coal types under the calibration working condition and extracting features to establish a coal type feature library;

[0011] Step S3: inputting the features of the real-time charge signals into the trained pattern recognition model to identify the dominant coal type being conveyed;

[0012] Step S4, calling corresponding reference characteristic parameters from the coal type characteristic library according to the recognition result;

[0013] Step S5, dynamically adjusting the early warning threshold according to the reference characteristic parameters;

[0014] Step S6, detecting the real-time charge signal based on the adjusted threshold value;

[0015] Step S7, when determining that it is abnormal, outputting graded early warning information containing early warning level and disposal suggestion.

[0016] As a preferred scheme of the coal powder pipeline blockage early warning method of the power plant, the electrostatic sensor array is at least three groups of electrode units arranged in a circular direction, and after the sampling data are synchronously collected, band-pass filtered, electrostatic drift compensated and time-aligned, time domain waveforms and frequency domain features obtained by fast Fourier transform are obtained.

[0017] As a preferred scheme of the coal powder pipeline blockage early warning method of the power plant, the pattern recognition model adopts wavelet packet decomposition when extracting features, selects energy distribution, spectral entropy and energy proportion of low frequency and high frequency subbands as coal type recognition basis, and trains the classifier offline through labeled data to output the dominant coal type.

[0018] As a preferred scheme of the coal powder pipeline blockage early warning method of the power plant, the dynamic adjustment of the early warning threshold includes:

[0019] Based on the difference in dielectric properties between the current coal type and the reference coal type and the amplitude and fluctuation statistics in the feature library, the threshold scaling coefficient and the upper and lower limits are calculated, and after the hysteresis and dead zone are suppressed to suppress the jitter, they are applied to the online threshold decision;

[0020] The step of calculating the threshold scaling coefficient and the upper and lower limits is:

[0021] Extracting the dielectric parameters of the current coal type and the reference coal type, the amplitude quantile and the fluctuation index;

[0022] The amplitude quantile takes the robust quantile of online observation, and the fluctuation index takes the robust dispersion in the sliding window;

[0023] Map the dielectric difference, amplitude quantile ratio and fluctuation ratio to a single scaling coefficient:

[0024]

[0025] Wherein, η t is the threshold scaling coefficient at time t, clip(·,a,b) represents truncating the input to the interval [a,b], is the mapping weight, de R is a dielectric difference metric q R is an amplitude quantile ratio v η is a fluctuation ratio min η max t is the discrete time index, and t is the lower and upper bound of the scaling coefficient, respectively.

[0026] In the formula:

[0027]

[0028] wherein ε r tan δ is the relative dielectric constant and loss tangent of the current coal type, (tan δ) ref are the corresponding parameters of the reference coal type, w ε w tan is a difference weighting and w ε +w tan =1; and D are the p quantiles of the current and reference normal flow amplitudes, respectively, D t is the current fluctuation index, and D ref is the reference fluctuation scalar;

[0029] On the basis of the reference quantile and the current fluctuation, dynamic upper and lower limits are generated and hysteresis is added:

[0030]

[0031] wherein L t and U t are the lower and upper thresholds at time t, are the two end quantiles of the reference amplitude, 0 t is a scaling coefficient, h≥0 is the hysteresis half-width, L min , L max , U min , U max is the engineering safety boundary; a dead zone δ0≥0 is added when making an online decision: only when the observed out-of-bound amplitude exceeds δ0 does the state flip.

[0032] As a preferred scheme of the coal powder pipeline blockage early warning method of the power plant, wherein: further comprising: using a generative adversarial network to train on normal flow data, calculating the similarity score of the real-time signal and the generated signal online, and combining the score with the threshold set based on the fixed false alarm rate for abnormality determination.

[0033] As a preferred scheme of the coal pipeline blockage early warning method for a thermal power plant, the method further comprises: obtaining a flow field of a bend region based on computational fluid dynamics simulation, aligning the flow field with a straight pipe section monitoring signal sample, training a convolutional long short-term memory network to establish a mapping relationship, and outputting a blockage state estimation of the bend region in operation and fusing the estimation with an abnormality detection result.

[0034] As a preferred scheme of the coal pipeline blockage early warning method for a thermal power plant, the method further comprises: obtaining a flow field of a bend region based on computational fluid dynamics simulation, aligning the flow field with a straight pipe section monitoring signal sample, training a convolutional long short-term memory network to establish a mapping relationship, and outputting a blockage state estimation of the bend region in operation and fusing the estimation with an abnormality detection result.

[0035] The step of performing single-sided or double-sided real-time accumulation and resetting comprises:

[0036] The step of performing single-sided or double-sided real-time accumulation and resetting comprises:

[0037] f hf =m s f b , m s ∈[5, 16],

[0038] wherein f hf is a high-frequency sampling frequency, in Hz, f b is a regular sampling frequency, in Hz, m s is a rate interval, is a recommended rate;

[0039] Zero drift compensation and scale unification are performed on the high-frequency sequence:

[0040]

[0041] wherein x t is a statistic quantity at a high-frequency time t, μ ref , σ ref is a reference position and scale under normal flow, and σ ref > 0;

[0042] An increment is constructed according to a designed detection amplitude and accumulated online, and an alarm is given and reset when the increment exceeds a threshold:

[0043]

[0044] Upper side alarm: Lower side alarm:

[0045] Reset after triggering: Or and freeze n locka high-frequency sample;

[0046] where s t is the log-likelihood increment for mean shift d0>0, which is designed to detect amplitude, are upper and lower CUSUM statistics, respectively, and A + >0 is the corresponding decision threshold, - is the number of frozen samples; one-sided is used when only amplitude surge is concerned, and two-sided is used when both surge and drop are concerned;

[0047] Under the standard normal distribution, based on the sequential ratio test approximation, the threshold and false alarm probability / ARL correspondence can be directly set:

[0048]

[0049] where α ± is the one-sided false alarm probability of each high-frequency sample, is the one-sided average alarm interval under the no-change condition, in terms of sample number; when both sides are effective at the same time, α is evenly divided between the two sides or allocated according to risk preference;

[0050] When m s increases, resulting in an increase in the correlation between samples, the equivalent threshold is relaxed:

[0051] The target ARL0 is modified to ARL0 / ρ eff ∈(0,1] by the effective independent sample ratio ρ eff ; ρ eff is obtained from the first-order autocorrelation estimate of the high-frequency sequence; the frozen time after the alarm is taken as T lock =n lock / f hf , covering the typical recovery time of a wind disturbance;

[0052] After the alarm, enter the frozen period, stop accumulation and only update the slow channel of μ ref , σ ref , and after the frozen period ends, resume accumulation; if a coal type switching event occurs within the window, refresh d0 and A ± for the coal type.

[0053] As a preferred scheme of the coal pipe blockage early warning method for a thermal power plant, the hierarchical early warning is classified according to abnormal duration, deviation degree, and spatial consistency, and outputs recommended disposal measures including adjusting the primary air volume, changing the speed of the coal feeder, discharging the sludge of the separator, and running at reduced load; the early warning information can be connected to the distributed control system through the communication interface.

[0054] ​In a second aspect, an embodiment of the present application provides a computer device comprising a memory and a processor, and the memory stores a computer program, wherein the computer program, when executed by the processor, implements any step of the method for early warning of coal pipe blockage in a thermal power plant according to the first aspect of the present application.

[0055] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the method for early warning of coal pipe blockage in a thermal power plant according to the first aspect of the present application.

[0056] The present application has the advantages that: the present application effectively solves the false alarm problem under the multi-coal blending combustion condition, can automatically identify the flow characteristic differences of different coal types, and dynamically adjusts the early warning threshold to avoid false alarms caused by changes in coal types; the combination of the generative adversarial network and the computational fluid dynamics simulation realizes virtual monitoring of key areas such as elbow pipes, and breaks through the limitation of relying only on straight pipe sections for monitoring in the past.

[0057] The present application significantly improves the transient anomaly capture capability through high-frequency sampling and cumulative sum algorithm, and can discover early signs of blockage. The linkage of the hierarchical early warning mechanism and the disposal suggestion enables the system not only to find problems, but also to provide targeted solutions, greatly improving the operation and maintenance efficiency. The software algorithm upgrade is mainly used to maximize the use of existing hardware resources, and reduce the modification cost and implementation difficulty.

[0058] The present application establishes a complete adaptive mechanism from coal type identification to threshold dynamic adjustment, so that the system can automatically optimize the operating parameters as the coal type changes, and truly realizes the autonomous learning and continuous improvement of the early warning system. This adaptive capability has important practical value for coal-fired power plants to cope with the complex and changeable coal market situation, and provides reliable technical support for the safe and stable operation of the power plant. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and should not be regarded as limiting the scope of the present application.

[0060] Figure 1 The flowchart of the method for early warning of coal pipe blockage in a thermal power plant in the embodiments. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solutions and advantages of the present application clearer, the following will further describe the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and should not be regarded as limiting the present application.

[0062] All terms used herein, including technical and scientific terms, have the meanings commonly understood by one of ordinary skill in the art unless otherwise defined. It should be noted that the use of certain terms herein, such as those listed in the following paragraphs, should not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0063] For example, the terms "first", "second", and the like as used herein are only used to distinguish between similar objects, to distinguish a first object from another object, and are not used to describe a specific order or sequence, nor can they be understood to indicate or imply relative importance.

[0064] The present application proposes a coal pipeline blockage early warning method for a thermal power plant, which combines Figure 1 As shown in the figure, the method comprises the following steps:

[0065] Step S1, collecting charge signals by an electrostatic sensor array arranged on the coal conveying pipeline;

[0066] In this embodiment, the electrostatic sensor array refers to a multi-electrode acquisition unit that is attached to the pipe wall and has negligible disturbance to the flow field in the pipe. In engineering, a charge amplifier + anti-saturation limiting is used as the basic channel, and shielding and single-point grounding are used to reduce common-mode interference. To ensure the time consistency of multiple channels, the same clock source or backplane bus trigger is used for synchronous acquisition to achieve alignment accuracy within ±1 sampling point across channels. The default conventional sampling frequency is 2 kHz, and the adjustable range is 1-5 kHz, which is set according to the existing device bandwidth and DCS throughput; the default single feature window is 1.0 s, and the adjustable range is 0.5-2.0 s, which is determined according to the compromise between abnormal response time scale and statistical stability. When there is a missing or single-channel saturation, interpolation and amplitude clipping of adjacent channels are used to maintain channel availability; if the synchronization error exceeds the preset upper limit, the feature calculation of this window is skipped and the sampling invalid state is reported.

[0067] Step S2, collecting normal flow signals of different coals under the calibration condition and extracting features to establish a coal characteristic library;

[0068] Specifically, the calibration condition refers to the stable interval of the primary air volume, the pulverized coal content, the moisture content of the incoming coal, and the unit load within the allowable range of the operation and maintenance regulations, with a duration of not less than 10 times the feature window to cover short-term fluctuations; at least 3 load segment samples are collected for each coal to reflect the robustness across loads. The robust quantile and robust dispersion in feature extraction refer to the exponential weighted quantile estimation and exponential weighted median absolute deviation index, respectively, to avoid the influence of extreme values. The default cumulative effective time for each coal is ≥30min, and the adjustable range is 15-120min, which is set according to the coal variation degree and statistical convergence speed. When the coal quality test parameters are incomplete, the dielectric parameter field is temporarily filled with online inversion estimation and marked with the source, so as to facilitate subsequent backfilling and correction.

[0069] Step S3, input the trained pattern recognition model based on the characteristics of the real-time charge signal to identify the dominant coal type of the current delivery;

[0070] For example, the dominant coal type refers to the coal type with the highest class confidence output by the classifier in the current window and exceeding the confidence threshold; when all confidence values are below the threshold, the output is uncertain and the last stable determination is maintained until the confidence of consecutive windows recovers. The confidence threshold is 0.6 by default and can be adjusted to 0.5-0.8, which is set according to the trade-off between precision and recall in offline cross-validation; the minimum holding time of stable determination is 20s by default and can be adjusted to 10-60s, which is valued according to the blending and deployment rhythm and the cost of false detection. Alternatively, when the number of categories increases, hierarchical determination is used to first coarsely classify coal types and then finely classify grades, so as to shorten the convergence time without changing the input and output definitions. The input features and training process of the classifier are consistent with those of the previous embodiments and are not changed.

[0071] Step S4, according to the identification result, the corresponding reference characteristic parameters are called from the coal type feature library;

[0072] Similarly, the reference characteristic parameters include amplitude quantile vectors, fluctuation degree scalars, and dielectric parameters; when calling, the records matching the current load segment are used first, and if there are none, the robust aggregate values of the same coal type are used. The refresh period is 30s by default and can be adjusted to 10-120s; when the identification result switches, it is refreshed once to reduce the time lag. If the feature library entry for this coal type is missing or too old, the online steady-state estimation within the last hour is used as a fallback in this embodiment and a temporary reference is marked.

[0073] Step S5, dynamically adjust the warning threshold according to the reference characteristic parameters;

[0074] Step S6, perform anomaly detection on the real-time charge signal based on the adjusted threshold;

[0075] Step S7, when the anomaly is determined, output the graded warning information containing the warning level and disposal suggestions;

[0076] In one embodiment, the electrostatic sensor array is at least three groups of electrode units arranged in a circular direction, and after synchronous acquisition, band-pass filtering, electrostatic drift compensation, and time alignment, time-domain waveforms and frequency-domain features obtained by fast Fourier transform are obtained;

[0077] Furthermore, the bandpass filter's passband defaults to covering the main frequency band carrying energy and particle collision information, with an initial value of 5-300Hz and an adjustable value of 3-500Hz, determined based on the experimental spectrum and mechanical resonance avoidance band. Electrostatic drift compensation is estimated through the slow channel and subtracted from the observations. The slow channel time constant defaults to 60-180s and is adjusted according to ambient temperature, humidity, and insulation conditions. 3-8 sets of ring electrodes meet the requirements for direction-insensitive acquisition. When the pipe diameter is large and installation is limited, a non-equidistant arrangement can be optionally used to maintain overlap in the field of view of adjacent electrodes, thus maintaining spatial consistency judgment capability. When the noise of a single channel exceeds the preset upper limit or DC saturation occurs, that channel participates in frequency domain statistics but is weighted less in spatial consistency calculations.

[0078] In one embodiment, the pattern recognition model uses wavelet packet decomposition when extracting features, selects the energy distribution, spectral entropy and energy ratio of low-frequency and high-frequency sub-bands as the basis for coal type identification, and trains a classifier offline through labeled data to output the dominant coal type.

[0079] In one embodiment, dynamically adjusting the warning threshold includes:

[0080] Based on the difference in dielectric properties between the current coal type and the reference coal type, as well as the amplitude and fluctuation statistics in the feature library, the threshold scaling factor and upper and lower limits are calculated, and then applied to online threshold decision-making after combining hysteresis and dead zone to suppress jitter.

[0081] The steps for calculating the threshold scaling factor and upper and lower limits are as follows:

[0082] Extract the dielectric parameters (relative permittivity and loss tangent), normal flow amplitude quantiles, and volatility indices of the current coal type and the reference coal type.

[0083] The amplitude quantile is taken from the robust quantile of online observation, and the volatility index is taken from the robust dispersion within the sliding window (such as the exponential weighted form of the median absolute deviation).

[0084] The dielectric difference, amplitude quantile, and fluctuation ratio are mapped to a single scaling factor to unify the stretching or compression threshold range:

[0085]

[0086] Where, η t Let be the threshold scaling factor at time t, and clip(·,a,b) denotes truncating the input to the interval [a,b]. For the mapping weights, d e R is a measure of dielectric difference. q R is the amplitude quantile ratio. v η is the volatility ratio. min ,η max The upper and lower bounds of the scaling factor are given, and t is the index of the discrete time step.

[0087] wherein:

[0088]

[0089] wherein, ε r is the relative permittivity and loss tangent of the current coal type, (tanδ) ref are the corresponding parameters of the reference coal type, w ε , w tan ∈ [0, 1] are the difference weights and w ε + w tan = 1; and D are the p-quantiles of the current and reference normal flow amplitudes, respectively, D t is the current fluctuation index, and D ref is the reference fluctuation scalar;

[0090] On the basis of the reference quantiles and the current fluctuation, dynamic upper and lower limits are generated and hysteresis is added:

[0091]

[0092] wherein, L t , U t are the lower and upper thresholds at time t, are the two end quantiles of the reference amplitude, 0 < p1 < p2 < 1, c > 0 is the fluctuation amplification coefficient, and η t is the scaling coefficient, h ≥ 0 is the hysteresis half-width, L min , L max , U min , U max are the engineering safety boundaries; a dead zone δ0 ≥ 0 is added when making online decisions: state inversion is triggered only when the observed out-of-bound amplitude exceeds δ0, which is used to suppress chattering;

[0093] The weights α, β, γ, the quantiles p, p1, p2, and the coefficients c, h, δ0 can be hierarchically configured according to coal type and load section, and fine-tuned online in small steps. When η t tends to the boundary for a long time, it is suggested to backtrack the calibration or update the reference quantiles and D ref ;

[0094] Optionally, the recommended value of the amplitude quantile is a combination of the median and high quantile to balance the sensitivity of position and tail; the volatility index uses the same robust scale measure as the previous embodiment, only for threshold width adaptation, without changing the definition of decision statistics. The default parameter values are: quantile pair (p1, p2) ≈ (0.5, 0.9), the hysteresis half-width h is adjusted according to the target false alarm rate by offline playback, and the initial value can be taken as 5-10% of the reference amplitude; the dead zone δ0 is equal to 0.5-1.0 times the current volatility by default. The adjustable range of parameters: p1∈[0.3, 0.6], p2∈[0.8, 0.95], h∈[0, 15% of the reference amplitude], δ0∈[0, 2×volatility]; the setting basis is the false alarm control of historical fault-free sections and the missed alarm constraint of manually labeled events. When the threshold scaling factor is close to the upper and lower clipping boundaries for a long time, record the edge forcing event and trigger the offline backtracking calibration prompt. In the abnormal boundary situation (missing or obviously distorted reference parameters), the online quantile and volatility of the latest stable window are replaced, and the parameter fine-tuning is temporarily frozen for no more than 5 minutes.

[0095] Specifically, the dielectric response difference, amplitude position and scale change are included in the same scaling framework here; the dielectric difference is used to reflect the statistical drift caused by coal replacement, the amplitude quantile ratio is used to capture the overall level of lifting or moving down, and the volatility ratio is used to give scale compensation for short-term disturbance and working condition fluctuation; the three form a single scaling factor in the same mapping, which is convenient for transferring the threshold between different coal types and load intervals; the upper and lower limits use the reference quantile as the benchmark, and are linearly expanded with the current volatility, and then the hysteresis and dead zone are superimposed to reduce the triggering frequency of the back-and-forth jitter; avoid rigid dependence on absolute amplitude, keep flexible response to distribution position and scale change in statistical sense; at the same time, through clipping and safety boundary limitation, the abnormal threshold is constrained in the range acceptable in engineering; parameter layering and online fine-tuning provide an evolution channel during operation, which is convenient for maintaining discriminant stability when there is long-term drift or gradual change in coal quality;

[0096] In one embodiment, it also includes: training on normal flow data using a generative adversarial network, calculating the similarity score of real-time signals and generated signals online, and using the score in combination with a threshold set based on a fixed false alarm rate for anomaly detection;

[0097] In this embodiment, the similarity score is used to measure the fitting degree of the real-time window and the normal distribution, and the threshold is obtained by calibrating the target false alarm rate on the non-anomaly playback data; the default target false alarm rate is 10 -5 level, adjustable 10 -4 -10 -6The training data covers at least three load sections and two typical pulverized coal content intervals. The effective sample length of a single coal type is recommended to be greater than or equal to 30 minutes. Alternatively, when the training sample is insufficient, the data augmentation method is used to simulate the amplitude and scale perturbations to supplement the boundary conditions, but the judgment caliber and output meaning are not changed. In the online stage, if the GAN score is temporarily unavailable, the system is degraded to threshold channel independent judgment.

[0098] In one embodiment, further comprising: based on the flow field of the elbow region obtained by the computational fluid dynamics simulation and the aligned sample of the straight pipe section monitoring signal, training a convolutional long short-term memory network to establish a mapping relationship, and outputting the accumulation state estimation of the elbow region in operation and fusing with the abnormal detection result;

[0099] Specifically, the aligned sample is realized by aligning the time base and conditioning the index with the working condition vector (air volume, pulverized coal content, load section), realizing one-to-one correspondence between the simulation frame and the measured window; the mapping model only outputs the probability or level of elbow accumulation, and the binary / multi-level results of abnormal detection are fused in the same time grid, without generating independent control instructions. The default fusion strategy is to take the high priority, and when the two conflict, the accumulation estimation is used to promote the warning level. In the boundary condition (CFD boundary condition deviation or out-of-domain working condition), the mapping output is automatically weighted to not higher than the level of a single threshold channel.

[0100] In one embodiment, the abnormal detection includes transient mutation detection based on cumulative sum, and the statistical quantity obtained by high-frequency sampling is accumulated and reset in real time on a single side or double side, and the high-frequency sampling frequency is not less than five times of the preset regular sampling frequency of the system;

[0101] Alternatively, the ratio of the regular sampling frequency and the high-frequency sampling frequency is initially set to 8, and is adjusted in the range of 5-16 according to the device bandwidth and CPU occupation; the detection amplitude and the decision threshold are set by setting the target average no-alarm sample length in the historical no-abnormal section, and then the target average no-alarm sample length is set, and then the target average no-alarm sample length is set. The alarm is not generated in the representative stable period. In order to suppress the repeated triggering caused by cluster fluctuations, the frozen sample strategy is used after triggering, and the frozen time length covers the typical recovery time of a wind disturbance, and the default is 50-200 ms. When the autocorrelation of the high-frequency sequence is significantly increased, the threshold is relaxed to maintain the target average no-alarm interval. If the high-frequency sampling channel is abnormal or the resources are insufficient, the system is degraded to single-sided detection of regular sampling and records the performance degradation flag.

[0102] The step of accumulating and resetting in real time on a single side or double side includes:

[0103] The regular sampling frequency is increased to high-frequency sampling, and the observation is standardized;

[0104] f hf = m s f b , m s ∈[5, 16],

[0105] where f hf is the high-frequency sampling frequency, Hz, f b is the regular sampling frequency, Hz, m s is the magnification interval, is the recommended magnification;

[0106] Zero drift compensation and scale unification are performed on the high-frequency sequence:

[0107]

[0108] where x t is the statistical quantity at high-frequency time t (the charge amplitude after band-pass or its robust envelope), μ ref , σ ref is the reference position and scale under normal flow (from the feature library of the identified coal type or the recent steady-state window), σ ref > 0;

[0109] The increment is constructed according to the designed detection amplitude and accumulated online, and the alarm is given and reset when the threshold is exceeded:

[0110]

[0111] Upper side alarm: Lower side alarm:

[0112] Reset after triggering: or and freeze n lock high-frequency samples;

[0113] where s t is the log-likelihood increment of the mean shift d0, d0 > 0 is the designed detection amplitude, are the upper and lower side CUSUM statistics, A + , A - > 0 is the corresponding decision threshold, is the number of frozen samples, used to suppress repeated reporting, typically covering 50-200 milliseconds of samples; when only the amplitude jump is concerned, one-sided is used, only when both the jump and the drop are concerned, two-sided is used, and

[0114] Under the standard normal distribution, based on the sequential ratio test approximation, the threshold and false alarm probability / ARL correspondence can be directly set:

[0115]

[0116] where α ±The one-sided false alarm probability for each high-frequency sample, The one-sided average alarm interval in the no-change situation, in terms of sample number; for example, the target (about equivalent to α + ≈10 -5 ) is taken as A + ≈11.51; when both sides are effective at the same time, α is divided equally between the two sides or allocated according to risk preference;

[0117] When m s increases, the correlation between samples increases, and the equivalent threshold is relaxed:

[0118] The target ARL0 is modified to ARL0 / ρ eff by the effective independent sample ratio ρ eff ∈(0, 1]; ρ eff is obtained from the first-order autocorrelation estimation of the high-frequency sequence; the freeze time after the alarm is taken as T lock =n lock / f hf , covering the typical recovery time of a wind disturbance;

[0119] After the alarm, enter the freeze period, stop accumulation and only update the slow channels of μ ref and σ ref , and resume accumulation after the freeze period ends; if a coal type switching event occurs within the window, refresh d0 and A ± for the coal type;

[0120] Specifically, this combines high-frequency sampling with sequential detection, uses a standardized statistic to construct a log-likelihood increment, and performs one-sided or two-sided accumulation online; when the threshold is reached, it is reported immediately; the increment construction, which takes the designed detection amplitude as the core, makes the sensitivity of the statistic adjustable to the target amplitude, while avoiding dependence on unknown scales; the reset uses the zeroing and freezing method, and the freeze period blocks short-time back-and-forth triggering, which is suitable for scenarios where the charge signal presents cluster fluctuations in a short time;

[0121] The correspondence between the threshold and the false alarm is approximately given by the sequential ratio test to give a closed expression, which can be directly set by the target average fault-free sample length in engineering, making it easy to control the alarm frequency within the acceptable range of operation and maintenance; when the coal type is switched and the load fluctuates, the reference parameters are refreshed, and the designed amplitude and threshold are configured in layers to maintain stable triggering characteristics;

[0122] In one embodiment, the hierarchical early warning is classified according to the duration of the anomaly, the degree of deviation, and the spatial consistency, and outputs recommended treatment measures including adjusting the primary air volume, changing the speed of the coal feeder, separating the slag of the separator, and running at reduced load. The early warning information can be connected to the distributed control system through the communication interface;

[0123] Similarly, the hierarchical caliber is determined in three dimensions of duration threshold, deviation amplitude threshold and multi-channel consistency threshold. The default duration threshold is 0.5s and 2s, the deviation amplitude is divided into medium / high according to the out-of-threshold super-limit ratio, and the spatial consistency is determined by more than half of the annular channels. The specific threshold can be fine-tuned according to the characteristics of the unit and the operation and maintenance experience. The early warning information includes timestamp, level, recommended measures and source channel summary, and the communication interface follows the existing protocol on the site. When the communication is unavailable, the local record and sound-light prompt are provided in this embodiment, and the message is sent after the link is restored.

[0124] The embodiment also provides a computer device, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the coal pipe blockage early warning method of the thermal power plant.

[0125] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, touchpad or mouse, etc.

[0126] The embodiment also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement a coal pipeline blockage early warning method for a thermal power plant as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or nonvolatile storage devices or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0127] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the above embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

[0128] In addition, those skilled in the art can understand that although some embodiments herein include certain features rather than other features included in other embodiments, the combination of features of different embodiments means to be within the scope of the present application and form different embodiments. For example, all the above embodiments can be used in any combination. The information disclosed in the background section is only intended to deepen the understanding of the overall background of the present application, and should not be regarded as acknowledging or implying in any form that the information constitutes the prior art known to those skilled in the art.

Claims

1. A method for early warning of blockage in pulverized coal pipelines of thermal power plants, characterized in that, Includes the following steps: Step S1: The charge signal is collected by an electrostatic sensor array installed in the pulverized coal conveying pipeline; Step S2: Under calibrated operating conditions, normal flow signals of different coal types are collected and features are extracted to establish a coal type feature library; Step S3: Based on the feature input of the real-time charge signal, the trained pattern recognition model is used to identify the dominant coal type being transported. Step S4: Retrieve the corresponding reference feature parameters from the coal type feature library according to the identification results; Step S5: Dynamically adjust the warning threshold based on the reference feature parameters; Step S6: Perform anomaly detection on the real-time charge signal based on the adjusted threshold; Step S7: When an anomaly is determined, output a graded warning message that includes the warning level and handling suggestions.

2. The method for early warning of blockage in pulverized coal pipelines of thermal power plants as described in claim 1, characterized in that, The electrostatic sensor array consists of at least three sets of electrode units arranged in a ring along the circumference. After synchronous acquisition, bandpass filtering, electrostatic drift compensation, and time alignment, the sampled data is used to obtain the time-domain waveform and the frequency-domain characteristics obtained by fast Fourier transform.

3. The method for early warning of pulverized coal pipeline blockage in a thermal power plant as described in claim 1, characterized in that, The pattern recognition model uses wavelet packet decomposition to extract features, selects the energy distribution, spectral entropy and energy ratio of low-frequency and high-frequency sub-bands as the basis for coal type identification, and trains a classifier offline through labeled data to output the dominant coal type.

4. The method for early warning of blockage in pulverized coal pipelines of thermal power plants as described in claim 1, characterized in that, The dynamically adjusted early warning threshold includes: Based on the difference in dielectric properties between the current coal type and the reference coal type, as well as the amplitude and fluctuation statistics in the feature library, the threshold scaling factor and upper and lower limits are calculated, and then applied to online threshold decision-making after combining hysteresis and dead zone to suppress jitter. The steps for calculating the threshold scaling factor and upper and lower limits are as follows: Extract the dielectric parameters, normal flow amplitude quantiles, and volatility indices of the current coal type and the reference coal type; The amplitude quantiles are taken from the robust quantiles observed online, and the volatility index is taken from the robust dispersion within the sliding window; Mapping dielectric difference, amplitude quantile, and fluctuation ratio to a single scaling factor: Where, η t Let be the threshold scaling factor at time t, and clip(·,a,b) denotes truncating the input to the interval [a,b]. For the mapping weights, d e R is a measure of dielectric difference. q R is the amplitude quantile ratio. v η is the volatility ratio. min η max The upper and lower bounds of the scaling factor are given, and t is the index of the discrete time step. In the formula: Where, ε r tanδ represents the relative permittivity and loss tangent of the current coal type. (tanδ) ref For reference parameters corresponding to different coal types, w ε w tan ∈[0,1] difference weighted and w ε +w tan =1; and These are the p-quantiles of the current and reference normal flow amplitudes, respectively, and D. t D is the current volatility indicator. ref Used as a reference volatility scalar; Based on the reference quantile and the current volatility, dynamic upper and lower limits are generated and hysteresis is added: Among them, L t U t Let be the lower and upper thresholds at time t. As the two quantiles of the reference amplitude, 0 <p1<p2<1,c> 0 represents the volatility amplification factor, η t The scaling factor is h ≥ 0, where h is the hysteresis half-width and L is the scaling factor. min ,L max U min U max For engineering safety boundaries; a dead zone δ0≥0 is added during online decision-making: the state flip is only triggered when the observation exceeds the boundary by more than δ0.

5. The method for early warning of pulverized coal pipeline blockage in a thermal power plant as described in claim 1, characterized in that, Also includes: Generative adversarial networks are trained on normal flowing data to calculate the similarity score between real-time signals and generated signals online. The score is then combined with a threshold set based on a fixed false alarm rate for anomaly detection.

6. The method for early warning of blockage in pulverized coal pipelines of thermal power plants as described in claim 1, characterized in that, Also includes: Based on the flow field in the bend region obtained from computational fluid dynamics simulation and the aligned samples of the monitoring signals of the straight pipe section, a convolutional long short-term memory network is trained to establish a mapping relationship. During operation, the estimated accumulation state of the bend region is output and fused with the anomaly detection results.

7. The method for early warning of blockage in pulverized coal pipelines of thermal power plants as described in claim 1, characterized in that, The anomaly detection includes transient mutation detection based on cumulative summation, which performs single / double-sided real-time accumulation and reset of statistics obtained through high-frequency sampling, wherein the high-frequency sampling frequency is not less than five times the system's preset conventional sampling frequency; The steps for performing single / dual-sided real-time accumulation and reset include: The sampling frequency was increased to a higher frequency based on the conventional sampling frequency, and the observations were standardized. f hf =m s f b ,m s ∈[5,16], Among them, f hf This is the high-frequency sampling frequency, measured in Hz, f. b This is the standard sampling frequency, measured in Hz (m). s The range is the multiplier. Recommended multiplier; Zero drift compensation and scaling for high-frequency sequences: Where, x t μ is a statistic for high-frequency time t. ref , σ ref σ represents the reference position and scale under normal flow conditions. ref >0; The incremental detection range is constructed according to the design detection range and accumulated online. An alarm is triggered and the system is reset when the threshold is exceeded. Upper alarm: Lower side alarm: Reset after triggering: or And freeze n lock One high-frequency sample; Among them, s t This represents the log-likelihood increment of shifting the mean upward by d0, where d0 > 0 represents the designed detection amplitude. These are the upper and lower CUSUM statistics, A. + A - >0 represents the corresponding decision threshold. To freeze the sample size; use one-sided when only focusing on sudden increases in amplitude, and use two-sided when focusing on both sudden increases and sudden decreases; Under the condition of no change in standard normality, the relationship between the threshold and the false alarm probability / ARL can be directly set based on the ordinal ratio test approximation: Where, α ± The one-sided false alarm probability for each high-frequency sample. The average alarm interval for one side under unchanged conditions is calculated based on the number of samples; when both sides are effective simultaneously, α is divided equally between the two sides or allocated according to risk preference. When m s When increasing the threshold leads to an increase in the correlation between samples, it is equivalent to relaxing the threshold: Target ARL0 is determined by the effective independent sample ratio ρ eff ∈(0,1] is corrected to ARL0 / ρ eff ;ρ eff Obtained from the first-order autocorrelation estimation of the high-frequency sequence; the freeze time after the alarm is taken as T. lock =n lock / f hf The typical recovery time for a single wind disturbance is covered; After the alarm is triggered, a freeze period begins, and the system stops. Accumulate and update only μ ref , σ ref The slow channel resumes accumulation after the freeze period ends; if a coal type switching event occurs within the window, then d0 and A are refreshed. ± Configured for this type of coal.

8. A method for early warning of blockage in pulverized coal pipelines of a thermal power plant as described in claim 1, characterized in that, The graded early warning is classified according to the duration of the abnormality, the degree of deviation, and the spatial consistency, and outputs recommended handling measures including adjusting the primary air volume, changing the coal feeder speed, separator slag discharge, and load reduction operation. The early warning information can be connected to the distributed control system through the communication interface.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for early warning of blockage in pulverized coal pipelines of thermal power plants as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for early warning of blockage in pulverized coal pipelines of thermal power plants as described in any one of claims 1 to 8.

Images