Early warning method and system based on power plant side big data operation and maintenance

Abstract

The invention relates to an early warning method and system based on power plant side big data operation and maintenance. The method comprises the steps of collecting time sequence data of a power plant side; performing trend monitoring on the change rate of the measuring point data, comparing the measuring point data with the corresponding upper limit value and lower limit value, and performing limit value monitoring to obtain an absolute alarm result; setting a plurality of measuring point objects according to the experience of an operator to compare amplitude intervals or variation amplitudes, and performing combination through judgment logic to obtain a combined early warning result; carrying out screening and correlation verification on the selected measuring point data, selecting various measuring point data with strong correlation as parameters, taking a historical operation working condition as a sample, obtaining an abnormal change interval of each parameter under the working condition same as the current working condition, and carrying out comparison to obtain a relative alarm result; and performing closed-loop adjustment on each step according to an alarm result. Compared with the prior art, equipment early warning is completed from three angles, and the early warning accuracy and reliability of the power plant equipment are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power plant early warning, and in particular to an early warning method and system based on big data operation and maintenance on the power plant side. Background Art

[0002] As a traditional secondary industry, power plants have relatively high requirements for the stability and reliability of equipment operation. At present, the real-time monitoring of operating conditions or equipment status is mainly carried out through the following two aspects:

[0003] Firstly, based on the existing DCS distributed control system, an alarm function is usually embedded in this system. For example, an alarm system, method and storage medium of a power plant DCS disclosed in the invention with the publication number CN115755813A includes: an acquisition module for obtaining real-time operation parameters of the power plant DCS and sending the real-time operation parameters to a data processing module; the data processing module for classifying the real-time operation parameters according to parameter categories; judging whether the corresponding real-time operation parameters are abnormal according to the alarm limit parameters set for each parameter category; generating corresponding alarm instructions according to the parameter categories corresponding to the abnormal operation parameters; and a display module for displaying the corresponding real-time operation parameters, alarm limit parameters and generating corresponding alarm screens according to the parameter categories according to the alarm instructions.

[0004] However, this kind of alarm function generally only covers the upper and lower limit alarms of equipment operation parameters, that is, absolute alarms. And since the limits are set as the upper and lower operation limits calibrated at the equipment factory, it is not sensitive to the trend changes of equipment parameters. And often such trend changes will accelerate equipment aging, cause potential equipment failures, and then affect the actual operation. Even when an abnormal operation occurs, it is necessary for the operation personnel to trace back the historical working condition parameters to analyze and judge the change trend, and it is difficult to quickly find the parameters that really cause the abnormal working condition. Thus, it can be seen that the above alarm methods all belong to "post-event alarm" and cannot achieve "early warning".

[0005] Secondly, it is the monitoring of operating conditions. For example, during the load increase and decrease stage, all parameters do not trigger the DCS upper and lower limit alarms. Therefore, the monitoring personnel need to monitor all parameters in the control system in detail: whether the operating current, moving blade opening, air volume, inlet and outlet pressure and temperature of the fan in the wind and smoke system match the operating conditions under the current actual load; similarly, whether the coal feeder and coal mill in the coal pulverizing system are operating normally and whether there is coal blockage. All the above methods require the operation monitoring personnel to have sufficient operation experience and a high enough attention, which puts relatively high requirements on the energy of the operation personnel. Summary of the Invention

[0006] The object of the present invention is to overcome the defects of the above-mentioned existing technologies and provide an early warning method and system based on big data operation and maintenance on the power plant side, so as to improve the early warning accuracy and reliability of power plant equipment.

[0007] The object of the present invention can be achieved by the following technical solutions:

[0008] An early warning method based on big data operation and maintenance on the power plant side, comprising the following steps:

[0009] Data acquisition step: acquiring time series data on the power plant side;

[0010] Absolute alarm step: for the measured point data in the time series data, by monitoring the trend of the change rate of the measured point data and comparing the measured point data with the corresponding upper limit value and lower limit value for limit monitoring to obtain an absolute alarm result; during the trend monitoring and limit monitoring processes, receiving the actual equipment operation feedback and dynamically adjusting the change rate threshold, upper limit value and lower limit value;

[0011] Combined judgment step: setting multiple measured point objects according to the experience of the operation personnel for comparison of amplitude intervals or change ranges, and combining them through a judgment logic to obtain a combined early warning result;

[0012] Relative alarm step: using an anomaly detection algorithm, selecting various measured point data from the time series data, screening and verifying the correlation of the selected measured point data, selecting various measured point data with strong correlation as parameters, and using the historical operation conditions as samples to train the anomaly detection algorithm model to obtain the abnormal change intervals of each parameter under the same working conditions as the current working conditions, and comparing the real-time results of each parameter with the corresponding abnormal change intervals to obtain a relative alarm result;

[0013] Closed-loop diagnosis step: obtaining the absolute alarm result, combined early warning result and relative alarm result, generating a record in the alarm list, and associating the corresponding expert knowledge base according to the equipment and its fault type to perform closed-loop adjustment on the absolute alarm step, combined judgment step and relative alarm step.

[0014] Further, the data acquisition step further includes: performing real-time calculation of characterization data on the acquired time series data for the judgment of the absolute alarm step, combined judgment step and relative alarm step, and the characterization data includes coal consumption, differential pressure, terminal difference, efficiency and enthalpy entropy.

[0015] Further, the limit monitoring in the absolute alarm step is implemented through a calculation configuration function block, and the configuration function block includes input constants, input measured points, judgment elements and logic elements. The input measured points and input constants are compared and judged through the judgment elements, and the output results of each judgment element are combined through the logic elements to obtain an absolute alarm result.

[0016] Further, the configuration function block further includes an if-else function block for performing additional judgment on primitive graphics in other cases.

[0017] Further, in the combined judgment step, the configuration function block performs combined judgment on multiple input measurement points to obtain a combined warning result.

[0018] Further, the configuration function block for performing combined judgment on the unit load and the lubricating oil temperature of the coal mill roller includes:

[0019] A unit load input measurement point, a 260MW input constant, a first greater-than judgment primitive graphic, an if-else function block, code block A, and code block B; the unit load input measurement point and the 260MW input constant are respectively input into the first greater-than judgment primitive graphic, the output of the first greater-than judgment primitive graphic is connected to the if-else function block, the outputs of the if-else function block are respectively connected to code block A and code block B. When the unit load input measurement point is greater than the 260MW input constant, code block A is executed; otherwise, code block B is executed.

[0020] Code block A includes a coal mill roller lubricating oil temperature input measurement point, a 90-degree Celsius input constant, a -10-degree Celsius input constant, a second greater-than judgment primitive graphic, a less-than judgment primitive graphic, and an AND logic primitive graphic. The coal mill roller lubricating oil temperature input measurement point and the 90-degree Celsius input constant are respectively input into the second greater-than judgment primitive graphic. The coal mill roller lubricating oil temperature input measurement point and the -10-degree Celsius input constant are respectively input into the less-than judgment primitive graphic. The second greater-than judgment primitive graphic and the less-than judgment primitive graphic are respectively input into the AND logic primitive graphic. If any condition that the coal mill roller lubricating oil temperature input measurement point is greater than 90 degrees Celsius or less than -10 degrees Celsius is satisfied, the status quantity 1 is output, corresponding to the alarm status; otherwise, the status quantity is 0.

[0021] Code block B includes a 0 input constant, a 1 input constant, and a third greater-than judgment primitive graphic. The 0 input constant and the 1 input constant are respectively input into the third greater-than judgment primitive graphic, and the third greater-than judgment primitive graphic outputs the status quantity false.

[0022] Further, the logic primitive graphics include AND logic primitive graphics and OR logic primitive graphics, and the judgment primitive graphics include greater-than judgment primitive graphics, less-than judgment primitive graphics, multiplication judgment primitive graphics, division judgment primitive graphics, addition judgment primitive graphics, and subtraction judgment primitive graphics.

[0023] Further, during the process of the combined judgment step for determining whether the coal mill is blocked, by taking the coal mill current, primary air differential pressure, pulverized coal-air mixture pressure, and outlet pulverized coal-air flow rate as input measurement points respectively, and conducting trend monitoring and limit monitoring respectively, and then synthesizing the monitoring results of each item, it is determined whether the coal mill is likely to be blocked.

[0024] The present invention also provides an early warning system based on big data operation and maintenance on the power plant side, including:

[0025] A lightweight data middle platform, which is used to collect, transmit, and store the time-series data on the power plant side, and conduct data governance and data visualization;

[0026] An enabling tool middle platform, which is used for the analysis of time-series data and alarm processing;

[0027] The enabling tool middle platform includes:

[0028] An absolute alarm module: for the measurement point data in the time-series data, by conducting trend monitoring on the change rate of the measurement point data, comparing the measurement point data with the corresponding upper limit value and lower limit value for limit monitoring, and obtaining an absolute alarm result; during the trend monitoring and limit monitoring processes, receiving the actual equipment operation feedback, and dynamically adjusting the change rate threshold, upper limit value, and lower limit value;

[0029] A combined judgment module: setting multiple measurement point objects according to the experience of the operation personnel to compare the amplitude intervals or change ranges, and combining them through a judgment logic to obtain a combined early warning result;

[0030] A relative alarm module: selecting various measurement point data from the time-series data, screening and verifying the correlation of the selected measurement point data, selecting various measurement point data with strong correlation as parameters, using the historical operation conditions as samples, obtaining the abnormal change intervals of each parameter under the conditions identical to the current condition, and comparing the real-time results of each parameter with the corresponding abnormal change intervals to obtain a relative alarm result;

[0031] A closed-loop diagnosis module: obtaining the absolute alarm result, combined early warning result, and relative alarm result, generating a record in the alarm list, and associating the corresponding expert knowledge base according to the equipment and its fault type to conduct closed-loop adjustment on the absolute alarm step, combined judgment step, and relative alarm step.

[0032] Further, the limit monitoring in the absolute alarm module is implemented through a calculation configuration function block. The configuration function block includes input constants, input measurement points, judgment elements, and logic elements. The judgment elements are used to compare and judge the input measurement points and input constants, and the logic elements are used to combine the output results of each judgment element to obtain an absolute alarm result;

[0033] The configuration function block also includes an if-else function block, which is used to execute additional judgment primitives in other cases;

[0034] The combined judgment module performs combined judgment on multiple input measurement points through the configuration function block to obtain a combined early warning result;

[0035] The logic primitives include AND gate logic primitives and OR gate logic primitives, and the judgment primitives include greater than judgment primitives, less than judgment primitives, multiplication judgment primitives, division judgment primitives, addition judgment primitives, and subtraction judgment primitives.

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] (1) The present invention performs early warning on the collected time-series data on the power plant side in various ways. Through trend monitoring and limit monitoring in absolute alarm, it realizes the investigation and monitoring of potential hidden dangers of equipment. And compared with the conventional long lower limit alarm scheme, the present invention also dynamically adjusts the change rate threshold, upper limit value, and lower limit value according to the actual equipment operation feedback to adapt to the margin acceptable to operators under different operating conditions, and at the same time strengthens the sensitivity of the alarm to data trend changes;

[0038] Through combined judgment early warning, by building a configuration function block, using input constants, input measurement points, judgment primitives, logic primitives, and if-else function blocks to realize the comprehensive automatic judgment of multiple measurement point objects, and obtain a combined early warning result, which further strengthens the comprehensive monitoring of the equipment operating conditions, ensuring that the comprehensive operating conditions of each device in the same control system also operate normally when the equipment body operates normally;

[0039] Through relative alarm, using the anomaly detection algorithm to compare multiple strongly correlated measurement points with historical operating conditions, it realizes the monitoring of parameters that have undergone abnormal changes and realizes relative alarm prediction;

[0040] Finally, through the closed-loop diagnosis process, it realizes the closed-loop diagnosis and adjustment of absolute alarm, combined judgment, and relative alarm, and realizes continuous closed-loop optimization.

[0041] (2) The present invention realizes absolute alarm for measurement point data through a trend and limit monitoring process that can dynamically adjust the change rate threshold, upper limit value, and lower limit value, ensuring the safe operation of the equipment; through the combination of judgment logics for multiple measurement point objects, it further realizes the comprehensive monitoring of the equipment operating conditions from the mechanism perspective, realizes the overall fault prediction of the equipment, and performs early warning processing; through the use of the anomaly detection algorithm to compare and analyze the changes in historical operating conditions of measurement point data, it realizes the perception and early warning of anomalies from the perspective of historical data; overall, it completes equipment early warning from three perspectives, greatly improving the early warning accuracy and reliability of power plant equipment.

[0042] (3) The present invention proposes to realize the automatic judgment of the combination of input measuring points by setting input constants, input measuring points, judgment elements, logic elements, and if-else function blocks. The judgment logic diagram built in this way runs reliably, the logic principle is clear, the operability is strong, and the safety is high, providing technical support for subsequent realization of less personnel and unmanned operation, and providing a basis for subsequent exploration of the economic operation of the unit to improve quality and efficiency.

[0043] (4) The present invention adds a rate-of-change alarm, which is more sensitive to trend changes.

[0044] (5) The present invention retains the experience of the operators in the alarm system through external logic alarms, reducing the learning cost of the operators. By constructing an expert knowledge base, it provides closed-loop guidance for subsequent early detection of faults, analysis of cause phenomena, and provision of recommended measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a schematic flowchart of an early warning method based on big data operation and maintenance on the power plant side provided in an embodiment of the present invention;

[0046] Figure 2 is a schematic diagram of the combined early warning process of multiple different types of measuring point data provided in an embodiment of the present invention;

[0047] Figure 3 is a schematic diagram of the combined early warning process of multiple same-type measuring point data provided in an embodiment of the present invention;

[0048] Figure 4 is a schematic diagram of the combined processing process of measuring point data provided in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and shown here can be arranged and designed in various different configurations.

[0050] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0051] It should be noted that similar reference numerals and letters indicate similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0052] Embodiment 1

[0053] As Figure 1 shown, this embodiment provides an early warning method based on big data operation and maintenance on the power plant side, including the following steps:

[0054] Data acquisition step S1: Acquire the time-series data on the power plant side;

[0055] Absolute alarm step S2: For the measured point data in the time-series data, through trend monitoring of the change rate of the measured point data, compare the measured point data with the corresponding upper limit value and lower limit value for limit monitoring to obtain the absolute alarm result; during the trend monitoring and limit monitoring processes, receive the actual equipment operation feedback and dynamically adjust the change rate threshold, upper limit value, and lower limit value;

[0056] Combined judgment step S3: Set multiple measured point objects according to the experience of the operation personnel to compare the amplitude range or change amplitude, and combine them through the judgment logic to obtain the combined early warning result;

[0057] Relative alarm step S4: Use the anomaly detection algorithm to select various measured point data from the time-series data, screen and verify the correlation of the selected measured point data, select various measured point data with strong correlation as parameters, and use the historical operation conditions as samples to train the anomaly detection algorithm model, obtain the abnormal change range of each parameter under the same working condition as the current working condition, and compare the real-time results of each parameter with the corresponding abnormal change range to obtain the relative alarm result;

[0058] Closed-loop diagnosis step S5: Obtain the absolute alarm result, combined early warning result, and relative alarm result, generate a record in the alarm list, and associate the corresponding expert knowledge base according to the equipment and its fault type to perform closed-loop adjustment on the absolute alarm step, combined judgment step, and relative alarm step.

[0059] Preferably, the data acquisition step S1 further includes: for the time-series data, after the acquisition is completed, it is necessary to synchronously manage and verify the measured point information;

[0060] And perform real-time calculation of the characterization data for the judgment of the absolute alarm step, combined judgment step, and relative alarm step. The characterization data includes coal consumption, differential pressure, terminal difference, efficiency, and enthalpy entropy.

[0061] The limit monitoring in the absolute alarm step S2 is implemented through a configuration function block, which includes an input constant, an input measurement point, a judgment graphic element, and a logic graphic element. The input measurement point and the input constant are compared and judged through the judgment graphic element, and the output results of each judgment graphic element are combined through the logic graphic element to obtain the absolute alarm result.

[0062] The configuration function block also includes an if-else function block, which is used to execute another judgment graphic element in other cases.

[0063] For the measured point data with mutations in the absolute alarm step S2, absolute alarm is carried out through the change rate and upper and lower limit alarm models, aiming to achieve the investigation and monitoring of potential hidden dangers of equipment. Compared with the upper and lower limit alarms of DCS, the absolute alarm model of this solution not only supports real-time comparison value modification, but also can simultaneously perform trend monitoring and limit monitoring on a single measurement point. By continuously adjusting and optimizing the comparison values of the change rate and the upper and lower limits, the margin acceptable to the operators under different operating conditions can be adapted, and at the same time, the sensitivity of the alarm to the change of data trend is strengthened.

[0064] The combined judgment step S3 performs combined judgment on multiple input measurement points through the configuration function block to obtain the combined early warning result.

[0065] For the alarms in some operating conditions in the combined judgment step S3, an external alarm logic is established through the DCS configuration logic, and the experience of the operators is incorporated into the consideration scope of the alarm through the combined judgment method. For example, when general operators judge whether there is coal blockage in the coal mill, they often include the coal mill current, primary air differential pressure, pulverized coal-air mixture pressure, and outlet pulverized coal-air flow rate in the consideration scope. By comparing the amplitude ranges and change amplitudes of the above points, it can be judged whether there may be coal blockage in the coal mill. It can be said that the combined logic further strengthens the comprehensive monitoring of the equipment operating conditions through logical means on the basis of the change rate of the upper and lower limits of a single measurement point, ensuring that the comprehensive operating conditions of each equipment in the same control system also maintain normal operation when the equipment body operates normally.

[0066] Specifically, as Figure 2 shown, the configuration function block for combined judgment of the unit load and the lubricating oil temperature of the coal mill roller includes:

[0067] The unit load input measurement point, the 260MW input constant, the first greater than judgment graphic element, the if-else function block, code block A, and code block B; the unit load input measurement point and the 260MW input constant are respectively input into the first greater than judgment graphic element, the output of the first greater than judgment graphic element is connected to the if-else function block, and the output of the if-else function block is respectively connected to code block A and code block B. When the unit load input measurement point is greater than the 260MW input constant, code block A is executed, otherwise code block B is executed;

[0068] Code block A includes a coal mill roller lubricating oil temperature input measurement point, a 90 degree Celsius input constant, a -10 degree Celsius input constant, a second greater than judgment primitive, a less than judgment primitive, and an AND gate logic primitive. The coal mill roller lubricating oil temperature input measurement point and the 90 degree Celsius input constant are respectively input into the second greater than judgment primitive, the coal mill roller lubricating oil temperature input measurement point and the -10 degree Celsius input constant are respectively input into the less than judgment primitive, the second greater than judgment primitive and the less than judgment primitive are respectively input into the AND gate logic primitive. If any of the conditions that the coal mill roller lubricating oil temperature input measurement point is greater than 90 degrees Celsius or less than -10 degrees Celsius is met, the output state quantity is 1, corresponding to the alarm state, otherwise the output state quantity is 0;

[0069] Code block B includes a 0 input constant, a 1 input constant and a third greater than judgment primitive. The 0 input constant and the 1 input constant are respectively input into the third greater than judgment primitive, and the third greater than judgment primitive outputs a state value of false.

[0070] The logic primitives include AND gate logic primitives and OR gate logic primitives, and the judgment primitives include greater than judgment primitives, less than judgment primitives, multiplication judgment primitives, division judgment primitives, addition judgment primitives and subtraction judgment primitives.

[0071] like Figure 3 As shown, for multiple coal mill roller bearing lubricating oil temperature change rates, they are respectively used as input measuring points, connected with absolute value logic elements, and 0.1 input constants and greater than judgment elements are set. The absolute value logic element and 0.1 input constant are both input into the greater than judgment element, and each greater than judgment element is connected to an OR gate logic element to realize the threshold judgment of the coal mill roller bearing lubricating oil temperature change rate under multiple measuring points.

[0072] The solution is to add preconditions to achieve control over alarm triggering, and combine with js code blocks to link the state quantity of the calculated configuration output to implement specific alarm suppression rules, so as to uniformly constrain multiple input points corresponding to the model and realize a multi-dimensional cross-warning method based on the anomaly detection model.

[0073] like Figure 4 As shown, for the input measuring point of the coal mill 5A current, by setting the time backtracking logic element, the subtraction logic element and the division logic element, the input measuring point of the coal mill 5A current is respectively input into the time backtracking logic element and the subtraction logic element, the output of the time backtracking logic element is input into the subtraction logic element and the division logic element, the subtraction logic element subtracts the input measuring point of the coal mill 5A current from the output of the time backtracking logic element, the output of the subtraction logic element is input into the division logic element, the division logic element divides the output of the time backtracking logic element by the output of the subtraction logic element, and obtains the current change rate of the coal mill 5A current, which is used for the change rate alarm.

[0074] In the relative alarm step S4, various anomaly detection algorithms are also used to classify and model the real-time operating condition data and the system's secondary calculation indicators, and then score and evaluate them to detect potential equipment faults in advance.

[0075] Compared with absolute alarm, model alarm requires multiple strongly correlated measurement points as parameters. Specifically: by selecting measurement points, screening the data of the selected measurement points, and verifying the correlation of the screened measurement point data, multiple strongly correlated measurement points are obtained as parameters;

[0076] Essentially, model alarm uses multiple strongly correlated measurement points as parameters, takes historical operating conditions as samples, obtains a trained anomaly detection algorithm model, and realizes the monitoring of abnormal changes in multiple parameters under similar operating conditions, which belongs to the category of relative alarm.

[0077] That is, in this embodiment, a standard modeling process is designed, which are: selecting measurement points, data screening, correlation verification, model training and model evaluation, and the model is established and put into use online. In the closed-loop diagnosis step S5, after the above three alarm steps are executed, once an abnormal situation is detected, an alarm will be issued and a record will be generated in the alarm list. At the same time, according to the equipment and its fault type, a custom expert knowledge base will be automatically associated to conduct closed-loop diagnosis and guidance on the fault cause, fault phenomenon and recommended measures.

[0078] This solution conducts early warnings on the collected time-series data on the power plant side in various ways. Through trend monitoring and limit monitoring in absolute alarm, the potential hidden dangers of equipment are investigated and monitored. And compared with the conventional long lower limit alarm scheme, this solution also dynamically adjusts the change rate threshold, upper limit value and lower limit value according to the actual equipment operation feedback to adapt to the margin that operators can accept under different operating conditions, and at the same time strengthens the sensitivity of the alarm to data trend changes;

[0079] Through combined judgment early warning, by building configuration function blocks, using input constants, input measurement points, judgment graphics elements, logic graphics elements and if-else function blocks to realize the comprehensive automatic judgment of multiple measurement point objects, and obtain combined early warning results, which further strengthens the comprehensive monitoring of the equipment operating conditions and ensures that the comprehensive operating conditions of each equipment in the same control system also operate normally when the equipment body operates normally;

[0080] Through relative alarm, using anomaly detection algorithms to compare multiple strongly correlated measurement points with historical operating conditions, the monitoring of parameters with abnormal changes is realized, and relative alarm prediction is achieved;

[0081] Finally, through the closed-loop diagnosis process, the closed-loop diagnosis and adjustment of absolute alarm, combined judgment and relative alarm are realized, and continuous closed-loop optimization is achieved.

[0082] This embodiment proposes three main algorithm selection schemes, namely the Gaussian Mixture Model (GMM) for anomaly detection based on statistics, the Local Outlier Factor (LOF) for anomaly detection based on traditional machine learning methods, and the Support Vector Machine (SVM) algorithm.

[0083] And a horizontal comparison is made on the training cycle, parameter tuning duration, and overall model configuration cycle of each model, and finally the following statistical table is formed:

[0084] Table 1

[0085]

[0086] Combined with Table 1 of the above statistics, it is not difficult to find that whether it is from the training cycle, parameter tuning difficulty or response time, it is more reasonable and efficient to use GMM as the main anomaly detection method; and then according to the algorithm requirements initially formulated by the group, the group finally determines to adopt the GMM Gaussian distribution anomaly detection algorithm as the early warning algorithm for this scheme. Thus, the best scheme is determined.

[0087] To ensure that the input features meet the requirements of the GMM algorithm and basically satisfy the Gaussian distribution, this embodiment will start from the point selection method, explore the relationship between the input feature dimensions on the premise of ensuring good correlation of the input measurement points, and standardize and unify the point selection method to continuously improve the accuracy and generalization ability of the model.

[0088] By calculating the correlation matrix and combining multiple rounds of tests, the point selection methods of the subsequent models are classified into the following four types:

[0089] (1) Homonymous measurement points (temperature, pressure)

[0090] (2) Measurement points at the same location (vibration

[0091] (3) Measurement points of the same type (important operating parameters of equipment)

[0092] (4) Correlated measurement points (objects within the control system)

[0093] Based on the model alarm constructed by the above point selection methods, while ensuring a reasonable number of measurement points, there is also a strong correlation between points. Therefore, when any measurement point of a feature dimension undergoes a mutation or deviation, accurate early warning can be achieved.

[0094] The input feature dimensions are deployed according to the measurement points at the same location, of the same type, and homonymous measurement points.

[0095] This solution realizes absolute alarm for the measured point data through a trend and limit monitoring process that can dynamically adjust the change rate threshold, upper limit value, and lower limit value, ensuring the safe operation of the equipment; through the combination of judgment logics for multiple measured point objects, it further realizes the comprehensive monitoring of the equipment operating conditions from a mechanistic perspective, realizes the overall fault prediction of the equipment, and conducts early warning processing; through the use of anomaly detection algorithms to compare and analyze the changes in historical operating conditions of the measured point data, it realizes the perception and early warning of anomalies from the perspective of historical data; overall, it completes equipment early warning from three perspectives, greatly improving the early warning accuracy and reliability of power plant equipment.

[0096] Embodiment 2

[0097] This embodiment also provides an early warning system based on big data operation and maintenance on the power plant side, including:

[0098] A lightweight data middle platform for collecting, transmitting, and storing time-series data on the power plant side, and performing data governance and data visualization;

[0099] An enabling tool middle platform for the analysis and alarm processing of time-series data;

[0100] The enabling tool middle platform includes:

[0101] An absolute alarm module: for the measured point data in the time-series data, through trend monitoring of the change rate of the measured point data, comparing the measured point data with the corresponding upper and lower limit values for limit monitoring to obtain an absolute alarm result; during the trend monitoring and limit monitoring processes, receiving the actual equipment operation feedback to dynamically adjust the change rate threshold, upper limit value, and lower limit value;

[0102] A combined judgment module: setting multiple measured point objects according to the experience of the operator to compare the amplitude intervals or change amplitudes, and combining them through judgment logics to obtain a combined early warning result;

[0103] A relative alarm module: selecting various measured point data from the time-series data, screening and verifying the correlation of the selected measured point data, selecting various measured point data with strong correlation as parameters, using the historical operating conditions as samples to obtain the abnormal change intervals of each parameter under the same operating conditions as the current one, and comparing the real-time results of each parameter with the corresponding abnormal change intervals to obtain a relative alarm result;

[0104] A closed-loop diagnosis module: obtaining the absolute alarm result, combined early warning result, and relative alarm result, generating a record in the alarm list, and associating the corresponding expert knowledge base according to the equipment and its fault type to perform closed-loop adjustment on the absolute alarm step, combined judgment step, and relative alarm step.

[0105] Specifically, the limit monitoring in the absolute alarm module is implemented through a configuration function block, which includes input constants, input measurement points, judgment elements, and logic elements. The judgment elements are used to compare and judge the input measurement points and input constants, and the logic elements are used to combine the output results of each judgment element to obtain the absolute alarm result;

[0106] The configuration function block also includes an if-else function block, which is used to execute other judgment elements in other cases;

[0107] The combined judgment module makes combined judgments on multiple input measurement points through the configuration function block to obtain the combined early warning result;

[0108] The logic elements include AND gate logic elements and OR gate logic elements, and the judgment elements include greater than judgment elements, less than judgment elements, multiplication judgment elements, division judgment elements, addition judgment elements, and subtraction judgment elements.

[0109] The early warning system based on the big data operation and maintenance on the power plant side constructed in this embodiment is named the Nebula system. The essence of the Nebula system is an early warning system based on the big data platform on the power plant side, relying on the data transmission platform and the data analysis platform, and integrating relevant specific business requirements (such as configuration calculation, anomaly detection, logical judgment, etc.). It can digitize important information such as equipment parameters, historical operating conditions data (trends, limits), and operating experience and permanently store it in the system as the basis for judging whether the actual working conditions are normal.

[0110] To achieve the above functions, seamless connection from the data end to the application end needs to be realized first. We have created two basic tool middle platforms on the big data platform side: the lightweight data middle platform and the empowerment tool middle platform. The function of the lightweight data middle platform is to complete data transmission, data collection and storage, data governance, and data visualization functions; it is convenient to classify data types in different forms (relational databases: MySQL, SQLSERVER, etc.; non-relational databases: InfuxDB, IOTDB, etc.) into unified data objects so that they can be called in each function module.

[0111] The empowerment tool middle platform focuses more on the analysis of time-series data, such as equipment management, configuration management, model management, baseline functions, and BI visualization. Among them, the measurement point management realizes the management of real-time / historical data of all time-series measurement point data, presents it to the user in the form of a line trend chart or a data list, and supports the export of batch measurement point data. Compared with the original SIS platform, this platform has better data loading, data output, and data calculation efficiency, and can be customized and developed and extended, and serves various specific businesses with the help of the powerful cloud computing ability of the platform;

[0112] The configuration management realizes the real-time calculation of time-series measurement points. By constructing formulas or using preset internal algorithms, another time-series measurement point is calculated and generated, and then stored in the measurement point management list. In this way, the output value of this calculation can be reused in other functional modules. Compared with DCS, it has stronger replicability and migratability. And because the output value realizes real-time tracking, all secondary calculation measurement points can be regarded as original measurement points.

[0113] In practical applications, the data platform completes the generation and storage of data objects by collecting business data; for time-series data, after the collection is completed, it is necessary to synchronously manage and verify the measurement point information. After completing the above work, the data will directly enter the front-end application side. The empowerment tool platform provides a high-timeliness and high-concurrency secondary empowerment configuration platform based on time-series data, integrating a series of practical functions such as measurement point management, calculation indicators, logical judgment, time management, and AI prediction. In specific production, we realized the real-time calculation of a series of characterization data such as coal consumption, differential pressure, terminal difference, efficiency, enthalpy entropy, etc. by constructing calculation formulas. These calculation results will be permanently retained in the time-series database to provide data support for subsequent exploration of the economic operation of the unit.

[0114] The Nebula system combines the model construction and alarm setting functions of the empowerment tool platform on the basis of the data platform to realize an all-round real-time warning system from the system to the equipment and then from the equipment to the measurement points. In practical applications, the operator judges potential abnormal situations by observing the real-time status of the unit equipment. After discovering a red alarm, clicking on the system parent node can reach the equipment-level interface to trace the location where the problem occurs.

[0115] The alarm function of the Nebula system is realized by the above absolute alarm module, combined judgment module and relative alarm module. For the specific content and beneficial effects, reference can be made to the above method embodiments, which will not be elaborated here.

[0116] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the protection scope determined by the claims.

Claims

1. An early warning method based on big data operation and maintenance at the power plant side, characterized in that: The following steps are involved: Data collection steps: Collect time series data from the power plant side; Absolute alarm step: for the measuring point data in the time series data, by performing trend monitoring on the rate of change of the measuring point data, comparing the measuring point data with the corresponding upper limit value and lower limit value, performing limit monitoring, and obtaining an absolute alarm result; during the trend monitoring and limit monitoring process, receiving actual equipment operation feedback, and dynamically adjusting the change rate threshold, upper limit value and lower limit value; Combined judgment step: according to the experience of the operator, multiple measuring point objects are set to compare the amplitude range or change amplitude, and combined through judgment logic to obtain a combined warning result; Relative alarm step: using an anomaly detection algorithm, selecting multiple measuring point data from the time series data, screening and verifying the correlation of the selected measuring point data, selecting multiple measuring point data with strong correlation as parameters, and using historical operating conditions as samples to train the anomaly detection algorithm model, obtaining the abnormal change interval of each parameter under the same operating condition as the current operating condition, comparing the real-time results of each parameter with the corresponding abnormal change interval, and obtaining a relative alarm result; Closed-loop diagnosis step: obtain the absolute alarm result, combined warning result and relative alarm result, generate records in the alarm list, and associate the corresponding expert knowledge base according to the equipment and its fault type to perform closed-loop adjustment on the absolute alarm step, combined judgment step and relative alarm step.

2. The early warning method based on big data operation and maintenance at the power plant side according to claim 1 is characterized in that: The data collection step also includes: real-time calculation of characterization data on the collected time series data for judgment in the absolute alarm step, the combined judgment step and the relative alarm step, wherein the characterization data includes coal consumption, differential pressure, terminal difference, efficiency and enthalpy entropy.

3. The early warning method based on big data operation and maintenance at the power plant side according to claim 1 is characterized in that: The limit value monitoring in the absolute alarm step is implemented by calculating the configuration function block, which includes input constants, input measuring points, judgment primitives and logic primitives. The input measuring points and input constants are compared and judged by the judgment primitives, and the output results of each judgment primitive are combined by the logic primitives to obtain an absolute alarm result.

4. The early warning method based on big data operation and maintenance at the power plant side according to claim 3 is characterized in that: The configuration function block also includes an if-else function block for executing other judgment primitives in other situations.

5. The early warning method based on big data operation and maintenance at the power plant side according to claim 4 is characterized in that: The combined judgment step performs combined judgment on multiple input measurement points through the configuration function block to obtain a combined warning result.

6. The early warning method based on big data operation and maintenance at the power plant side according to claim 5 is characterized in that: The configuration function blocks for combined judgment of unit load and coal mill roller lubricating oil temperature include: Unit load input measurement point, 260MW input constant, first greater than judgment element, if-else function block, code block A and code block B; the unit load input measurement point and 260MW input constant are respectively input into the first greater than judgment element, the output of the first greater than judgment element is connected to the if-else function block, the output of the if-else function block is respectively connected to code block A and code block B, when the unit load input measurement point is greater than the 260MW input constant, code block A is executed, otherwise code block B is executed; The code block A includes a coal mill roller lubricating oil temperature input measurement point, a 90 degree Celsius input constant, a -10 degree Celsius input constant, a second greater than judgment primitive, a less than judgment primitive and an AND gate logic primitive. The coal mill roller lubricating oil temperature input measurement point and the 90 degree Celsius input constant are respectively input into the second greater than judgment primitive, the coal mill roller lubricating oil temperature input measurement point and the -10 degree Celsius input constant are respectively input into the less than judgment primitive, the second greater than judgment primitive and the less than judgment primitive are respectively input into the AND gate logic primitive. If any condition that the coal mill roller lubricating oil temperature input measurement point is greater than 90 degrees Celsius or less than -10 degrees Celsius is met, the output state quantity 1 corresponds to the alarm state, otherwise the output state quantity is 0; The code block B includes a 0 input constant, a 1 input constant and a third greater than judgment primitive, wherein the 0 input constant and the 1 input constant are respectively input into the third greater than judgment primitive, and the third greater than judgment primitive outputs a state value of false.

7. The early warning method based on big data operation and maintenance at the power plant side according to claim 3 is characterized in that: The logic primitives include AND gate logic primitives and OR gate logic primitives, and the judgment primitives include greater than judgment primitives, less than judgment primitives, multiplication judgment primitives, division judgment primitives, addition judgment primitives and subtraction judgment primitives.

8. The early warning method based on big data operation and maintenance at the power plant side according to claim 1 is characterized in that: In the process of judging whether the coal mill is blocked by coal in the combined judgment step, the coal mill current, primary air differential pressure, air-powder mixture pressure and outlet air-powder flow rate are respectively used as input measurement points, and trend monitoring and limit monitoring are respectively performed. Then, the various monitoring results are comprehensively combined to judge whether the coal mill is likely to be blocked by coal.

9. An early warning system based on big data operation and maintenance at the power plant side, characterized in that: include: Lightweight data center, used to collect, transmit and store time series data from power plants, and perform data governance and data visualization; Empowerment tool platform for time series data analysis and alarm processing; The enabling tool platform includes: Absolute alarm module: for the measuring point data in the time series data, by performing trend monitoring on the rate of change of the measuring point data, comparing the measuring point data with the corresponding upper limit and lower limit, performing limit monitoring, and obtaining an absolute alarm result; during the trend monitoring and limit monitoring process, receiving actual equipment operation feedback, and dynamically adjusting the change rate threshold, upper limit and lower limit; Combined judgment module: according to the experience of the operator, multiple measuring point objects are set to compare the amplitude range or change range, and combined through judgment logic to obtain a combined warning result; Relative alarm module: select multiple measuring point data from the time series data, screen and verify the correlation of the selected measuring point data, select multiple measuring point data with strong correlation as parameters, and use historical operating conditions as samples to obtain the abnormal change interval of each parameter under the same working condition as the current working condition, compare the real-time results of each parameter with the corresponding abnormal change interval, and obtain a relative alarm result; Closed-loop diagnosis module: obtains the absolute alarm results, combined warning results and relative alarm results, generates records in the alarm list, and associates the corresponding expert knowledge base according to the equipment and its fault type to make closed-loop adjustments to the absolute alarm steps, combined judgment steps and relative alarm steps.

10. The early warning system based on big data operation and maintenance at the power plant side according to claim 9 is characterized in that: The limit value monitoring in the absolute alarm module is realized by calculating the configuration function block, which includes input constants, input measuring points, judgment primitives and logic primitives. The input measuring points and input constants are compared and judged by the judgment primitives, and the output results of each judgment primitive are combined by the logic primitives to obtain the absolute alarm result. The configuration function block also includes an if-else function block for executing other judgment primitives in other situations; The combined judgment module performs combined judgment on multiple input measurement points through the configuration function block to obtain a combined warning result; The logic primitives include AND gate logic primitives and OR gate logic primitives, and the judgment primitives include greater than judgment primitives, less than judgment primitives, multiplication judgment primitives, division judgment primitives, addition judgment primitives and subtraction judgment primitives.

Images