Method and system for detecting abnormal operation state data of pumped storage unit

Abstract

The invention provides an operation state data anomaly detection method and system of a pumped storage unit, and is applied to the technical field of state monitoring data anomaly detection of the pumped storage unit. The method comprises the steps of performing interval division on obtained real-time operation state data of the pumped storage unit based on a Gaussian mixture model, and determining a working condition interval to which the real-time operation state data belongs; for the real-time operation state data of each working condition interval, inserting the real-time operation state data into a robustness random segmentation model corresponding to the working condition interval, and calculating an abnormal score of the real-time operation state data of the working condition interval; and determining an anomaly detection result of the working condition interval based on the anomaly score of the working condition interval and an anomaly threshold corresponding to the working condition interval. The problem of low unit state evaluation accuracy caused by abnormal data in the original state monitoring data of the pumped storage unit is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormality detection of state monitoring data of a pumped storage unit, and in particular to a method and system for detecting abnormality of operating state data of a pumped storage unit. Background Art

[0002] At present, my country is building a high-proportion new energy power system. The large-scale access of intermittent new energy has increased the peak load and frequency regulation pressure on the power supply side and reduced the stability level of the system. Pumped storage is an important means to reduce the negative impact of the randomness and volatility of intermittent new energy on the power system and promote the flexible consumption and efficient utilization of new energy.

[0003] In order to cope with the volatility of intermittent renewable energy, pumped storage units need to switch between multiple operating conditions such as power generation, pumping, phase adjustment, start and stop, and the units often need to operate at the boundary of the stable zone. The unit shaft system is subject to hydraulic shock, mechanical failure, and electromagnetic imbalance during frequent start and stop and operating condition switching, which may induce various faults and accidents such as abnormal vibration, structural fatigue, electrical failure, and damage to the operating mode of the unit, which are extremely harmful. Carrying out status maintenance on the unit based on the unit operation status monitoring data is of great significance to ensure the safety and stability of pumped storage units.

[0004] However, abnormal data generated by factors such as short-term sensor failure, external interference and transmission errors reduce the quality of the original status monitoring data, resulting in low accuracy of the unit's status assessment and affecting the unit's status maintenance work. Summary of the invention

[0005] In order to overcome the above-mentioned problem of low accuracy of unit status assessment due to the presence of abnormal data in the original status monitoring data of the pumped-storage unit, the present invention provides a method and system for detecting abnormality in operating status data of a pumped-storage unit.

[0006] In one aspect, the present invention provides a method for detecting abnormality in operating status data of a pumped storage unit, the method comprising:

[0007] Based on the Gaussian mixture model, the acquired real-time operation status data of the pumped storage unit is divided into intervals to determine the operating condition interval to which the real-time operation status data belongs;

[0008] For each of the real-time operating status data of the operating condition interval, inserting the real-time operating status data into the robust random segmentation model corresponding to the operating condition interval, and calculating the abnormality score of the real-time operating status data of the operating condition interval;

[0009] Based on the abnormality score of the operating condition interval and the abnormality threshold corresponding to the operating condition interval, an abnormality detection result of the operating condition interval is determined.

[0010] Optionally, before dividing the acquired real-time operating status data of the pumped-storage unit into intervals based on the Gaussian mixture model, the method further includes:

[0011] Extracting multi-dimensional features of the real-time operating status data under multiple operating condition dimensions; the multi-dimensional features include at least two of the lower frame vibration features, guide vane opening features, unit power features, excitation voltage features, and excitation current features;

[0012] The multi-dimensional features of the real-time operating status data at each operating moment are used as data points at the corresponding operating moment.

[0013] Optionally, dividing the acquired real-time operating status data of the pumped-storage unit into intervals based on the Gaussian mixture model to determine the operating condition interval to which the real-time operating status data belongs includes:

[0014] Calculate the probability that each of the data points belongs to each type of cluster based on the Gaussian mixture model;

[0015] Divide each data point into the cluster with the largest probability to obtain a multi-class clustering result;

[0016] A clustering result is taken as an operating condition interval to obtain the operating condition interval of each data point.

[0017] Optionally, the training process of the Gaussian mixture model includes:

[0018] Acquire historical operating status data of the pumped-storage unit, and extract multi-dimensional features of the historical operating status data to obtain training samples;

[0019] The model parameters of the Gaussian mixture model are learned based on the training samples and the maximum expectation algorithm to complete the training process of the Gaussian mixture model.

[0020] Optionally, the generation process of the robust random segmentation model includes:

[0021] Using the Gaussian mixture model to divide the training samples into operating condition intervals, to obtain training samples corresponding to each operating condition interval;

[0022] For each of the operating condition intervals, each training sample under the operating condition interval is taken as a node, and a corresponding robust random segmentation tree is constructed based on a random cutting algorithm of a binary search tree until the robust random segmentation tree reaches a preset height, thereby obtaining a corresponding robust random segmentation model, wherein the robust random segmentation model includes multiple robust random segmentation trees.

[0023] Optionally, after the robust random segmentation model is generated, the method further includes:

[0024] For each training sample in the operating condition interval, the abnormality score of each training sample is calculated based on the robust random segmentation model corresponding to the operating condition interval to obtain an abnormality score sequence of the operating condition interval;

[0025] Based on the abnormality score sequence of the operating condition interval, generating an abnormality score curve;

[0026] An abnormal threshold corresponding to the operating condition interval is determined based on an inflection point in the abnormal scoring curve.

[0027] Optionally, inserting the real-time operating status data into a robust random segmentation model corresponding to the operating condition interval to calculate an abnormality score of the real-time operating status data includes:

[0028] Inserting each data point of the real-time operating status data into each robust random segmentation tree of the robust random segmentation model of the corresponding operating condition interval;

[0029] Based on each of the data points traversing each of the robust random segmentation trees of the robust random segmentation model, calculating the depth of each of the data points in each of the robust random segmentation trees of the robust random segmentation model;

[0030] An anomaly score for each of the data points is calculated based on the depth of each of the data points in each robust random segmentation tree of the robust random segmentation model.

[0031] Optionally, determining the abnormality detection result based on the abnormality score of the operating condition interval and the abnormality threshold corresponding to the operating condition interval includes:

[0032] For each data point in each operating condition interval, if the abnormality score is greater than the abnormality threshold of the corresponding operating condition interval, the data point is determined to be an abnormal data point;

[0033] The method further comprises:

[0034] The abnormal data points are removed or marked.

[0035] On the other hand, the present invention also provides a pumped storage unit operation status data abnormality detection system, comprising:

[0036] An interval division module, used for performing interval division on the acquired real-time operation status data of the pumped storage unit based on a Gaussian mixture model, and determining the operating condition interval to which the real-time operation status data belongs;

[0037] The anomaly detection module is used to insert the real-time operating status data of each operating condition interval into a robust random segmentation model corresponding to the operating condition interval, calculate the anomaly score of the real-time operating status data of the operating condition interval; and determine the anomaly detection result of the operating condition interval based on the anomaly score of the operating condition interval and the anomaly threshold corresponding to the operating condition interval.

[0038] Optionally, the system further comprises a feature extraction module, wherein the feature extraction module is used to:

[0039] Extracting multi-dimensional features of the real-time operating status data under multiple operating condition dimensions; the multi-dimensional features include at least two of the lower frame vibration features, guide vane opening features, unit power features, excitation voltage features, and excitation current features;

[0040] The multi-dimensional features of the real-time operating status data at each operating moment are used as data points at the corresponding operating moment.

[0041] Optionally, the interval division module includes:

[0042] A probability calculation submodule, used for calculating the probability that each of the data points belongs to each type of cluster based on the Gaussian mixture model;

[0043] A clustering submodule is used to divide each data point into the cluster with the highest probability to obtain a multi-class clustering result;

[0044] The operating condition interval determination submodule is used to take a clustering result as a working condition interval to obtain the working condition interval of each data point.

[0045] Optionally, the system further includes a Gaussian model training module, and the Gaussian model training module includes:

[0046] A sample acquisition submodule is used to acquire the historical operation status data of the pumped storage unit and extract the multi-dimensional features of the historical operation status data to obtain training samples;

[0047] The training submodule is used to learn the model parameters of the Gaussian mixture model based on the training samples and the maximum expectation algorithm to complete the training process of the Gaussian mixture model.

[0048] Optionally, the system further includes a segmentation model generation module, and the segmentation model generation module includes:

[0049] An interval division submodule, used to divide the training samples into operating condition intervals using the Gaussian mixture model to obtain training samples corresponding to each operating condition interval;

[0050] The segmentation tree construction submodule is used to take each training sample in the operating condition interval as a node, and construct a corresponding robust random segmentation tree based on the random cutting algorithm of the binary search tree until the robust random segmentation tree reaches a preset height, thereby obtaining a corresponding robust random segmentation model, wherein the robust random segmentation model includes multiple robust random segmentation trees.

[0051] Optionally, the system further includes an abnormality threshold determination module, and the abnormality threshold determination module includes:

[0052] A sample scoring submodule, for calculating the abnormality score of each training sample under each operating condition interval based on the robust random segmentation model corresponding to the operating condition interval, to obtain an abnormality score sequence of the operating condition interval;

[0053] A curve generation submodule, used for generating an abnormality score curve based on the abnormality score sequence of the operating condition interval;

[0054] A threshold determination submodule is used to determine the abnormal threshold corresponding to the operating condition interval based on the inflection point in the abnormal scoring curve.

[0055] Optionally, the anomaly detection module includes:

[0056] A data point insertion submodule, used for inserting each data point of the real-time operating status data into each robust random segmentation tree of the robust random segmentation model of the corresponding operating condition interval;

[0057] A depth calculation submodule, used for traversing each robust random segmentation tree of the robust random segmentation model based on each data point, and calculating the depth of each data point in each robust random segmentation tree of the robust random segmentation model;

[0058] The anomaly score calculation submodule is used to calculate the anomaly score of each data point based on the depth of each data point in each robust random segmentation tree of the robust random segmentation model.

[0059] Optionally, the anomaly detection module further includes:

[0060] an abnormality identification submodule, for determining, for each data point in each operating condition interval, that the data point is an abnormal data point if the abnormality score is greater than an abnormality threshold of the corresponding operating condition interval;

[0061] The system further comprises:

[0062] The exception processing module is used to remove or mark the abnormal data points.

[0063] On the other hand, the present invention also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0064] The memory is used to store one or more programs;

[0065] When the one or more programs are executed by the at least one processor, any one of the above-mentioned methods for detecting abnormality in operating status data of a pumped-storage unit is implemented.

[0066] On the other hand, the present invention further provides a readable storage medium having an execution program stored thereon, and when the execution program is executed, the method for detecting abnormality in operating status data of a pumped-storage unit described in any one of the above is implemented.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] The present invention provides a method and system for detecting anomalies in operating status data of a pumped-storage unit. The method comprises the following steps: firstly, dividing the real-time operating status data of the pumped-storage unit into operating condition intervals based on a Gaussian mixture model, and determining the operating condition interval to which the real-time operating status data belongs; inserting the real-time operating status data into a robust random segmentation model corresponding to the operating condition interval, and calculating an anomaly score of the real-time operating status data; determining an anomaly detection result of the operating condition interval based on the anomaly score of the operating condition interval and an anomaly threshold corresponding to the operating condition interval; dividing data with the same probability distribution into the same operating condition interval by using a Gaussian mixture model, and constructing corresponding robust random segmentation models for different operating condition intervals to perform anomaly detection, so that each robust random segmentation model has stronger feature pertinence, thereby improving the accuracy and efficiency of anomaly detection, reducing the influence of abnormal data on state assessment and maintenance work of the pumped-storage unit, and improving the accuracy of unit state assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is one of the flow charts of a method for detecting abnormality of operating status data of a pumped storage unit according to the present invention;

[0070] Figure 2 A three-dimensional graph of time T-sequence number-amplitude V of upper frame vibration data and lower frame vibration data of a pumped storage unit according to an example of the present invention;

[0071] Figure 3 A schematic diagram of the construction process of the robust random partitioning tree of the present invention;

[0072] Figure 4 Schematic diagram of the anomaly detection process of the robust random segmentation model of the present invention;

[0073] Figure 5The second flowchart of the method for detecting abnormality of operating status data of a pumped storage unit according to the present invention;

[0074] Figure 6 It is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION

[0075] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0076] Embodiment 1:

[0077] The present invention provides a method for detecting abnormal operation status data of a pumped storage unit, as shown in the schematic diagram Figure 1 As shown, including:

[0078] Step S110, dividing the acquired real-time operating status data of the pumped storage unit into intervals based on a Gaussian mixture model to determine the operating condition interval to which the real-time operating status data belongs;

[0079] Step S120, for each of the real-time operating status data of the operating condition interval, inserting the real-time operating status data into the robust random segmentation model corresponding to the operating condition interval, and calculating the abnormality score of the real-time operating status data of the operating condition interval;

[0080] Step S130 , determining an abnormality detection result of the operating condition interval based on the abnormality score of the operating condition interval and an abnormality threshold corresponding to the operating condition interval.

[0081] In this example implementation, the real-time operating status data may be the full operating condition data of the pumped storage unit for real-time monitoring. Since the operating state of the pumped storage unit is complex, its operating condition data includes transient data and steady-state data. In order to ensure the statistical performance of the data, after obtaining the real-time operating status data, the stable operating condition data in the real-time operating status data may be extracted. For example, the operating status data from one hour after each unit operation to ten minutes before shutdown is extracted as the stable operating condition data. The operating status data of the pumped storage unit may include multi-dimensional data such as lower frame vibration data, guide vane opening, unit active power, unit reactive power, excitation voltage, excitation current, etc. The abnormal threshold may be pre-set according to the characteristics and requirements of the data, or it may be determined based on historical operating status data. The Gaussian mixture model is used to cluster data with the same probability distribution, that is, to divide the real-time operating status data into intervals, for example, to divide the real-time operating status data stream into flow intervals to achieve interval detection. The robust random segmentation model is used to detect abnormal data. The corresponding robust random segmentation model is constructed for different operating conditions, and the segmentation model construction based on statistical performance is realized, which can improve the detection accuracy of the robust random segmentation model for different operating conditions. This example uses the Gaussian mixture model to divide the data with the same probability distribution into the same operating condition interval, and constructs the corresponding robust random segmentation model for different operating condition intervals for anomaly detection, so that each robust random segmentation model has stronger feature targeting, thereby improving the accuracy and efficiency of anomaly detection, reducing the impact of abnormal data on the status assessment and maintenance of pumped storage units, and improving the accuracy of unit status assessment.

[0082] In an example implementation, before dividing the acquired real-time operating status data of the pumped storage unit into intervals based on the Gaussian mixture model in step S110, the method further includes:

[0083] Extracting multi-dimensional features of the real-time operating status data under multiple operating condition dimensions; the multi-dimensional features include at least two of the lower frame vibration features, guide vane opening features, unit power features, excitation voltage features, and excitation current features;

[0084] The multi-dimensional features of the real-time operating status data at each operating moment are used as data points at the corresponding operating moment.

[0085] In this example implementation, the real-time operation status data includes operation data of multiple operating dimensions of the pumped storage unit, for example, upper frame vibration data, lower frame vibration data, guide vane opening data, unit active power data, unit reactive power data, excitation voltage data, and excitation current data, etc. The dimension selection can be performed according to needs or experience, for example, the dimensions of lower frame vibration, guide vane opening, and unit active power are selected as feature dimensions. Figure 2 As shown, it is the three-dimensional data of the upper frame vibration data and the lower frame vibration data of a pumped storage unit. The three-dimensional coordinate axes are time T, frame numbers 1, 2, 3, 4, and vibration amplitude V. It can be seen from the figure that the vibration data characteristics of the upper frame are similar to those of the lower frame, and the lower frame is the main load-bearing component. In order to reduce the amount of data calculation, only the vibration data of the lower frame can be selected as the feature data. After determining the data dimension, the data of the corresponding dimension is extracted to form a multi-dimensional feature. The real-time operating status data of the pumped storage unit is a time series data stream, and the degree feature of each operating moment is used as the data point of the operating moment, that is, each data point includes a multi-dimensional feature. Through anomaly detection based on multi-dimensional features, multiple dimensions can be combined for anomaly identification to improve the accuracy of anomaly detection.

[0086] In some example implementations, the step S110 of dividing the acquired real-time operating status data of the pumped storage unit into intervals based on the Gaussian mixture model to determine the operating condition interval to which the real-time operating status data belongs includes the following process:

[0087] Calculate the probability that each of the data points belongs to each type of cluster based on the Gaussian mixture model;

[0088] Divide each data point into the cluster with the largest probability to obtain a multi-class clustering result;

[0089] A clustering result is taken as an operating condition interval to obtain the operating condition interval of each data point.

[0090] In this example implementation, the Gaussian mixture model is a model trained based on historical operating status data. The probability of each data point belonging to each type of cluster is first calculated. Each type of cluster corresponds to a single Gaussian distribution. Each data point is divided into the cluster with the largest probability to obtain multiple clustering results, and then multiple operating condition intervals are obtained. Exemplarily, the probability of a data point belonging to each single Gaussian distribution is calculated by the following formula:

[0091]

[0092] In the formula, γ(i,k) represents the i-th data point x i The probability of belonging to the kth Gaussian distribution, π k represents the mixing coefficient of the kth Gaussian distribution, p k (x i |θ k ) represents the i-th data point x i The probability of belonging to the kth Gaussian distribution, θ k =(μ k ,σ 2 k) is the parameter of the kth Gaussian distribution, μ k is the mean of the kth Gaussian distribution, σ 2 k is the variance of the kth Gaussian distribution, π j represents the mixing coefficient of the jth Gaussian distribution, p j (x i |θ j ) represents the i-th data point x i The probability of belonging to the jth Gaussian distribution, θ j is the parameter of the jth Gaussian distribution, and j ranges from 1 to k.

[0093] Exemplarily, the training process of the Gaussian mixture model includes:

[0094] Acquire historical operating status data of the pumped-storage unit, and extract multi-dimensional features of the historical operating status data to obtain training samples;

[0095] The model parameters of the Gaussian mixture model are learned based on the training samples and the maximum expectation algorithm to complete the training process of the Gaussian mixture model.

[0096] In this example implementation, the historical operation data may be operation data corresponding to a preset historical time period, such as operation status data for the past month. Similarly, multi-dimensional features of the historical operation status data may be extracted, and the multi-dimensional features of each historical operation moment may be used as a training sample. The process of learning the model parameters of the Gaussian mixture model based on the training sample and the maximum expectation algorithm is as follows:

[0097] The first step is the initialization phase: select the number of Gaussian distributions in the Gaussian mixture model (GMM) and initialize the parameters of each Gaussian distribution, including the mean, covariance, and mixing coefficient.

[0098] In the second step, for each training sample, the posterior probability that it belongs to each Gaussian distribution is calculated based on the current parameters.

[0099] The third step is to update the parameters: recalculate the mean, covariance and mixing coefficient of each Gaussian distribution based on the posterior probability of each Gaussian distribution.

[0100] Step 4: Convergence check: Determine whether the algorithm has converged, usually based on the log-likelihood of the data or the change in model parameters. Repeat steps 2 to 4 until the model converges.

[0101] In the above steps, the probability density function of the Gaussian mixture model is expressed as follows:

[0102]

[0103] Where: P(xs |θ) represents the probability density function of the Gaussian mixture model, x s is the sth training sample, θ is the parameter of the Gaussian mixture model, π k represents the mixing coefficient of the kth Gaussian distribution, p k (x s |θ k ) represents the probability density function of the kth Gaussian distribution, and K represents the number of single Gaussian distributions in the Gaussian mixture model.

[0104] The probability density function of a single Gaussian distribution is defined as:

[0105]

[0106] The expectation-maximization algorithm (EM) is used to estimate the parameter π of the Gaussian mixture model. k and θ k :

[0107] First determine the probability that the sample belongs to each single Gaussian distribution, and then use the maximum likelihood method to calculate the parameter π of a single Gaussian distribution k and θ k , the maximum likelihood function is as follows:

[0108]

[0109] Calculate π k and θ k :

[0110]

[0111] in, represents the maximum likelihood function of the parameters, γ(s,k) is the probability that the sth sample belongs to the kth Gaussian distribution, s represents the training sample number, N is the number of training samples, Represents the total probability of samples from the kth Gaussian distribution.

[0112] The above steps are repeated and iterated until the likelihood function value converges, completing the training process of the Gaussian mixture model.

[0113] Furthermore, a robust random segmentation model is generated based on the trained Gaussian mixture model, specifically including:

[0114] Using the Gaussian mixture model to divide the training samples into operating condition intervals, to obtain training samples corresponding to each operating condition interval;

[0115] For each of the operating condition intervals, each training sample under the operating condition interval is taken as a node, and a corresponding robust random segmentation tree is constructed based on a random cutting algorithm of a binary search tree until the robust random segmentation tree reaches a preset height, thereby obtaining a corresponding robust random segmentation model, wherein the robust random segmentation model includes multiple robust random segmentation trees.

[0116] In this example implementation, the robust random segmentation model of each operating condition interval includes multiple robust random segmentation trees. For example, the robust random segmentation model of each operating condition can be set to 100 robust random segmentation trees, and the size of each tree is 256. The robust random segmentation model is based on a group of robust random cut trees (RRCT), which constructs an effective binary tree to segment the time series samples and converts the description of abnormal phenomena into the path length of the quantified leaf node in the tree. The training sample set can be divided into training sample subsets of different operating condition intervals. For each operating condition interval, there can be a data set with n data in the interval, such as Figure 3 As shown in the figure, when constructing an RRCT, randomly sample from n data as the training sample set S of the tree. In the sample, randomly select a feature, and randomly select a value in the range of all values ​​of this feature (between the minimum and maximum values) as the split value C, and perform binary partitioning on the sample, dividing the sample with less than the value to the left of the node, and dividing the sample with greater than or equal to the value to the right of the node. In this way, a split condition and left and right data sets are obtained, and then the above process is repeated on the left and right data sets respectively, and the termination condition is directly reached (the data itself cannot be divided or the tree reaches a preset height). All nodes in the robust random partition tree RRCT correspond to the training sample set of historical operating status data, where non-leaf nodes correspond to a training sample. A robust random partition tree is created from the sample set S, and a new leaf node L is generated for sample x. If the partition separates sample x from the rest of the point set, the point is deleted from the point set. The above process is recursively performed on each training sample set to obtain a robust random partition model for each operating interval. The robust random segmentation model can better handle high-dimensional data and streaming data and has better detection effect on time series data.

[0117] In some example embodiments, after the robust random segmentation model is generated, the method further comprises:

[0118] For each training sample in the operating condition interval, the abnormality score of each training sample is calculated based on the robust random segmentation model corresponding to the operating condition interval to obtain an abnormality score sequence of the operating condition interval;

[0119] Based on the abnormality score sequence of the operating condition interval, generating an abnormality score curve;

[0120] An abnormal threshold corresponding to the operating condition interval is determined based on an inflection point in the abnormal scoring curve.

[0121] In this example implementation, the abnormality score of the sample can be calculated based on the training sample formed by the historical operation status data and the generated robust random segmentation model to obtain the corresponding abnormality score sequence, and then generate an abnormality score curve with the horizontal axis being the sample / time and the vertical axis being the abnormality score, and the last inflection point of the curve or the maximum value of the inflection point can be used as the abnormality threshold. This example uses the statistical characteristics of the time series sample to determine the abnormality threshold, which can improve the accuracy of anomaly detection based on the abnormality threshold.

[0122] In some example implementations, inserting the real-time operating status data into the robust random segmentation model corresponding to the operating condition interval in step S120 to calculate the abnormality score of the real-time operating status data includes the following process:

[0123] Inserting each data point of the real-time operating status data into each robust random segmentation tree of the robust random segmentation model of the corresponding operating condition interval;

[0124] Based on each of the data points traversing each of the robust random segmentation trees of the robust random segmentation model, calculating the depth of each of the data points in each of the robust random segmentation trees of the robust random segmentation model;

[0125] An anomaly score for each of the data points is calculated based on the depth of each of the data points in each robust random segmentation tree of the robust random segmentation model.

[0126] In this example implementation, for each data point, the RRCF algorithm calculates an anomaly score, which is determined based on a combination of the path length or other statistics of the data point in the corresponding segmentation model. The path length reflects the typicality or similarity of the data point with other data points, and the higher the anomaly score, the more likely it is an anomaly. Given a tree T(S) of a point set S and a data point p, when the joint distribution including the data point p is significantly different from the distribution excluding the data point p, the data point p is marked as an anomaly. Specifically, the anomaly score is calculated by the following formula:

[0127]

[0128] In the formula, |M(T)| is the data complexity of tree T, f(y,S,T) is the depth of point set S in tree T, and y represents the data point in point set S.

[0129] Anomaly score calculation formula:

[0130]

[0131] In the formula, s DC (x,S) represents the anomaly score for the data point x in the point set S, reflecting the typicality or anomaly length of the data point x in the entire random forest. f(y,S\{x},T) is the depth of the tree T after removing the data point x from the point set S.

[0132] In some example implementations, determining the abnormality detection result based on the abnormality score of the operating condition interval and the abnormality threshold corresponding to the operating condition interval in step S130 includes the following process:

[0133] For each data point in each operating condition interval, if the abnormality score is greater than the abnormality threshold of the corresponding operating condition interval, the data point is determined to be an abnormal data point;

[0134] In this example implementation, for each operating interval data point, the data point with an abnormality score greater than the abnormality threshold can be determined as an abnormal data point. Figure 4 As shown, for a data point x, M robust random split trees can be used Anomaly scoring is performed and the voting results of M trees are combined to obtain the final detection result.

[0135] Exemplarily, for detected abnormal data points, the abnormal data points are removed or marked to prepare for subsequent status assessment and maintenance work of the unit, thereby reducing the impact of abnormal data on the status assessment and maintenance work of the unit.

[0136] The following is a specific example to illustrate the specific steps of a method for detecting abnormal operation status data of a pumped storage unit of the present invention. Figure 5 As shown, the following processes may be included:

[0137] (1) Obtain the real-time monitoring data of the pumped storage unit in the historical period, that is, the historical operating status data, from which the vibration of the lower frame, the guide vane opening, and the active power are selected to form a three-dimensional data set. The data from one hour after each unit operation to ten minutes before shutdown is regarded as the stable operating state, and the steady-state value is extracted from the original three-dimensional data.

[0138] (2) Based on the Gaussian mixture model, the extracted steady-state data is divided into different operating ranges. The three-dimensional data is divided into multiple operating ranges, and the data in each range conforms to the same probability distribution.

[0139] (3) Use the robust random segmentation forest algorithm to perform anomaly scoring on the sample points in each interval. Specifically, calculate and sort the anomaly scores of the samples in each area, set the model to have 100 robust random segmentation trees in each operating interval, and the size of each tree is 256, and calculate the anomaly scores of all samples.

[0140] (4) Generate an abnormality score curve based on the abnormality scores of samples in each operating range, and set the inflection point of the curve as the abnormal threshold.

[0141] (5) The frame vibration, guide vane opening, and active power data (i.e., real-time operating status data) collected in real time by the unit are brought into the Gaussian mixture model to calculate the probability of the data in the overall distribution, thereby dividing the real-time operating status data into each operating condition interval. After that, the data is inserted into the robust random partitioning tree under the operating condition to calculate the data anomaly score.

[0142] (6) Compare the data anomaly score with the anomaly threshold of the corresponding operating condition, and output the anomaly detection result in real time. Specifically, compare the real-time sample anomaly score with the anomaly threshold. If the sample anomaly score is less than the anomaly score threshold of the operating condition area, the sample is a normal sample. If the sample anomaly value is greater than the anomaly score threshold, the sample is an abnormal sample.

[0143] The present invention continuously obtains the full operating condition data of the real-time monitoring of the pumped storage unit, and extracts the operating condition data of the stable operation of the unit. On this basis, the extracted steady-state data is divided into different operating condition intervals based on the Gaussian mixture model. Afterwards, all sample points in each operating condition interval are traversed respectively by the robust random segmentation forest algorithm to obtain the abnormal score of each sample point. Finally, the abnormal threshold is determined according to the inflection point of the abnormal score curve, and the samples with abnormal scores greater than the abnormal threshold are regarded as abnormal values ​​to obtain the abnormal value detection results. In this way, abnormal samples in the unit operation monitoring data can be detected in real time, the full operating condition data cleaning work of the pumped storage unit can be realized, and the quality of the monitoring data can be ensured; the detection of abnormal values ​​of the status monitoring data plays an important role in promoting the status maintenance work of the unit.

[0144] Embodiment 2:

[0145] Based on the same inventive concept, the present invention also provides a pumped storage unit operation status data abnormality detection system, the system comprising:

[0146] An interval division module, used for performing interval division on the acquired real-time operation status data of the pumped storage unit based on a Gaussian mixture model, and determining the operating condition interval to which the real-time operation status data belongs;

[0147] The anomaly detection module is used to insert the real-time operating status data of each operating condition interval into a robust random segmentation model corresponding to the operating condition interval, calculate the anomaly score of the real-time operating status data of the operating condition interval; and determine the anomaly detection result of the operating condition interval based on the anomaly score of the operating condition interval and the anomaly threshold corresponding to the operating condition interval.

[0148] In a possible implementation, the system further includes a feature extraction module, wherein the feature extraction module is used to:

[0149] Extracting multi-dimensional features of the real-time operating status data under multiple operating condition dimensions; the multi-dimensional features include at least two of the lower frame vibration features, guide vane opening features, unit power features, excitation voltage features, and excitation current features;

[0150] The multi-dimensional features of the real-time operating status data at each operating moment are used as data points at the corresponding operating moment.

[0151] In a possible implementation, the interval division module includes:

[0152] A probability calculation submodule, used for calculating the probability that each of the data points belongs to each type of cluster based on the Gaussian mixture model;

[0153] A clustering submodule is used to divide each data point into the cluster with the highest probability to obtain a multi-class clustering result;

[0154] The operating condition interval determination submodule is used to take a clustering result as a working condition interval to obtain the working condition interval of each data point.

[0155] In a possible implementation, the system further includes a Gaussian model training module, and the Gaussian model training module includes:

[0156] A sample acquisition submodule is used to acquire the historical operation status data of the pumped storage unit and extract the multi-dimensional features of the historical operation status data to obtain training samples;

[0157] The training submodule is used to learn the model parameters of the Gaussian mixture model based on the training samples and the maximum expectation algorithm to complete the training process of the Gaussian mixture model.

[0158] In a possible implementation, the system further includes a segmentation model generation module, and the segmentation model generation module includes:

[0159] An interval division submodule, used to divide the training samples into operating condition intervals using the Gaussian mixture model to obtain training samples corresponding to each operating condition interval;

[0160] The segmentation tree construction submodule is used to take each training sample in the operating condition interval as a node, and construct a corresponding robust random segmentation tree based on the random cutting algorithm of the binary search tree until the robust random segmentation tree reaches a preset height, thereby obtaining a corresponding robust random segmentation model, wherein the robust random segmentation model includes multiple robust random segmentation trees.

[0161] In a possible implementation, the system further includes an abnormality threshold determination module, and the abnormality threshold determination module includes:

[0162] A sample scoring submodule, for calculating the abnormality score of each training sample under each operating condition interval based on the robust random segmentation model corresponding to the operating condition interval, to obtain an abnormality score sequence of the operating condition interval;

[0163] A curve generation submodule, used for generating an abnormality score curve based on the abnormality score sequence of the operating condition interval;

[0164] A threshold determination submodule is used to determine the abnormal threshold corresponding to the operating condition interval based on the inflection point in the abnormal scoring curve.

[0165] In a possible implementation, the anomaly detection module includes:

[0166] A data point insertion submodule, used for inserting each data point of the real-time operating status data into each robust random segmentation tree of the robust random segmentation model of the corresponding operating condition interval;

[0167] A depth calculation submodule, used for traversing each robust random segmentation tree of the robust random segmentation model based on each data point, and calculating the depth of each data point in each robust random segmentation tree of the robust random segmentation model;

[0168] The anomaly score calculation submodule is used to calculate the anomaly score of each data point based on the depth of each data point in each robust random segmentation tree of the robust random segmentation model.

[0169] In a possible implementation, the anomaly detection module further includes:

[0170] an abnormality identification submodule, for determining, for each data point in each operating condition interval, that the data point is an abnormal data point if the abnormality score is greater than an abnormality threshold of the corresponding operating condition interval;

[0171] The system further comprises:

[0172] The exception processing module is used to remove or mark the abnormal data points.

[0173] Embodiment 3:

[0174] like Figure 6As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0175] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding functions, so as to implement the steps of a method for detecting abnormal operation status data of a pumped-storage unit in the above-mentioned embodiment.

[0176] Example 4

[0177] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It can be understood that the storage medium here can include both built-in storage media in electronic devices and, of course, extended storage media supported by electronic devices. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of a method for detecting abnormal operation status data of a pumped storage unit in the above embodiment.

[0178] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0179] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0180] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0181] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0182] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A method for detecting abnormality in operating status data of a pumped storage unit, characterized in that: include: Based on the Gaussian mixture model, the acquired real-time operation status data of the pumped storage unit is divided into intervals to determine the operating condition interval to which the real-time operation status data belongs; For each of the real-time operating status data of the operating condition interval, inserting the real-time operating status data into the robust random segmentation model corresponding to the operating condition interval, and calculating the abnormality score of the real-time operating status data of the operating condition interval; Based on the abnormality score of the operating condition interval and the abnormality threshold corresponding to the operating condition interval, an abnormality detection result of the operating condition interval is determined.

2. The method according to claim 1, characterized in that Before dividing the acquired real-time operating status data of the pumped-storage unit into intervals based on the Gaussian mixture model, the method further includes: Extracting multi-dimensional features of the real-time operating status data under multiple operating condition dimensions; the multi-dimensional features include at least two of the lower frame vibration features, guide vane opening features, unit power features, excitation voltage features, and excitation current features; The multi-dimensional features of the real-time operating status data at each operating moment are used as data points at the corresponding operating moment.

3. The method according to claim 2, characterized in that The step of dividing the acquired real-time operation status data of the pumped-storage unit into intervals based on the Gaussian mixture model to determine the operating condition interval to which the real-time operation status data belongs includes: Calculate the probability that each of the data points belongs to each type of cluster based on the Gaussian mixture model; Divide each data point into the cluster with the largest probability to obtain a multi-class clustering result; A clustering result is taken as an operating condition interval to obtain the operating condition interval of each data point.

4. The method according to claim 2, characterized in that: The training process of the Gaussian mixture model includes: Acquire historical operating status data of the pumped-storage unit, and extract multi-dimensional features of the historical operating status data to obtain training samples; The model parameters of the Gaussian mixture model are learned based on the training samples and the maximum expectation algorithm to complete the training process of the Gaussian mixture model.

5. The method according to claim 4, characterized in that The generation process of the robust random segmentation model includes: Using the Gaussian mixture model to divide the training samples into operating condition intervals, to obtain training samples corresponding to each operating condition interval; For each of the operating condition intervals, each training sample under the operating condition interval is taken as a node, and a corresponding robust random segmentation tree is constructed based on a random cutting algorithm of a binary search tree until the robust random segmentation tree reaches a preset height, thereby obtaining a corresponding robust random segmentation model, wherein the robust random segmentation model includes multiple robust random segmentation trees.

6. The method according to claim 5, characterized in that After the robust random segmentation model is generated, the method further includes: For each training sample in the operating condition interval, the abnormality score of each training sample is calculated based on the robust random segmentation model corresponding to the operating condition interval to obtain an abnormality score sequence of the operating condition interval; Based on the abnormality score sequence of the operating condition interval, generating an abnormality score curve; An abnormal threshold corresponding to the operating condition interval is determined based on an inflection point in the abnormal scoring curve.

7. The method according to claim 5, characterized in that The step of inserting the real-time operating status data into a robust random segmentation model corresponding to the operating condition interval and calculating an abnormality score of the real-time operating status data includes: Inserting each data point of the real-time operating status data into each robust random segmentation tree of the robust random segmentation model of the corresponding operating condition interval; Based on each of the data points traversing each of the robust random segmentation trees of the robust random segmentation model, calculating the depth of each of the data points in each of the robust random segmentation trees of the robust random segmentation model; An anomaly score for each of the data points is calculated based on the depth of each of the data points in each robust random segmentation tree of the robust random segmentation model.

8. The method according to claim 2, characterized in that: The determining of the abnormality detection result based on the abnormality score of the operating condition interval and the abnormality threshold corresponding to the operating condition interval includes: For each data point in each operating condition interval, if the abnormality score is greater than the abnormality threshold of the corresponding operating condition interval, the data point is determined to be an abnormal data point; The method further comprises: The abnormal data points are removed or marked.

9. A pumped storage unit operation status data anomaly detection system, characterized in that: The system comprises: An interval division module, used for performing interval division on the acquired real-time operation status data of the pumped storage unit based on a Gaussian mixture model, and determining the operating condition interval to which the real-time operation status data belongs; The anomaly detection module is used to insert the real-time operating status data of each operating condition interval into a robust random segmentation model corresponding to the operating condition interval, calculate the anomaly score of the real-time operating status data of the operating condition interval; and determine the anomaly detection result of the operating condition interval based on the anomaly score of the operating condition interval and the anomaly threshold corresponding to the operating condition interval.

10. The system according to claim 9, characterized in that The system further comprises a feature extraction module, wherein the feature extraction module is used to: Extracting multi-dimensional features of the real-time operating status data under multiple operating condition dimensions; the multi-dimensional features include at least two of the lower frame vibration features, guide vane opening features, unit power features, excitation voltage features, and excitation current features; The multi-dimensional features of the real-time operating status data at each operating moment are used as data points at the corresponding operating moment.

11. The system according to claim 10, characterized in that The interval division module includes: A probability calculation submodule, used for calculating the probability that each of the data points belongs to each type of cluster based on the Gaussian mixture model; A clustering submodule is used to divide each data point into the cluster with the highest probability to obtain a multi-class clustering result; The operating condition interval determination submodule is used to take a clustering result as a working condition interval to obtain the working condition interval of each data point.

12. The system according to claim 10, characterized in that The system also includes a Gaussian model training module, and the Gaussian model training module includes: A sample acquisition submodule is used to acquire the historical operation status data of the pumped storage unit and extract the multi-dimensional features of the historical operation status data to obtain training samples; The training submodule is used to learn the model parameters of the Gaussian mixture model based on the training samples and the maximum expectation algorithm to complete the training process of the Gaussian mixture model.

13. The system according to claim 12, characterized in that The system further includes a segmentation model generation module, wherein the segmentation model generation module includes: An interval division submodule, used to divide the training samples into operating condition intervals using the Gaussian mixture model to obtain training samples corresponding to each operating condition interval; The segmentation tree construction submodule is used to take each training sample in the operating condition interval as a node, and construct a corresponding robust random segmentation tree based on the random cutting algorithm of the binary search tree until the robust random segmentation tree reaches a preset height, thereby obtaining a corresponding robust random segmentation model, wherein the robust random segmentation model includes multiple robust random segmentation trees.

14. The system according to claim 13, characterized in that The system further includes an abnormal threshold determination module, wherein the abnormal threshold determination module includes: A sample scoring submodule, for calculating the abnormality score of each training sample under each operating condition interval based on the robust random segmentation model corresponding to the operating condition interval, to obtain an abnormality score sequence of the operating condition interval; A curve generation submodule, used for generating an abnormality score curve based on the abnormality score sequence of the operating condition interval; A threshold determination submodule is used to determine the abnormal threshold corresponding to the operating condition interval based on the inflection point in the abnormal scoring curve.

15. The system according to claim 13, characterized in that The anomaly detection module comprises: A data point insertion submodule, used for inserting each data point of the real-time operating status data into each robust random segmentation tree of the robust random segmentation model of the corresponding operating condition interval; A depth calculation submodule, used for traversing each robust random segmentation tree of the robust random segmentation model based on each data point, and calculating the depth of each data point in each robust random segmentation tree of the robust random segmentation model; The anomaly score calculation submodule is used to calculate the anomaly score of each data point based on the depth of each data point in each robust random segmentation tree of the robust random segmentation model.

16. The system according to claim 10, characterized in that The anomaly detection module also includes: an abnormality identification submodule, for determining, for each data point in each operating condition interval, that the data point is an abnormal data point if the abnormality score is greater than an abnormality threshold of the corresponding operating condition interval; The system further comprises: The exception processing module is used to remove or mark the abnormal data points.

17. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the method for detecting abnormality in operating status data of a pumped-storage unit according to any one of claims 1 to 8 is implemented.

18. A readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, the method for detecting abnormality in operating status data of a pumped storage unit according to any one of claims 1 to 8 is implemented.

Images