A power plant boiler fault diagnosis method, system, terminal and medium based on an artificial neural network

Abstract

This invention belongs to the field of fault diagnosis technology, specifically disclosing a method, system, terminal, and medium for fault diagnosis of power plant boilers based on artificial neural networks. It includes: collecting multi-source operating parameter data during boiler operation, constructing a feature vector of the current operating state, and identifying the operating condition; processing the operating parameters based on the identified operating condition information to obtain operating condition adaptation feature data; inputting the operating condition adaptation feature data into a fault discrimination model to obtain a fault discrimination value; adaptively determining a dynamic fault discrimination threshold matching the current operating condition; further determining whether the current operating state is in a transition zone or a feature distribution drift range; when in a transition zone, outputting an early warning and triggering a verification process; and outputting a confirmation alarm result when the verification conditions are met. This invention achieves adaptive diagnosis of boiler status under multiple operating conditions and non-stable operating environments, effectively reducing false alarms and missed alarms, and improving the accuracy and stability of fault early warning.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis, specifically relating to a method, system, terminal, and medium for fault diagnosis of power plant boilers based on artificial neural networks. Background Technology

[0002] As the core heat energy conversion equipment in coal-fired power plants, the boiler's operating status directly affects the safety, economy, and stability of the unit. In actual operation, if an abnormality or malfunction occurs in the boiler system and is not identified and addressed promptly, it can easily lead to equipment damage, unplanned shutdowns, or even safety accidents.

[0003] To improve the automation level of boiler operation status monitoring and fault diagnosis, various fault diagnosis and early warning methods based on operating parameter analysis have been proposed in the existing technology. These include monitoring methods based on empirical rules and threshold judgments, diagnostic methods based on statistical analysis and correlation modeling, and data-driven diagnostic methods that incorporate artificial neural networks, fuzzy logic, or deep learning models. These methods typically collect multi-source monitoring data such as temperature, pressure, flow rate, and composition during boiler operation, extract features, and construct fault discrimination models to identify and warn of boiler operational anomalies.

[0004] However, existing fault diagnosis and early warning technologies for thermal power plants / boilers are mostly based on fixed threshold rules, single model training, or simple online training mechanisms. These are difficult to adapt to the frequent switching and non-stationary characteristics of multiple operating conditions caused by factors such as peak shaving, load fluctuations, and changes in coal quality. In this situation, the statistical characteristics of operating data drift over time, and the behavior of normal operating condition changes on some key parameters may be similar to the characteristics of actual faults, leading to both false alarms and false alarms. At the same time, traditional online update strategies lack discrimination gating between "operating condition changes / actual faults," and the model is prone to catastrophic amnesia during continuous updates, making it unable to stably and reliably accommodate historical operating conditions in the long term. In addition, fixed thresholds cannot reflect operating condition drift in a timely manner, further reducing the accuracy and real-time performance of online monitoring. Summary of the Invention

[0005] This invention addresses the problems in the prior art by providing a power plant boiler fault diagnosis method, system, terminal, and medium based on artificial neural networks. This solves the problem that fault diagnosis and early warning technologies in the prior art are mostly based on fixed threshold rules, single model training, or simple online training mechanisms, which are difficult to adapt to the frequent switching of multiple operating conditions and the non-stationary characteristics of multiple operating conditions caused by factors such as peak shaving, load fluctuations, and changes in coal quality.

[0006] The technical solution adopted in this invention is as follows: In a first aspect, this application provides a method for fault diagnosis of power plant boilers based on artificial neural networks, the method comprising the following steps: Step S1: Collect multi-source operating parameter data during boiler operation. The operating parameter data includes process parameters that reflect combustion status, steam status, and fluid transport status. Step S2: Construct a feature vector of the current operating status based on the operating parameter data, identify the operating conditions, and obtain the current operating condition information; Step S3: Based on the current operating condition status information, perform operating condition-related feature mapping or feature reconstruction processing on the operating parameter data to obtain operating condition adaptation feature data; Step S4: Input the working condition adaptation feature data into the fault discrimination model to obtain the fault discrimination value output by the model; Step S5: Based on the current operating condition status information and the statistical distribution characteristics of historical operating data under the corresponding operating condition, adaptively determine the fault discrimination threshold that matches the current operating condition; Step S6: Determine whether the current operating state is in the transition zone or the characteristic distribution drift zone; When the operating condition is in the transition zone or the characteristic distribution drift zone, the fault judgment value is output as the early warning level and the review process is triggered. When the review process meets the preset conditions, the confirmation alarm result is output. When the fault is not in the transition zone or the characteristic distribution drift zone, the fault judgment value is compared with the fault judgment threshold. When the fault judgment value exceeds the fault judgment threshold, a direct alarm result is output.

[0007] Furthermore, step S5 includes: Construct corresponding spatial distribution models of working condition features for different working condition categories; Based on the distribution position of the current operating state feature vector in the corresponding operating condition feature space, calculate the feature offset or probability density value; Based on feature offset or probability density values, a dynamic fault discrimination threshold matching the current operating state is generated through a threshold mapping function. Among them, the dynamic fault discrimination threshold is continuously adjusted as the distribution of the operating state feature vector in the operating condition feature space changes.

[0008] Furthermore, calculate the feature offset of the current running state feature vector:

[0009] in, Let t be the vector of working condition mean values, and t be the current time. This is the feature vector of the current running state. It is a 2-norm. The trend sensitivity coefficient The covariance matrix is ​​constructed based on historical stable operating samples of the k-th type of operating condition; Construct a threshold mapping function based on feature offsets:

[0010] in, As the baseline threshold, For the historical offset quantiles corresponding to the operating conditions, , This is the adjustment coefficient.

[0011] Furthermore, an adaptive weighting function based on joint time and state adjustment is used to construct the working condition mean vector:

[0012] in, The set of samples of the k-th type of operating condition based on historical stable operating intervals. Construct an adaptive weight function that jointly regulates time and state:

[0013] in, These are the feature vectors of the i-th and j-th historical running states, respectively. Let be the number of samples used for modeling within the historical stable operating range of the k-th type of operating condition. , These represent the sampling times corresponding to the i-th and j-th historical samples, respectively. The time decay coefficient characterizes the degree to which the time difference decays the weights; it is a constant or a function of time / state. This represents the state similarity weight coefficient.

[0014] Furthermore, step S6, determining whether the current operating state is in the transition zone or the characteristic distribution drift zone, includes: Based on the relative relationship between the distribution position of the current operating state feature vector in the corresponding working condition feature space and the dynamic fault discrimination threshold, the deviation degree of the current operating state is calculated. Based on whether the degree of deviation is within the neighborhood of the dynamic fault discrimination threshold and the trend of the degree of deviation within the continuous sampling period, it is determined whether the current operating state is in the working condition transition zone or the characteristic distribution drift zone. When the degree of deviation is within the neighborhood of the dynamic fault discrimination threshold and shows a continuous increasing trend, it is determined to be the working condition transition zone or the characteristic distribution drift zone.

[0015] Furthermore, step S6, determining whether the current operating state is in the transition zone or the characteristic distribution drift range, includes: Based on feature offset With continuous dynamic threshold Constructing gating margin indicators:

[0016] The current operating state is determined to be in the transition zone or characteristic distribution drift range when any of the following conditions are met: or ; in, The threshold for the transition zone gating bandwidth. This is the threshold for the gated drift rate.

[0017] Furthermore, when it is determined that the current operating state is in the transition zone or the characteristic distribution drift range, the review process includes: Within the preset time window Internal gating margin index Conduct continuous monitoring and construct a cumulative review index:

[0018] When the cumulative indicators for review meet When the alarm result is confirmed, output confirmation. in, To verify the length of the time window, The threshold for review and judgment.

[0019] Secondly, this application provides a power plant boiler fault diagnosis system based on artificial neural networks, used to implement the power plant boiler fault diagnosis method based on artificial neural networks as described in the first aspect. The system includes: The data acquisition unit is used to collect multi-source operating parameter data during boiler operation. The operating parameter data includes process parameters that reflect combustion status, steam status, and fluid transport status. The working condition identification unit, connected to the data acquisition unit, is used to construct a feature vector of the current operating state based on the operating parameter data, identify the operating condition of the operating state, and obtain the current operating condition information. The feature processing unit, connected to the working condition identification unit, is used to perform working condition-related feature mapping or feature reconstruction processing on the operating parameter data based on the current working condition status information to obtain working condition-adaptive feature data. The fault discrimination unit, connected to the feature processing unit, is used to input the working condition adaptation feature data into the artificial neural network fault discrimination model to obtain the fault discrimination value. The threshold determination unit, connected to the fault discrimination unit, is used to adaptively determine the fault discrimination threshold that matches the current operating condition based on the current operating condition status information and the statistical distribution characteristics of historical operating data under the corresponding operating condition. The gating determination unit, connected to the threshold determination unit, is used to determine whether the current operating state is in the working condition transition zone or the feature distribution drift zone. The output control unit, connected to the gating judgment unit, is used to output the fault judgment value as a warning level and trigger the review process when the current operating state is in the working condition transition zone or the characteristic distribution drift zone. When the review process meets the preset conditions, the confirmed alarm result is output. When the current operating state is not in the working condition transition zone or the characteristic distribution drift zone, the fault judgment value is compared with the fault judgment threshold. When the fault judgment value exceeds the fault judgment threshold, the direct alarm result is output.

[0020] Thirdly, this application provides a terminal, including: Memory, used to store power plant boiler fault diagnosis programs based on artificial neural networks; A processor is configured to implement the steps of the artificial neural network-based power plant boiler fault diagnosis method as described in the first aspect when executing the artificial neural network-based power plant boiler fault diagnosis program.

[0021] Fourthly, this application provides a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the power plant boiler fault diagnosis method based on artificial neural networks as described in the first aspect.

[0022] As can be seen from the above technical solutions, the advantages of the present invention are: By constructing a complete diagnostic process that includes multi-source operating parameter acquisition, operating condition identification, operating condition adaptation feature processing, artificial neural network fault discrimination, dynamic threshold determination, and gating and diversion output, a structured closed-loop control from data input to alarm output is achieved. Compared with the traditional method of direct judgment based on fixed thresholds or a single model, this application introduces a dual-path processing mechanism of "operating condition discrimination + dynamic threshold + transition zone gating" at the overall framework level. This means that the fault detection process no longer simply relies on a single discrimination result, but instead combines the operating environment to perform path diversion judgment. As a result, in complex operating scenarios with frequent switching of multiple operating conditions, load fluctuations, and non-stationary parameter changes, the false alarm rate and false negative rate are significantly reduced, the stability and adaptability of fault diagnosis are improved, and the reliability of the system during long-term online operation is enhanced.

[0023] By introducing an adaptive threshold construction method based on the statistical distribution characteristics of historical operating data for corresponding operating conditions during the dynamic threshold determination stage, the fault discrimination threshold is no longer a fixed constant but can automatically match different operating condition categories, thereby avoiding misjudgment problems caused by using the same threshold for different operating load ranges. This technical solution can automatically form corresponding judgment criteria according to the distribution characteristics of each operating condition, effectively improving the judgment accuracy under wide load ranges, peak shaving operations, and fuel quality fluctuation conditions, making fault judgment more consistent with the actual operating background.

[0024] By constructing feature offsets with a covariance structure and introducing a rate of change term for operating state into the offsets, the deviation of the current operating state from the feature space simultaneously reflects spatial distribution differences and temporal trends. This design breaks through the traditional method of measuring the degree of anomaly solely by static distance, enabling the model to more sensitively identify progressive anomalies, slow degradation, and trend drift, thereby discovering potential risks in advance and improving the foresight and continuity of early warnings.

[0025] By constructing a working condition mean vector through an adaptive weight function jointly adjusted by time and state, historical samples are differentiated and weighted according to time difference and state similarity during distribution modeling. This technique enhances the responsiveness to recent operating conditions while preventing interference from long-term samples in current judgments, thereby improving the adaptability of the distribution model under non-stationary operating conditions and effectively mitigating model inaccuracies caused by operating condition drift.

[0026] By refining the determination of whether the system is in a transitional operating condition zone or a characteristic distribution drift range, an environmental state recognition mechanism is added before fault diagnosis. This mechanism can trigger gating judgment when the judgment result is critical or the operating state is at a changing boundary, avoiding direct output of alarm signals at the moment of operating condition switching, thereby reducing false alarms caused by natural changes in operating conditions and improving the overall robustness of the system's judgment.

[0027] By constructing a gating margin index and combining it with a dynamic fault discrimination threshold for quantitative judgment, the transition zone identification is given a clear mathematical expression. This design uses the difference between the offset and the dynamic threshold as the core gating basis, enabling the system to accurately distinguish between "transitional states approaching the threshold" and "abnormal states that are clearly exceeding the limit," thereby achieving reasonable diversion of alarm paths and improving the system's judgment accuracy within the boundary interval.

[0028] By introducing a time-window-based integral verification mechanism in the transition zone, the duration and magnitude of deviation are cumulatively evaluated, thus avoiding false alarms caused by instantaneous fluctuations. This mechanism continuously monitors gating indicators and outputs a confirmed alarm result only when preset cumulative conditions are met. This ensures that alarm decisions are based on "continuous anomalies" rather than "instantaneous anomalies," significantly enhancing the reliability of fault diagnosis and improving the system's stable operation in complex environments. Attached Figure Description

[0029] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of the power plant boiler fault diagnosis method based on artificial neural networks according to the present invention; Figure 2 This is an architecture diagram of the power plant boiler fault diagnosis system based on artificial neural networks of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 As shown, this application provides a power plant boiler fault diagnosis method based on artificial neural networks, including the following steps: Step S1: Collect multi-source operating parameter data during boiler operation. The operating parameter data includes process parameters that reflect combustion status, steam status, and fluid transport status. In practical implementation, multi-source operating parameter data can be acquired in real time through a distributed control system or field instrument acquisition system. The collected data includes key parameters such as furnace temperature, primary air volume, secondary air volume, coal feed rate, feedwater flow rate, main steam pressure, main steam temperature, reheat steam temperature, flue gas oxygen content, and the operating status of the forced draft fan and induced draft fan. These parameters are collected synchronously at a preset sampling period and uniformly connected to the data processing platform through a data interface. In one embodiment, when the unit is operating at rated load, various operating parameters are collected once per second, and outliers are removed and filtered to ensure data quality. In another embodiment, when the unit is in peak-shaving operation or a phase of rapid load change, the sampling frequency can be appropriately increased to enhance the ability to capture changes in operating status. Through the above multi-source data acquisition method, a complete and continuous input information foundation is ensured for the subsequent diagnostic model.

[0033] Step S2: Construct a feature vector of the current operating status based on the operating parameter data, identify the operating conditions, and obtain the current operating condition information; In the specific implementation process, the multi-source operating parameters collected in step S1 are combined according to a preset dimensional order to form a feature vector representing the current operating state. This feature vector may include the original parameter values ​​or derived features after normalization or moving average processing. Subsequently, the feature vector is classified by a pre-trained operating condition identification model to obtain the operating condition category to which the current operating state belongs, such as low load, medium load, high load, or start-stop transition, etc. In one embodiment, the operating condition identification model can be trained using supervised learning, through historical operating data and manually labeled operating conditions; in another embodiment, clustering can be used to automatically group historical data, thereby forming operating condition divisions for different operating intervals. Through the above operating condition identification process, subsequent fault diagnosis is no longer performed under a globally unified model, but rather under a clearly defined operating condition background, improving the specificity of diagnosis.

[0034] Step S3: Based on the current operating condition status information, perform operating condition-related feature mapping or feature reconstruction processing on the operating parameter data to obtain operating condition adaptation feature data; In the specific implementation process, based on the operating condition information obtained in step S2, the feature processing rules for the corresponding operating condition are invoked to reconstruct the operating parameters. For example, under low-load conditions, combustion stability-related parameters have a significant impact on fault determination, and these parameters can be weighted more heavily; under high-load conditions, steam pressure and temperature fluctuations are more sensitive, so these parameters are highlighted. In one embodiment, different feature mapping matrices can be established according to different operating conditions to convert the original feature vectors into operating condition-adaptive feature vectors; in another embodiment, the features under different operating conditions can be reconstructed using standardized intervals to make the feature distributions between different operating conditions comparable. Through operating condition-related feature processing, the input of the fault discrimination model is made more consistent with the current operating background, thereby improving the stability and accuracy of the model output.

[0035] Step S4: Input the working condition adaptation feature data into the fault discrimination model to obtain the fault discrimination value output by the model; In the specific implementation process, the operating condition adaptation feature data obtained in step S3 is input into a pre-trained artificial neural network fault discrimination model. The model outputs a fault discrimination value representing the degree of abnormality in the current operating state. This fault discrimination value can be a continuous numerical value, used to represent the degree of deviation between the current state and the normal operating state. In one embodiment, the fault discrimination model can be trained based on historical normal operating data and fault data. By continuously optimizing the model parameters, it can be made capable of identifying fault types such as boiler heating surface leakage, unstable combustion, and coal mill blockage. In another embodiment, cross-validation can be used during model training to improve generalization ability and adapt to different unit operating conditions. The fault discrimination value output by the neural network model provides a basis for subsequent threshold determination.

[0036] In this embodiment, the fault discrimination model is constructed using a multi-layer feedforward artificial neural network structure. The model includes an input layer, several hidden layers, and an output layer. The number of nodes in the input layer is consistent with the dimensionality of the operating condition adaptation feature data, and each input node corresponds to an operating state feature. The hidden layers are used to extract the non-linear correlations between features and can be configured as two or three layers, with each layer containing several neurons. The output layer outputs a fault discrimination value representing the degree of abnormality in the current operating state. The output can be a single continuous numerical value, used to characterize the degree of deviation between the current state and the normal operating state.

[0037] During the model training phase, a training dataset is first constructed. This dataset includes historical normal operation data and confirmed fault operation data. Historical normal operation data can be selected from operating records during the unit's long-term stable operation phase, ensuring coverage of typical operating conditions under different load ranges, coal combustion conditions, and seasonal environments. Fault operation data can be selected from actual operation data such as heating surface leakage, unstable combustion, coal mill blockage, and abnormal air supply, and annotated by operating or maintenance records. To improve the model's generalization ability, the data undergoes denoising, outlier removal, and standardization to ensure that different parameters are within comparable ranges.

[0038] In one embodiment, the processed historical data is divided into a training set and a validation set according to a preset ratio. The training set is used for model parameter optimization, and the validation set is used to evaluate the model's performance on unseen data. During training, the internal connection weights of the network are continuously adjusted iteratively to gradually reduce the difference between the model output and the sample labeling results. When the model's error on the validation set stabilizes and meets the preset accuracy requirements, training is stopped and the model parameters are saved.

[0039] In another embodiment, to adapt to the differences between different power plants or different boiler models, the model can be retrained or fine-tuned using historical data of the corresponding power plant while keeping the basic model structure unchanged, so that the model parameters are more in line with the specific equipment characteristics.

[0040] During actual operation, the trained fault discrimination model is deployed in the online monitoring system, receiving the operating condition adaptation feature data generated in step S3 in real time and outputting fault discrimination values. These fault discrimination values ​​then participate in dynamic threshold judgment and gating triage mechanisms, thus forming a complete fault diagnosis process.

[0041] Through the above implementation methods, it is ensured that the fault discrimination model not only has good recognition capabilities, but also can work in conjunction with the working condition recognition and dynamic threshold mechanism proposed in this application to achieve stable and reliable fault diagnosis in complex operating environments with multiple working conditions.

[0042] Step S5: Based on the current operating condition status information and the statistical distribution characteristics of historical operating data under the corresponding operating condition, adaptively determine the fault discrimination threshold that matches the current operating condition; Specifically, step S5 includes: Construct corresponding spatial distribution models of working condition features for different working condition categories; In the specific implementation process, an independent feature space distribution model is established for each type of operating condition. This model, based on historical stable operating data, reflects the correlation structure and distribution range between various features under that operating condition. In one embodiment, historical samples can be filtered chronologically, retaining only data from time periods when the equipment is stable and has no fault records, to construct a reference distribution model for that operating condition. In this way, each type of operating condition has an independent feature space representation, avoiding mutual interference between different operating conditions.

[0043] Based on the distribution position of the current operating state feature vector in the corresponding operating condition feature space, calculate the feature offset or probability density value; In the specific implementation process, the feature vector of the current operating state is mapped to the feature space model of the corresponding working condition, and its positional relationship relative to the historical stable distribution is evaluated. This evaluation result can be expressed as the degree of deviation of the current state from the distribution center, or as the probability level of the current state's occurrence under that distribution. When the current state is located in the distribution center region, it indicates a high degree of consistency with the stable operating state; when it gradually approaches the distribution boundary or falls into a low-probability region, it indicates a difference from historical stable operation. In one embodiment, the offset is reflected by comparing the distance between the current state and the historical sample clustering region; in another embodiment, the probability density value is reflected by statistically analyzing the frequency of the current state's occurrence in the historical sample distribution. Through this method, multi-dimensional feature differences are transformed into a single indicator, providing a basis for dynamic threshold generation.

[0044] Based on feature offset or probability density values, a dynamic fault discrimination threshold matching the current operating state is generated through a threshold mapping function. In practice, based on the obtained feature offset or probability density value, a preset threshold mapping rule is invoked to adjust the original baseline threshold. This mapping rule can relax or tighten the threshold according to the magnitude of the offset. For example, when the current state is still in the core area of ​​a historically stable distribution, a relatively strict threshold standard can be maintained; when the current state is close to the distribution boundary, the threshold level can be moderately adjusted to avoid misjudging normal fluctuations as faults. In one embodiment, the threshold mapping process is implemented as a continuous function, allowing the threshold to change smoothly with the degree of offset, rather than being adjusted abruptly, thereby enhancing the stability of the system's judgment.

[0045] Among them, the dynamic fault discrimination threshold is continuously adjusted as the distribution of the operating state feature vector in the operating condition feature space changes.

[0046] In practical implementation, the system reassesses the position of the current operating state in the feature space within each sampling period and updates the fault discrimination threshold based on the latest offset result. As the operating state moves in the feature space, the threshold level changes synchronously, ensuring that the judgment boundary always fits the current operating context. In one embodiment, when the unit gradually transitions from a stable load to a new operating range, the threshold will gradually transition with the distribution change without abrupt changes, thereby avoiding frequent alarms or missed alarms. Through the continuous adjustment mechanism, the fault determination process has good real-time adaptability and dynamic robustness.

[0047] Step S6: Determine whether the current operating state is in the transition zone or the characteristic distribution drift zone; In the specific implementation process, the fault discrimination value obtained in step S4 and the dynamic fault discrimination threshold obtained in step S5 are combined to evaluate the positional relationship of the current operating state in the corresponding operating condition feature space. When the operating state is detected to be within the threshold neighborhood, or when the operating state shows a continuous shift trend within a continuous sampling period, the current state is determined to be in the operating condition transition zone or the feature distribution drift zone. In one embodiment, when the unit transitions from low load to medium load, some parameters may fluctuate in a short period of time. At this time, the system can identify it as a transition zone rather than immediately determining it as a fault. In another embodiment, when the fuel quality changes and causes the overall combustion parameters to shift, the system can identify it as a feature distribution drift zone to avoid direct alarm. This transition zone determination mechanism enhances the robustness of the system in dynamic operating environments.

[0048] When the operating condition is in the transition zone or the characteristic distribution drift zone, the fault judgment value is output as the early warning level and the review process is triggered. When the review process meets the preset conditions, the confirmation alarm result is output. In practical implementation, when a transition zone is identified, the final alarm result is not immediately output. Instead, a warning state is entered, and the changes in the fault discrimination value are continuously tracked within a preset time window. In one embodiment, the system can monitor abnormal trends over several consecutive sampling periods, and only output a confirmed alarm result when the abnormality persists or further expands; if the abnormality returns to the normal range within a short period of time, the warning state is canceled. This verification mechanism effectively avoids false alarms caused by short-term fluctuations and improves the reliability of alarm information.

[0049] When the fault is not in the transition zone or the characteristic distribution drift zone, the fault judgment value is compared with the fault judgment threshold. When the fault judgment value exceeds the fault judgment threshold, a direct alarm result is output.

[0050] In practical implementation, when the system determines that the current operating state is within a stable operating range and there are no signs of distribution drift, it directly compares the fault discrimination value with the dynamic fault discrimination threshold. When the fault discrimination value significantly exceeds the threshold, the system immediately outputs an alarm signal and can simultaneously send alarm information to the upper-level monitoring system so that operators can handle it promptly. In one embodiment, the alarm signal can also be recorded to the operation log for subsequent analysis and model optimization. The design of a direct alarm path enables the system to respond quickly in clearly defined abnormal situations, meeting the real-time requirements of boiler operation safety management.

[0051] In some embodiments, the feature offset of the current running state feature vector is calculated:

[0052] in, Let t be the vector of working condition mean values, and t be the current time. This is the feature vector of the current running state. It is a 2-norm. The trend sensitivity coefficient The covariance matrix is ​​constructed based on historical stable operating samples of the k-th type of operating condition; Construct a threshold mapping function based on feature offsets:

[0053] in, As the baseline threshold, For the historical offset quantiles corresponding to the operating conditions, , This is the adjustment coefficient.

[0054] In some embodiments, an adaptive weighting function based on joint time and state adjustment is used to construct the operating condition mean vector:

[0055] in, The set of samples of the k-th type of operating condition based on historical stable operating intervals. Construct an adaptive weight function that jointly regulates time and state:

[0056] in, These are the feature vectors of the i-th and j-th historical running states, respectively. Let be the number of samples used for modeling within the historical stable operating range of the k-th type of operating condition. , These represent the sampling times corresponding to the i-th and j-th historical samples, respectively. The time decay coefficient characterizes the degree to which the time difference decays the weights; it is a constant or a function of time / state. This represents the state similarity weight coefficient.

[0057] In some embodiments, determining whether the current operating state is in the transition zone or the characteristic distribution drift zone in step S6 includes: In the specific implementation process, step S6, as a key diversion link in the entire diagnostic process, focuses on distinguishing between two different scenarios: "real anomalies" and "natural changes in operating conditions." Therefore, before proceeding to alarm judgment, the system first performs a background consistency analysis of the current operating state based on the operating condition feature space model constructed in the preceding steps. In one embodiment, when the unit is in a load switching, fuel blending ratio adjustment, or start-up / shutdown phase, some operating parameters may experience periodic fluctuations. These fluctuations are not caused by equipment failure but by changes in operating strategies. Therefore, by setting up a dedicated transition zone judgment mechanism, the system achieves higher judgment robustness in complex operating environments, avoiding the direct output of false alarm results.

[0058] Based on the relative relationship between the distribution position of the current operating state feature vector in the corresponding working condition feature space and the dynamic fault discrimination threshold, the deviation degree of the current operating state is calculated. In practical implementation, the feature vector of the current operating state is mapped to the feature space model of the corresponding working condition, and its position relative to the historical stable operating distribution of that working condition is evaluated. This deviation reflects the degree of difference between the current state and the stable operating state. In one embodiment, the system first obtains the feature value of the current operating state and compares it with the distribution range of historical stable samples to obtain an index characterizing the degree of deviation. When the current state is close to the center of the historical stable distribution, the degree of deviation is small; when the current state gradually moves away from the boundary of the stable region, the degree of deviation increases. In this way, the complex multidimensional differences in operating states are transformed into a quantifiable degree of deviation, providing a basis for subsequent transition zone judgment.

[0059] Based on whether the degree of deviation is within the neighborhood of the dynamic fault discrimination threshold and the trend of the degree of deviation within the continuous sampling period, it is determined whether the current operating state is in the working condition transition zone or the characteristic distribution drift zone. In practical implementation, the system compares the current deviation level with the dynamic fault discrimination threshold and focuses on its range. When the deviation level is within the neighborhood of the threshold, it indicates that the current state is close to the abnormal boundary but has not yet significantly exceeded the normal range. In one embodiment, when the unit gradually increases from low load to medium load, combustion status and steam parameters may experience short-term fluctuations, at which point the deviation level may approach the threshold boundary. The system tracks the trend of this deviation level over multiple consecutive sampling periods. If it shows a gradual upward trend, it indicates that the operating state may be in a transition phase or there is a risk of distribution drift; if the deviation level fluctuates within the neighborhood but does not continue to increase, it can be determined as a normal disturbance. By combining the dual judgment mechanism of range location and time trend, the accuracy of transition zone identification is improved.

[0060] When the degree of deviation is within the neighborhood of the dynamic fault discrimination threshold and shows a continuous increasing trend, it is determined to be the working condition transition zone or the characteristic distribution drift zone.

[0061] In practical implementation, when the system detects that the deviation is within the threshold neighborhood and shows a significant increasing trend within a continuous sampling period, it determines that the current operating state is in the transition zone or the characteristic distribution drift zone. In one embodiment, when changes in coal quality lead to a gradual decrease in combustion efficiency, relevant operating parameters continuously deviate over several minutes, and the deviation gradually increases, which the system uses to determine it as a distribution drift zone. In another embodiment, when the unit executes a load adjustment command, some parameters show increasing changes in a short period of time, but gradually recover as the operating conditions stabilize. If the increasing trend does not continue, it will not enter the transition zone determination. Through the above mechanism, the system can effectively distinguish between "transient deviations caused by changes in operating conditions" and "continuous anomalies caused by potential faults," providing a reliable basis for subsequent early warning and review processing.

[0062] In some embodiments, determining whether the current operating state is in the transition zone or the characteristic distribution drift zone in step S6 includes: Based on feature offset With continuous dynamic threshold Constructing gating margin indicators:

[0063] The current operating state is determined to be in the transition zone or characteristic distribution drift range when any of the following conditions are met: or ; in, The threshold for the transition zone gating bandwidth. This is the threshold for the gated drift rate.

[0064] In some embodiments, when it is determined that the current operating state is in the transition zone or the characteristic distribution drift zone, the review process includes: Within the preset time window Internal gating margin index Conduct continuous monitoring and construct a cumulative review index:

[0065] When the cumulative indicators for review meet When the alarm result is confirmed, output confirmation. in, To verify the length of the time window, The threshold for review and judgment.

[0066] Please see Figure 2 As shown, in some embodiments, this application provides a power plant boiler fault diagnosis system based on artificial neural networks, used to implement a power plant boiler fault diagnosis method based on artificial neural networks. The system includes: The data acquisition unit is used to collect multi-source operating parameter data during boiler operation. The operating parameter data includes process parameters that reflect combustion status, steam status, and fluid transport status. The working condition identification unit, connected to the data acquisition unit, is used to construct a feature vector of the current operating state based on the operating parameter data, identify the operating condition of the operating state, and obtain the current operating condition information. The feature processing unit, connected to the working condition identification unit, is used to perform working condition-related feature mapping or feature reconstruction processing on the operating parameter data based on the current working condition status information to obtain working condition-adaptive feature data. The fault discrimination unit, connected to the feature processing unit, is used to input the working condition adaptation feature data into the artificial neural network fault discrimination model to obtain the fault discrimination value. The threshold determination unit, connected to the fault discrimination unit, is used to adaptively determine the fault discrimination threshold that matches the current operating condition based on the current operating condition status information and the statistical distribution characteristics of historical operating data under the corresponding operating condition. The gating determination unit, connected to the threshold determination unit, is used to determine whether the current operating state is in the working condition transition zone or the feature distribution drift zone. The output control unit, connected to the gating judgment unit, is used to output the fault judgment value as a warning level and trigger the review process when the current operating state is in the working condition transition zone or the characteristic distribution drift zone. When the review process meets the preset conditions, the confirmed alarm result is output. When the current operating state is not in the working condition transition zone or the characteristic distribution drift zone, the fault judgment value is compared with the fault judgment threshold. When the fault judgment value exceeds the fault judgment threshold, the direct alarm result is output.

[0067] In some embodiments, this application provides a terminal, including: Memory, used to store power plant boiler fault diagnosis programs based on artificial neural networks; A processor is used to implement the steps of the artificial neural network-based power plant boiler fault diagnosis method when executing the artificial neural network-based power plant boiler fault diagnosis program.

[0068] In some embodiments, this application provides a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the power plant boiler fault diagnosis method based on artificial neural networks.

[0069] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the scope of protection of one or more embodiments of this specification.

Claims

1. A method for fault diagnosis of power plant boilers based on artificial neural networks, characterized in that, Includes the following steps: Step S1: Collect multi-source operating parameter data during boiler operation. The operating parameter data includes process parameters that reflect combustion status, steam status, and fluid transport status. Step S2: Construct a feature vector of the current operating status based on the operating parameter data, identify the operating conditions, and obtain the current operating condition information; Step S3: Based on the current operating condition status information, perform operating condition-related feature mapping or feature reconstruction processing on the operating parameter data to obtain operating condition adaptation feature data; Step S4: Input the working condition adaptation feature data into the fault discrimination model to obtain the fault discrimination value output by the model; Step S5: Based on the current operating condition status information and the statistical distribution characteristics of historical operating data under the corresponding operating condition, adaptively determine the fault discrimination threshold that matches the current operating condition; Step S6: Determine whether the current operating state is in the transition zone or the characteristic distribution drift zone; When the operating condition is in the transition zone or the characteristic distribution drift zone, the fault judgment value is output as the early warning level and the review process is triggered. When the review process meets the preset conditions, the confirmation alarm result is output. When the fault is not in the transition zone or the characteristic distribution drift zone, the fault judgment value is compared with the fault judgment threshold. When the fault judgment value exceeds the fault judgment threshold, a direct alarm result is output.

2. The power plant boiler fault diagnosis method based on artificial neural networks according to claim 1, characterized in that, Step S5 includes: Construct corresponding spatial distribution models of working condition features for different working condition categories; Based on the distribution position of the current operating state feature vector in the corresponding operating condition feature space, calculate the feature offset or probability density value; Based on feature offset or probability density values, a dynamic fault discrimination threshold matching the current operating state is generated through a threshold mapping function. Among them, the dynamic fault discrimination threshold is continuously adjusted as the distribution of the operating state feature vector in the operating condition feature space changes.

3. The power plant boiler fault diagnosis method based on artificial neural networks according to claim 2, characterized in that, Calculate the feature offset of the current running state feature vector: in, Let t be the vector of working condition mean values, and t be the current time. This is the feature vector of the current running state. It is a 2-norm. The trend sensitivity coefficient The covariance matrix is ​​constructed based on historical stable operating samples of the k-th type of operating condition; Construct a threshold mapping function based on feature offsets: in, As the baseline threshold, For the historical offset quantiles corresponding to the operating conditions, , This is the adjustment coefficient; A dynamic fault detection threshold that matches the current operating state is generated through a threshold mapping function.

4. The power plant boiler fault diagnosis method based on artificial neural networks according to claim 3, characterized in that, Constructing the operating condition mean vector based on an adaptive weight function that combines time and state adjustment: in, The set of samples of the k-th type of operating condition based on historical stable operating intervals. Construct an adaptive weight function that jointly regulates time and state: in, These are the feature vectors of the i-th and j-th historical running states, respectively. Let be the number of samples used for modeling within the historical stable operating range of the k-th type of operating condition. , These represent the sampling times corresponding to the i-th and j-th historical samples, respectively. The time decay coefficient characterizes the degree to which the time difference decays the weights; it is a constant or a function of time / state. This represents the state similarity weight coefficient.

5. The power plant boiler fault diagnosis method based on artificial neural networks according to claim 3 or 4, characterized in that, Step S6, determining whether the current operating state is in the transition zone or the characteristic distribution drift range, includes: Based on the relative relationship between the distribution position of the current operating state feature vector in the corresponding working condition feature space and the dynamic fault discrimination threshold, the deviation degree of the current operating state is calculated. Based on whether the degree of deviation is within the neighborhood of the dynamic fault discrimination threshold and the trend of the degree of deviation within the continuous sampling period, it is determined whether the current operating state is in the working condition transition zone or the characteristic distribution drift zone. When the degree of deviation is within the neighborhood of the dynamic fault discrimination threshold and shows a continuous increasing trend, it is determined to be the working condition transition zone or the characteristic distribution drift zone.

6. The power plant boiler fault diagnosis method based on artificial neural networks according to claim 5, characterized in that, Step S6, determining whether the current operating state is in the transition zone or the characteristic distribution drift range, includes: Based on feature offset With continuous dynamic threshold Constructing gating margin indicators: The current operating state is determined to be in the transition zone or characteristic distribution drift range when any of the following conditions are met: or ; in, The threshold for the transition zone gating bandwidth. This is the threshold for the gated drift rate.

7. The power plant boiler fault diagnosis method based on artificial neural networks according to claim 6, characterized in that, When it is determined that the current operating status is in the transition zone or the characteristic distribution drift zone, the review process includes: Within the preset time window Internal gating margin index Conduct continuous monitoring and construct a cumulative review index: When the cumulative indicators for review meet When the alarm result is confirmed, output confirmation. in, To verify the length of the time window, The threshold for review and judgment.

8. A power plant boiler fault diagnosis system based on artificial neural networks, used to implement the power plant boiler fault diagnosis method based on artificial neural networks as described in claim 1, characterized in that, The system includes: The data acquisition unit is used to collect multi-source operating parameter data during boiler operation. The operating parameter data includes process parameters that reflect combustion status, steam status, and fluid transport status. The working condition identification unit, connected to the data acquisition unit, is used to construct a feature vector of the current operating state based on the operating parameter data, identify the operating condition of the operating state, and obtain the current operating condition information. The feature processing unit, connected to the working condition identification unit, is used to perform working condition-related feature mapping or feature reconstruction processing on the operating parameter data based on the current working condition status information to obtain working condition-adaptive feature data. The fault discrimination unit, connected to the feature processing unit, is used to input the working condition adaptation feature data into the artificial neural network fault discrimination model to obtain the fault discrimination value. The threshold determination unit, connected to the fault discrimination unit, is used to adaptively determine the fault discrimination threshold that matches the current operating condition based on the current operating condition status information and the statistical distribution characteristics of historical operating data under the corresponding operating condition. The gating determination unit, connected to the threshold determination unit, is used to determine whether the current operating state is in the working condition transition zone or the feature distribution drift zone. The output control unit, connected to the gating judgment unit, is used to output the fault judgment value as a warning level and trigger the review process when the current operating state is in the working condition transition zone or the characteristic distribution drift zone. When the review process meets the preset conditions, the confirmed alarm result is output. When the current operating state is not in the working condition transition zone or the characteristic distribution drift zone, the fault judgment value is compared with the fault judgment threshold. When the fault judgment value exceeds the fault judgment threshold, the direct alarm result is output.

9. A terminal, characterized in that, include: Memory, used to store power plant boiler fault diagnosis programs based on artificial neural networks; A processor is configured to implement the steps of the power plant boiler fault diagnosis method based on artificial neural networks as described in claim 1 when executing the power plant boiler fault diagnosis program based on artificial neural networks.

10. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions. When the computer reads the computer instructions from the storage medium, the computer executes the power plant boiler fault diagnosis method based on artificial neural networks as described in claim 1.

Images