Intelligent monitoring disk-based anomaly prediction method, system, device and storage medium

Abstract

The application discloses an abnormality prediction method, system and device based on intelligent monitoring and a storage medium, and comprises the following steps: according to a to-be-monitored device, selecting corresponding device parameters, and establishing an abnormality prediction model; obtaining boundary parameters of the to-be-monitored device, inputting the obtained boundary parameters and preset early warning parameters into the abnormality prediction model, and obtaining an abnormal probability value of the early warning parameters under the given boundary parameters; comparing the abnormal probability value with a global probability threshold value, and if the abnormal probability value is less than the global probability threshold value, triggering an abnormality prompt by the abnormality prediction model. The application can normalize the sensor measurement data probability with the change of the device working condition, avoid the misleading caused by the uneven distribution of the training data to the health degree evaluation, be used for solving the multi-mode problem, avoid the work of manually setting the parameter alarm threshold value by setting the global probability threshold value, and further improve the prediction efficiency and the prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for automatic control systems, and in particular to an anomaly prediction method, system, device, and storage medium based on intelligent monitoring. Background Technology

[0002] Currently, conventional intelligent monitoring systems for thermal power plants mainly employ two technical approaches. The first approach involves establishing characteristic models of the power plant's main systems and detecting anomalies by analyzing whether the actual relationships between measured values ​​conform to the theoretical relationships between variables. This method is easy to quickly identify the causes of anomalies, but it is prone to problems such as underdetermining due to incomplete sensors, overdetermining due to sensor redundancy, and discrepancies between theory and reality due to sensor measurement errors. Furthermore, because establishing characteristic models requires extensive specialized knowledge, equipment coverage is typically low, failing to reduce the workload of manual monitoring. The second approach utilizes historical data from the power plant's DCS to establish data-driven predictive models, detecting anomalies by analyzing the deviation between actual and predicted values ​​of each measured value or the probability of a set of data occurring.

[0003] The second category of technical approaches is mainly divided into two types based on the machine learning algorithms employed. The first type is supervised learning algorithms, which directly predict output parameters based on equipment boundary parameters and issue warnings by comparing the deviation between predicted and measured values, combined with manually set thresholds. The drawback of this method is its difficulty in handling multi-mode system problems, such as multiple stable states of equipment under the same boundary parameters, or situations where the health status of equipment parameters within their defined domain exhibits a non-Gaussian distribution with prediction deviations. The second type is unsupervised learning algorithms, which typically calculate the probability of simultaneous occurrence of related equipment parameters or their reconstruction error, and then issue warnings based on manually set thresholds. This type of method is inherently unsuitable for addressing situations where the operating conditions of thermal power units frequently change due to external factors such as load commands; that is, abundant training data does not necessarily imply a high probability of health. Furthermore, both of these methods require manually setting alarm thresholds, which increases the experience required for modeling and the workload involved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an anomaly prediction method, system, device, and storage medium based on intelligent monitoring. This invention uses a multi-mode variable operating condition analog quantity probability prediction algorithm to normalize the probability of sensor measurement data occurrence as the equipment operating condition changes, thereby avoiding the misleading effect of uneven training data distribution on health assessment. It can be used to solve multi-mode problems, and can also avoid the need for manual setting of alarm thresholds for each parameter by setting a global probability threshold.

[0005] In a first aspect, embodiments of the present invention provide an anomaly prediction method based on intelligent monitoring, comprising:

[0006] Based on the device to be monitored, select the corresponding device parameters, and establish an anomaly prediction model based on the device parameters, which include boundary parameters and early warning parameters;

[0007] Obtain the boundary parameters of the device to be monitored, and input the obtained boundary parameters and the preset warning parameters into the anomaly prediction model to obtain the anomaly probability value of the warning parameters occurring when the boundary parameters are given;

[0008] The anomaly probability value is compared with the global probability threshold. If it is less than the global probability threshold, the anomaly prediction model triggers an anomaly alert.

[0009] The anomaly probability value is calculated using the following formula:

[0010]

[0011] In the formula, y + Indicates the warning parameter, x + Let K represent the boundary parameter, y represent the historical data of the warning parameter, K represent the hyperparameter, and Δy represent the infinitesimal measure in the vector space of the warning parameter. max This represents the upper limit of the warning parameter in historical data, y min α represents the lower limit of the warning parameter in historical data. k μ represents the weighting coefficient. k Let Σ represent the mean vector. k Represents the covariance matrix. Indicated by μ k The mean vector, with Σ k The normal distribution of the covariance matrix in y i ,x i The probability density at that location.

[0012] Furthermore, the step of selecting appropriate equipment parameters based on the equipment to be monitored, and establishing an anomaly prediction model based on the equipment parameters, includes:

[0013] Based on the device to be monitored, the external parameters representing the device's operating status are used as boundary parameters, and the internal parameters of the device to be monitored are used as early warning parameters;

[0014] The model parameters of the anomaly prediction model are set, including hyperparameters, weight coefficients, mean vectors, and covariance matrices. The initial values ​​of the hyperparameters are calculated from the number of device parameters. The number of weight coefficients, mean vectors, and covariance matrices are the same as the values ​​of the hyperparameters. The initial values ​​of the weight coefficients, mean vectors, and covariance matrices are all obtained by random initialization.

[0015] The historical device parameters of the device to be monitored are obtained. Based on the historical device parameters and preset iteration conditions, the weight coefficients, the mean vector and the covariance are iteratively calculated to obtain the final weight coefficients, the mean vector and the covariance matrix.

[0016] The anomaly prediction model is generated based on the weight coefficients, mean vector, and covariance matrix obtained through iterative calculation.

[0017] Furthermore, the hyperparameters are expressed using the following formula:

[0018]

[0019] In the formula, m represents the number of boundary parameters, and n represents the number of warning parameters;

[0020] The weighting coefficients are expressed by the following formula:

[0021]

[0022] The mean vector is represented by the following formula:

[0023]

[0024] The covariance matrix is ​​expressed by the following formula:

[0025]

[0026] In the formula, x i y represents the historical data of the i-th set of boundary parameters. i Let N represent the historical data of the i-th group of warning parameters, and N represent the N groups of historical data.

[0027]

[0028]

[0029] Indicated by μ k The mean vector, with Σ k The normal distribution of the covariance matrix in y i ,x iThe probability density at that location.

[0030] Further, the step of acquiring the boundary parameters of the device to be monitored, and inputting the acquired boundary parameters and the preset warning parameters into the anomaly prediction model to obtain the anomaly probability value of the warning parameters occurring given the boundary parameters includes:

[0031] Obtain the boundary parameters of the device to be monitored and generate a boundary parameter vector;

[0032] Select the parameter to be monitored from the warning parameters as the first warning parameter, and use the remaining warning parameters as the second warning parameter. Obtain the second warning parameter of the device to be monitored, and generate a warning parameter vector based on the obtained second warning parameter and the preset first warning parameter.

[0033] The boundary parameter vector and the warning parameter vector are input into the anomaly prediction model for prediction, and the anomaly probability value of the first warning parameter occurring when the boundary parameters are given is obtained.

[0034] Furthermore, after the anomaly prediction model triggers an anomaly alert if the probability is less than the global probability threshold, the method further includes:

[0035] Based on the historical equipment parameters of the device to be monitored, a health score curve for the device to be monitored is generated.

[0036] The abnormal probability value is input into the health score curve to obtain the health score value of the device to be monitored.

[0037] Furthermore, based on the monitoring time of the warning parameter, the warning parameter is divided into a first sub-warning parameter and a second sub-warning parameter;

[0038] Input the historical device parameters corresponding to the first sub-early warning parameter into the anomaly prediction model, calculate the first anomaly probability value, and generate a first health score curve based on the first anomaly probability value and a preset first slope.

[0039] The historical device parameters corresponding to the second sub-early warning parameter are input into the anomaly prediction model to calculate the second anomaly probability value, and a second health score curve is generated based on the second anomaly probability value and the preset second slope.

[0040] Furthermore, the preset iteration condition is that the values ​​of the weight coefficients, the mean vector, and the covariance matrix remain unchanged.

[0041] Secondly, embodiments of the present invention provide an anomaly prediction system based on intelligent monitoring, comprising:

[0042] An anomaly prediction model is used to select appropriate equipment parameters based on the equipment to be monitored, and to establish an anomaly prediction model based on the equipment parameters, which include boundary parameters and early warning parameters.

[0043] An anomaly probability calculation module is used to obtain the boundary parameters of the device to be monitored, input the obtained boundary parameters and the preset warning parameters into the anomaly prediction model, and obtain the anomaly probability value of the warning parameters occurring when the boundary parameters are given.

[0044] The threshold comparison module is used to compare the anomaly probability value with the global probability threshold. If it is less than the global probability threshold, the anomaly prediction model triggers an anomaly prompt.

[0045] The anomaly probability value is calculated using the following formula:

[0046]

[0047] In the formula, y + Indicates the warning parameter, x + Let K represent the boundary parameter, y represent the historical data of the warning parameter, K represent the hyperparameter, and Δy represent the infinitesimal measure in the vector space of the warning parameter. max This represents the upper limit of the warning parameter in historical data, y min α represents the lower limit of the warning parameter in historical data. k μ represents the weighting coefficient. k Let Σ represent the mean vector. k Represents the covariance matrix. Indicated by μ k The mean vector, with Σ k The normal distribution of the covariance matrix in y i ,x i The probability density at that location.

[0048] Thirdly, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0049] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0050] This invention provides an anomaly prediction method, system, device, and storage medium based on intelligent monitoring. Compared with existing technologies, this invention proposes a multi-mode variable operating condition analog quantity probability prediction algorithm, which can normalize the probability of sensor measurement data occurrence as the equipment operating condition changes, thereby avoiding the misleading effect of uneven distribution of training data on health assessment. It can not only be used to solve multi-mode problems, but also avoid the work of manually setting alarm thresholds for each parameter by setting a global probability threshold, reducing the experience requirements and workload of modeling. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating the anomaly prediction method based on intelligent monitoring in an embodiment of the present invention.

[0052] Figure 2 This is a schematic diagram of the existing feedwater heater of a thermal power unit;

[0053] Figure 3 This is a schematic diagram of the water temperature monitoring panel at the water supply outlet in an embodiment of the present invention;

[0054] Figure 4 This is a schematic diagram of the heater water level monitoring panel curve in an embodiment of the present invention;

[0055] Figure 5 This is a schematic diagram of the hydrophobic outlet temperature monitoring curve in an embodiment of the present invention;

[0056] Figure 6 This is a schematic diagram of a health score curve in an embodiment of the present invention;

[0057] Figure 7 This is a schematic diagram of another health score curve in an embodiment of the present invention;

[0058] Figure 8 This is a schematic diagram of the structure of the anomaly prediction system based on intelligent monitoring in an embodiment of the present invention;

[0059] Figure 9 This is an internal structural diagram of the computer device in an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0061] Please see Figure 1The first embodiment of the present invention proposes an anomaly prediction method based on intelligent monitoring, comprising steps S10 to S30:

[0062] Step S10: Select the corresponding equipment parameters according to the equipment to be monitored, and establish an anomaly prediction model based on the equipment parameters, including boundary parameters and early warning parameters.

[0063] This invention provides a multi-mode variable operating condition analog probability prediction algorithm for intelligent monitoring panels in thermal power plants. It is an engineering method that establishes a probabilistic model of early warning parameters based on boundary parameters, and can quickly estimate the probability of early warning parameters occurring given the boundary parameters. To achieve this, boundary parameters and early warning parameters must first be selected. For a specific piece of equipment in a thermal power unit, the selection of boundary parameters is closely related to the equipment under study. Different boundary parameters are selected based on the scope of the equipment. If the unit is the research object, external parameters affecting the unit's operating state, such as unit load and environmental parameters, can be selected as boundary parameters. If the feedwater heater is the research object, external parameters affecting the equipment's operating state, such as extraction steam pressure, feedwater flow rate, and feedwater temperature, can be selected as boundary parameters. Early warning parameters are selected from the measured parameter variables of the equipment under study, and boundary parameters cannot be selected as early warning parameters. For example, if the unit is the research object, internal parameters such as main steam temperature and main steam pressure can be selected as early warning parameters. If the feedwater heater is the research object, internal parameters such as heater water level, feedwater outlet temperature, and condensate temperature can be selected as early warning parameters.

[0064] After selecting the equipment parameters, an anomaly prediction model is built based on these parameters. First, the model parameters are determined, including hyperparameters, weight coefficients, mean vector, and covariance matrix. The values ​​of the hyperparameters are calculated from the number of equipment parameters; that is, the hyperparameter K is a positive integer, and its values ​​are:

[0065]

[0066] In the formula, m represents the number of boundary parameters, and n represents the number of warning parameters. The floor symbol;

[0067] Then randomly initialize K weight coefficients α. k Mean vector μ k The covariance matrix Σ k , where k = 1, 2, 3, ..., K.

[0068] Since the model parameters are initialized randomly, the model also needs to be trained using historical data. Specifically, this involves obtaining the historical device parameters of the device to be monitored. Assuming there are N sets of historical data for the boundary parameter x and the warning parameter y, let the i-th set be x. i =[x i,1 ,x i,2 ,...,x i,m ]、y i =[y i,1 ,y i,2 ,...,y i,n Then repeat steps A and B below until the weighting coefficient α is reached. k Mean vector μ k The covariance matrix Σ k It will no longer change.

[0069] Step A:

[0070]

[0071] Where i = 1, 2, ..., N, k = 1, 2, ..., K, Represented by μ k The mean vector, with Σ k The normal distribution of the covariance matrix in y i ,x i The probability density at a given location is calculated using the following formula:

[0072]

[0073] Step B:

[0074]

[0075] By training the parameters of the anomaly prediction model through the above steps, stable model parameters can be obtained. After obtaining the determined model parameters, the next step of anomaly probability calculation can be performed.

[0076] Step S20: Obtain the boundary parameters of the device to be monitored, input the obtained boundary parameters and the preset warning parameters into the anomaly prediction model, and obtain the anomaly probability value of the warning parameters occurring when the boundary parameters are given.

[0077] Given the working condition x = x + Below, the warning parameter y = y + The probability P(y) + |x + This refers to the anomaly probability value to be calculated in this invention. For example, taking a generator unit as an example, given the operating condition as [x] 1+ ,x 2+This refers to the unit load x1 being equal to a given load value x. 1+ Ambient temperature x2 equals the given temperature x 2+ , [y 1+ ,y 2+ This refers to the occurrence of main steam temperature y. 1+ And the main steam pressure is y 2+ For example, taking a water heater as an example, given the operating condition [x] 1+ ,x 2+ ,x 3+ This refers to the extraction steam pressure x1 being equal to a given pressure x. 1+ Water supply flow rate x2 equals the given flow rate x 2+ The water supply temperature x3 is equal to the given temperature x 3+ , [y 1+ ,y 2+ ,y 3+ This refers to the situation where the heater water level is y. 1+ The water supply outlet temperature is y 2+ And the hydrophobic temperature is y 3+ The situation.

[0078] Combining the probability formula, we can obtain P(y) + |x + )for:

[0079]

[0080] In engineering implementation, solving the multiple integrals in the denominator of the above formula is complex. P(y) can be approximated using the following formula. + |x + ):

[0081]

[0082] Where, Δy = Δy1Δy2…Δy n For the infinitesimal measure y in the vector space of the warning parameter, min =[y 1,minn ,y 2,min ,...,y n,min ]、y max =[y 1,max ,y 2,max ,...,y n,max ] are the lower and upper bound vectors of the warning parameter vector in the training samples, respectively. max y min Take the upper and lower limits of each warning parameter in the historical dataset, and take the infinitesimal measure.

[0083] Step S30: Compare the anomaly probability value with the global probability threshold. If it is less than the global probability threshold, the anomaly prediction model triggers an anomaly alert.

[0084] To avoid the increased experience and workload required for modeling due to manually setting alarm thresholds, this invention uses a global probability threshold instead of manual setting. Specifically, the anomaly prediction model will trigger an anomaly alert when the following conditions are met:

[0085] P(y + |x + )<ε

[0086] Wherein, ε∈(0,1) is the global probability threshold, which is set according to the actual situation of the thermal power plant. In this invention, it is preferably set to 0.01.

[0087] The following uses the #2 high-pressure heater of a thermal power unit as an example to illustrate the practical application of the anomaly prediction method provided by this invention in monitoring. Please refer to [link / reference]. Figure 2 The typical structure of a feedwater heater for a thermal power unit is shown, wherein 1 is the casing, 2 is the U-shaped tube, 3 is the tube bundle support, 4 is the liquid level, 5 is the liquid level sensor, 6 is the feedwater outlet, 7 is the steam inlet, 8 is the baffle of the steam cooling section, 9 is the baffle of the cooling section, 10 is the control valve, 11 is the top water chamber, 12 is the bottom water chamber, 13 is the drain outlet, and 14 is the feedwater inlet.

[0088] During equipment operation, low-temperature feedwater flows into the bottom water chamber 12 through feedwater inlet 14, passes parallel to the U-shaped tube 2, absorbs heat from the metal wall, and finally enters the top water chamber 11 and flows to the next stage heater. The steam side is divided into three zones based on steam phase: a superheated steam cooling zone, a condensation zone, and a condensate cooling zone. Superheated steam first enters the superheated steam cooling zone through steam inlet 7, and heats the U-shaped tube segment by segment through cross-convective heating via baffles 8. After the superheated steam cools to saturation, it enters the condensation zone, where steam molecules condense into water droplets on the surface of the U-shaped tube and fall to the bottom of the heater. Condensate enters the condensate cooling zone through a water seal, and heats the U-shaped tube segment by segment through cross-convective heating via baffles 9. The supercooled condensate is finally discharged from the heater and flows into the next stage heater through control valve 10. Control valve 10 controls the condensate flow rate by throttling, adjusting the heater water level to achieve a preset target water level. The relevant measuring points for the heater include feedwater flow rate, feedwater inlet temperature, feedwater outlet temperature, steam inlet pressure, heater water level, and condensate outlet temperature.

[0089] For the boundary parameters, this embodiment selects three parameters—feedwater flow rate, feedwater inlet temperature, and steam inlet pressure—to form the boundary parameter vector [x1, x2, x3]. For the warning parameters, three parameters—feedwater outlet temperature, heater water level, and condensate outlet temperature—are selected to form the warning parameter vector [y1, y2, y3]. Given that the number of boundary parameters is m = 3 and the number of warning parameters is n = 3, the hyperparameters can be obtained.

[0090] Three months of historical operational data for the above six parameters were selected, with a sampling interval of 1 minute. Four weighting coefficients α were obtained through random initialization. k Mean vector μ k The covariance matrix Σ k That is, the weighting coefficient is:

[0091] α1=0.22, α2=0.23, α3=0.28, α4=0.27;

[0092] The mean vector is:

[0093] μ1=[231,0.6,199,114,205,5.7] T ,μ2=[230,0.5,181,120,214,5.4] T ,

[0094] μ3=[221,0.63,210,103,201,6.1] T ,μ4=[229,0.59,210,132,215,5.8] T ;

[0095] The covariance matrix is:

[0096]

[0097] Repeat steps A and B until α 1～4 μ 1～4 and Σ 1～4 No further changes are observed, thus we obtain:

[0098] α1=0.79, α2=0.14, α3=0.06, α4=0.01;

[0099] μ1=[252,0.6,198,115,204,5.8] T ,μ2=[261,0.5,177,130,213,5.5] T ,

[0100] μ3=[234,0.52,185,103,201,6.1] T,μ4=[299,0.61,212,121,214,5.77] T ;

[0101]

[0102]

[0103] Based on historical data, the upper and lower limits of each warning parameter can be determined, namely:

[0104] y min =[230,0.35,164]、y max =[316,0.69,247],

[0105] In this case, the infinitesimal measure is taken as Δy = 0.0024.

[0106] Based on the parameters calculated above, an anomaly prediction model can be obtained with the high-pressure heater as the research object. Then, the boundary parameters and early warning parameters can be input into the anomaly prediction model to make anomaly predictions.

[0107] First, obtain the current boundary parameters of the high-pressure heater, namely, the real-time data of feedwater flow rate x1, feedwater inlet temperature x2, and steam inlet pressure x3. There are three warning parameters: feedwater outlet temperature y1, heater water level y2, and condensate outlet temperature y3. Anomaly predictions are then performed for these three warning parameters respectively:

[0108] ①Prediction of anomalies in water supply outlet temperature y1:

[0109] Obtain the measured values ​​of heater water level y2 and condensate outlet temperature y3. Based on the upper and lower limits of the warning parameters, calculate in real time P(y) for each value of feedwater outlet temperature y1 within the range of 230 to 316. + |x + );

[0110] ②Prediction of abnormal water level y2 in heater:

[0111] Obtain the measured values ​​of the feedwater outlet temperature y1 and the condensate outlet temperature y3. Based on the upper and lower limits of the warning parameters, calculate in real time the P(y) value for each value of the heater water level y2 within the range of 0.35 to 0.69. + |x + );

[0112] ③Prediction of anomalies in hydrophobic outlet temperature y3:

[0113] Obtain the measured values ​​of the feedwater outlet temperature y1 and the heater water level y1. Based on the upper and lower limits of the warning parameters, calculate in real time P(y) for each value of the condensate outlet temperature y3 within the range of 164 to 247. + |x+ ).

[0114] Then, based on the three sets of probability values ​​calculated at each time point, P(y) is plotted. + |x + Contour lines are drawn to obtain the following figures. Figure 3 The water temperature monitoring curve at the water supply outlet shown, and the attached... Figure 4 The heater water level monitoring curve and attached figure are shown. Figure 5 The diagram shows the temperature monitoring curve for the condensate outlet. Operators determine whether the equipment is operating normally by observing the relative position of the measured values ​​on the contour lines, i.e., P(y). + |x + If the elevation falls below the contour line corresponding to the global probability threshold, an alarm will be triggered. Furthermore, since the historical data simultaneously collects both high-pressure heater activation and deactivation operation modes, therefore, [the following is missing from the original text]. Figure 3 The given feedwater outlet temperature monitoring curve shows typical contour lines for multi-mode objects. That is, if the feedwater outlet temperature is in the higher or lower temperature range, it can be judged as normal, while if it is in the middle value, it will be judged as abnormal. This is consistent with the physical characteristics of the equipment and actual needs. This is a feature that conventional anomaly detection algorithms do not have.

[0115] Furthermore, to make the monitoring more efficient and accurate, this invention also provides a health scoring method for monitoring equipment, the specific steps of which are as follows:

[0116] Step S40: Generate a health score curve for the device to be monitored based on the historical device parameters of the device to be monitored.

[0117] Step S50: Input the abnormal probability value into the health score curve to obtain the health score value of the device to be monitored.

[0118] That is, calculating P(y) of the selected device in real time. + |x + Afterwards, a health score for the device can be obtained based on a preset scoring curve, with a value range of 0 to 100. The scoring curve provided by this invention is based on P(y + |x + The curve is a monotonically increasing curve with the x-axis representing the health score and the y-axis representing the health score. Its probability value is calculated based on historical data, and the design of the curve varies depending on the monitoring parameters.

[0119] For parameters that require long-term monitoring, the scoring curve can be designed as follows: Figure 6 As shown, the curve maintains a certain slope at all points, such that the health score varies with P(y). + |x +It has high sensitivity in each section, and the equipment's operating status can be clearly reflected in the health score, which helps operators to monitor it over a long period of time.

[0120] For parameters that do not require long-term monitoring and only need explicit alerts when an anomaly occurs, their scoring curves can be designed as follows: Figure 7 As shown, the characteristic of this type of parameter curve is based on P(y + |x + The comparison between y and ε divides the curve into two regions, where P(y) = ε. + |x + In the region where )≥ε, the health score is high and the curve slope is small, while in P(y) + |x + In the region where y < ε, the health score is low and the curve slope is steep; the score at the location of the global probability threshold ε is greater than 50. Therefore, under the condition of device health, i.e., P(y) < ε, the health score is low and the curve slope is steep. + |x + When )≥ε, the health score increases with P(y) + |x + The change in P(y) is not obvious, and operators do not need to monitor it. In the event of equipment malfunction, i.e., P(y) + |x + When ) < ε, P(y) + |x + Even small changes can be clearly reflected in the health score, helping operators to understand the actual operating status of the equipment.

[0121] The present invention provides an anomaly prediction method based on intelligent monitoring. Compared with traditional methods, which are difficult to handle multi-mode system problems and are not suitable for handling situations where the operating conditions of equipment change frequently due to external factors such as load commands, the present invention can solve multi-mode problems through the provided anomaly prediction model. Furthermore, by setting a global probability threshold, the work of manually setting alarm thresholds for each parameter is avoided, thereby reducing the experience requirements and workload of modeling and improving the prediction efficiency and accuracy of the model.

[0122] Please see Figure 8 Based on the same inventive concept, the second embodiment of the present invention provides an anomaly prediction system based on intelligent monitoring, comprising:

[0123] Anomaly prediction model 10 is used to select appropriate equipment parameters based on the equipment to be monitored, and to establish anomaly prediction model based on the equipment parameters, wherein the equipment parameters include boundary parameters and early warning parameters.

[0124] Anomaly probability calculation module 20 is used to obtain the boundary parameters of the device to be monitored, input the obtained boundary parameters and the preset warning parameters into the anomaly prediction model, and obtain the anomaly probability value of the warning parameters occurring when the boundary parameters are given.

[0125] The threshold comparison module 30 is used to compare the anomaly probability value with the global probability threshold. If the anomaly probability value is less than the global probability threshold, the anomaly prediction model triggers an anomaly prompt.

[0126] The technical features and effects of the anomaly prediction system based on intelligent monitoring proposed in this invention are the same as those of the method proposed in this invention, and will not be repeated here. Each module in the above-mentioned anomaly prediction system based on intelligent monitoring can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or it can be stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0127] Please see Figure 9 The diagram illustrates the internal structure of a computer device in one embodiment. This computer device can specifically be a terminal or a server. The computer device includes a processor, memory, network interface, display, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an anomaly prediction method based on intelligent monitoring. The display screen of the computer device can be a liquid crystal display (LCD) or an e-ink display. The input devices of the computer device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse, etc.

[0128] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than shown in the diagram, or combine certain components, or have the same component arrangement.

[0129] Furthermore, embodiments of the present invention also propose a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0130] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0131] In summary, this invention provides an anomaly prediction method, system, device, and storage medium based on intelligent monitoring. The anomaly prediction method involves selecting appropriate device parameters based on the device to be monitored, and establishing an anomaly prediction model based on these parameters. The device parameters include boundary parameters and warning parameters. The boundary parameters of the device to be monitored are acquired, and the acquired boundary parameters and preset warning parameters are input into the anomaly prediction model to obtain the anomaly probability value of the warning parameter occurring given the boundary parameters. The anomaly probability value is compared with a global probability threshold; if it is less than the global probability threshold, the anomaly prediction model triggers an anomaly alert. This invention can normalize the probability of sensor measurement data occurrence according to changes in device operating conditions, avoiding the misleading effect of uneven training data distribution on health assessment. It can not only solve multi-mode problems but also avoid the need for manually setting alarm thresholds for each parameter by setting a global probability threshold, further improving prediction efficiency and accuracy.

[0132] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0133] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. An abnormality prediction method based on intelligent monitoring, characterized by, The method comprises the following steps: According to the equipment to be monitored, the corresponding equipment parameters are selected, and an abnormality prediction model is established according to the equipment parameters, wherein the equipment parameters include boundary parameters and early warning parameters; The boundary parameters of the equipment to be monitored are obtained, and the obtained boundary parameters and preset early warning parameters are input into the abnormality prediction model to obtain an abnormal probability value of the early warning parameters under the given boundary parameters; The abnormal probability value is compared with a global probability threshold value, and if it is less than the global probability threshold value, the abnormality prediction model triggers an abnormality prompt; Wherein, the abnormal probability value is calculated by the following formula: wherein y + represents a warning parameter, x + represents a boundary parameter, y represents historical data of a warning parameter, and K represents a hyperparameter, represents a microelement measure in a warning parameter vector space, y max represents an upper limit of a warning parameter in historical data, y min represents a lower limit of a warning parameter in historical data, represents a weight coefficient, represents a mean vector, represents a covariance matrix, represents a probability density of a normal distribution with as a mean vector and as a covariance matrix at . Wherein, the step of selecting the corresponding equipment parameters according to the equipment to be monitored and establishing the abnormality prediction model according to the equipment parameters comprises: According to the equipment to be monitored, the external parameters representing the running state of the equipment are taken as the boundary parameters, and the internal parameters of the equipment to be monitored are taken as the early warning parameters; The model parameters of the abnormality prediction model are set, wherein the model parameters include hyperparameters, weight coefficients, mean vectors and covariance matrices, wherein the initial value of the hyperparameters is calculated from the number of equipment parameters, the number of weight coefficients, mean vectors and covariance matrices is the same as the value of the hyperparameters, and the initial values of the weight coefficients, mean vectors and covariance matrices are obtained by random initialization method; The historical equipment parameters of the equipment to be monitored are obtained, and the weight coefficients, mean vectors and covariance matrices are iteratively calculated according to the historical equipment parameters and preset iteration conditions to obtain the final weight coefficients, mean vectors and covariance matrices; The abnormality prediction model is generated according to the weight coefficients, mean vectors and covariance matrices obtained by iteration calculation; The hyperparameters are represented by the following formula: Wherein, m represents the number of boundary parameters, and n represents the number of early warning parameters; The weight coefficients are represented by the following formula: The mean vectors are represented by the following formula: The covariance matrices are represented by the following formula: wherein represents the i-th set of boundary parameter historical data, y i represents the i-th set of early warning parameter historical data, N represents N sets of historical data, wherein denotes the probability density of a normal distribution with mean vector and covariance matrix at ​ The step of obtaining the boundary parameters of the equipment to be monitored, inputting the obtained boundary parameters and preset early warning parameters into the abnormality prediction model to obtain the abnormal probability value of the early warning parameters under the given boundary parameters comprises: The boundary parameters of the equipment to be monitored are obtained, and a boundary parameter vector is generated; From the early warning parameters, the parameters to be monitored are selected as the first early warning parameters, and the remaining early warning parameters are taken as the second early warning parameters, the second early warning parameters of the equipment to be monitored are obtained, and a warning parameter vector is generated according to the obtained second early warning parameters and preset first early warning parameters; The boundary parameter vector and the early warning parameter vector are input into the abnormality prediction model for prediction to obtain the abnormal probability value of the first early warning parameters under the given boundary parameters. 2.The smart inventory-based anomaly prediction method of claim 1, wherein, After the step of if the abnormal probability value is less than the global probability threshold value, the abnormality prediction model triggers an abnormality prompt, the method further comprises the following steps: According to the historical equipment parameters of the equipment to be monitored, a health score curve of the equipment to be monitored is generated; Input the abnormal probability value into the health score curve to obtain a health score value of the device to be monitored. 3.The smart inventory-based anomaly prediction method of claim 2, wherein, The step of generating the health score curve of the device to be monitored according to historical device parameters of the device to be monitored comprises: According to the monitoring time of the early warning parameter, the early warning parameter is divided into a first sub-early warning parameter and a second sub-early warning parameter; The historical device parameters corresponding to the first sub-early warning parameter are input into the abnormality prediction model to obtain a first abnormal probability value, and a first health score curve is generated according to the first abnormal probability value and a preset first slope; The historical device parameters corresponding to the second sub-early warning parameter are input into the abnormality prediction model to obtain a second abnormal probability value, and a second health score curve is generated according to the second abnormal probability value and a preset second slope. 4.The smart inventory-based anomaly prediction method of claim 1, wherein, The preset iteration condition is that the numerical values of the weight coefficient, the mean vector and the covariance matrix remain unchanged.

5. An anomaly prediction system based on intelligent monitoring, characterized by, The system is applied to the method of any one of claims 1 to 4, comprising: An abnormality prediction model is used to select corresponding device parameters according to a device to be monitored, and to establish an abnormality prediction model according to the device parameters, wherein the device parameters comprise boundary parameters and early warning parameters; An abnormal probability calculation module is used to obtain the boundary parameters of the device to be monitored, and to input the obtained boundary parameters and preset early warning parameters into the abnormality prediction model to obtain an abnormal probability value of the early warning parameters under the given boundary parameters; A threshold comparison module is used to compare the abnormal probability value with a global probability threshold value, and if the abnormal probability value is less than the global probability threshold value, the abnormality prediction model triggers an abnormality prompt; The abnormal probability value is calculated using the following formula: where y + represents the early warning parameter, x + represents the boundary parameter, y represents the historical data of the early warning parameter, and K represents a hyperparameter, represents the microelement measure in the early warning parameter vector space, y max represents the upper limit of the early warning parameter in the historical data, y min represents the lower limit of the early warning parameter in the historical data, represents the weight coefficient, represents the mean vector, represents the covariance matrix, represents the probability density of the normal distribution with the mean vector and the covariance matrix at .

6. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 4.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 4.

Images