Hydroelectric generator set abnormal state detection method and device and computer equipment

Abstract

This invention relates to the field of hydropower technology and discloses a method, apparatus, and computer equipment for detecting abnormal states of hydro-turbine generator units. The method includes: acquiring current characteristic data of the hydro-turbine generator unit, including unit operating parameters and stability monitoring parameters; inputting the characteristic data into a pre-trained abnormal state detection model to obtain mapped features; calculating the distance from the mapped features to a single-class center, which is determined based on positive sample data collected when the hydro-turbine generator unit is in normal operation; and determining that the hydro-turbine generator unit is in a normal state if the distance is less than a preset threshold. This invention considers the impact of different operating conditions on the operating state of the hydro-turbine generator unit when detecting abnormal states, thus enabling accurate detection of abnormal states.

Description

Technical Field

[0001] This invention relates to the field of hydropower technology, specifically to a method, device, and computer equipment for detecting abnormal states of hydro-generator units. Background Technology

[0002] Hydropower turbine generator sets are the core and critical equipment of hydropower stations. Condition monitoring technology for hydropower turbine generator sets can monitor and analyze their operating status at any time, promptly identify potential hazards and defects, prevent destructive accidents, and guide the operation and maintenance of the units, thereby ensuring the safe and economical operation of the units.

[0003] The condition monitoring data of hydro-generator units is multivariate time-series data, characterized by high dimensionality and massive volume, reflecting the real-time condition information of the hydro-generator units. Current hydropower station turbine generator monitoring systems collect information such as equipment swing and peak-to-peak vibration values, using simple threshold and slope methods to issue early warnings of abnormal conditions. However, practice shows that the parameter characteristics of the unit are closely related to its operating conditions. Therefore, how to scientifically mine monitoring data and accurately detect abnormal equipment conditions is a key challenge in hydro-generator condition monitoring. Summary of the Invention

[0004] In view of this, the present invention provides a method, apparatus and computer equipment for detecting abnormal states of hydro-generator sets, in order to solve the problem of being unable to accurately detect abnormal states of equipment.

[0005] In a first aspect, the present invention provides a method for detecting abnormal states of a hydro-generator unit, comprising: acquiring current feature data of the hydro-generator unit, the feature data including unit operating parameters and stability monitoring parameters; inputting the feature data into a pre-trained abnormal state detection model to obtain the mapping features of the feature data; calculating the distance from the mapping features to a single-class center, the single-class center being determined based on positive sample data collected when the hydro-generator unit is in normal operation; and determining that the hydro-generator unit is in a normal state if the distance is less than a preset threshold.

[0006] The abnormal state detection method for hydro-generator units provided in this invention acquires the unit's operating parameters and stability monitoring parameters, analyzes these parameters using an abnormal state detection model, obtains mapping features of the feature data, and determines whether the hydro-generator unit is in a normal state based on the distance between the mapping features and the single-class center. Since the operating state of the hydro-generator unit varies under different operating conditions, the abnormal state detection model fully learns the relationship between different operating conditions and monitoring parameters. When detecting abnormal states of the hydro-generator unit, it can consider the impact of different operating conditions on the unit's operating state, thereby accurately detecting abnormal states, avoiding destructive accidents, and guiding unit operation and maintenance, thus ensuring the safe and economical operation of the unit.

[0007] In one optional implementation, the abnormal state detection model includes an encoder and a mapping network. The encoder is used to map the feature data to obtain initial multidimensional features and send the initial multidimensional features to the mapping network. The mapping network is used to perform secondary mapping on the initial multidimensional features to obtain mapped features.

[0008] In one optional implementation, the steps of training the abnormal state detection model include: acquiring training feature data of the hydro-generator unit at different time points; constructing multiple sets of sample data based on the training feature data of the hydro-generator unit at different time points; establishing positive sample data and negative sample data using the multiple sets of sample data; training the initial encoder using the positive sample data to obtain the encoder; and training the initial mapping network using the encoder, positive sample data, and negative sample data to obtain the mapping network.

[0009] In one optional implementation, multiple sets of sample data are constructed based on the training features of the hydro-generator unit at different time points, including: constructing a first two-dimensional feature matrix based on the training feature data at different time points; normalizing the first two-dimensional feature matrix to obtain a second two-dimensional feature matrix; and dividing the second two-dimensional feature matrix in the time dimension using overlapping sampling to obtain multiple sets of sample data.

[0010] In one optional implementation, the step of establishing negative sample data using multiple sets of sample data includes: keeping the unit operating parameters in each sample data unchanged, randomly replacing the original data of the stability parameters in the sample data with out-of-standard values ​​to obtain out-of-standard sample data; and normalizing the out-of-standard sample data to obtain negative sample data.

[0011] In one optional implementation, training an initial encoder using positive sample data to obtain an encoder includes: keeping the unit operating parameters in each sample data unchanged, randomly setting the stability parameters in the sample data to 0 according to a preset probability to obtain enhanced sample data; mapping the enhanced sample data to a first multi-dimensional feature using the initial encoder; reconstructing the first multi-dimensional feature into the original sample using a decoder network; inputting each sample data and its corresponding original sample into a first loss function to obtain a first loss value; if the sum of the first loss values ​​does not meet a first preset condition, optimizing the parameters of the initial encoder and decoder network, and returning to the step of mapping the enhanced sample data to the first multi-dimensional feature using the initial encoder, until the first loss value meets the first preset condition to obtain the encoder.

[0012] In one optional implementation, an initial mapping network is trained using an encoder, positive sample data, and negative sample data to obtain a mapping network. This includes: mapping positive sample data and negative sample data to a second multi-dimensional feature and a third multi-dimensional feature using the encoder, respectively; mapping the second and third multi-dimensional features to a fourth and a fifth multi-dimensional feature using the initial mapping network, respectively; calculating a second loss value for the initial mapping network using the distance between the fourth multi-dimensional feature and the single-class center; calculating a third loss value for the initial mapping network using the fifth multi-dimensional feature, the single-class center, and a preset radius of the hypersphere of the single-class center; if the sum of the second and third loss values ​​does not satisfy a second preset condition, optimizing the parameters of the initial mapping network, and returning to the step of mapping positive sample data and negative sample data to the second and third multi-dimensional features using the encoder, until the sum of the second and third loss values ​​satisfies the second preset condition, thus obtaining the mapping network.

[0013] In one optional implementation, the sample data includes multiple sets of data. The step of determining a preset threshold includes: inputting each set of data in the positive sample data into the abnormal state detection model to obtain the sixth multidimensional feature corresponding to each set of data; calculating the distance between each sixth multidimensional feature and the single class center; and determining the preset threshold based on the distance corresponding to each sixth multidimensional feature, so that the proportion of distances less than the preset threshold meets the preset ratio.

[0014] Secondly, the present invention provides an abnormal state detection device for a hydro-generator unit, comprising: a data acquisition module for acquiring current feature data of the hydro-generator unit, the feature data including unit operating parameters and stability monitoring parameters; a mapping feature acquisition module for inputting the feature data into a pre-trained abnormal state detection model to obtain the mapping features of the feature data; a distance calculation module for calculating the distance from the mapping features to a single-class center, the single-class center being determined based on positive sample data collected when the hydro-generator unit is in normal operating condition; and a state detection module, which determines that the hydro-generator unit is in a normal state if the distance is less than a preset threshold.

[0015] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the above-described method for detecting abnormal states of a hydro-generator set or any corresponding embodiment thereof. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the abnormal state detection method for a hydro-generator unit according to an embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram of the data processing flow of the abnormal state detection model for processing feature data in an embodiment of the present invention;

[0019] Figure 3 This is a schematic diagram of the process of training the abnormal state detection model in an embodiment of the present invention;

[0020] Figure 4 This is a schematic diagram illustrating how the second two-dimensional feature matrix is ​​divided using overlapping sampling in an embodiment of the present invention to obtain multiple sets of sample data;

[0021] Figure 5 This is a schematic diagram of the encoder training process in an embodiment of the present invention;

[0022] Figure 6 This is a schematic diagram of the process of training the mapping network in an embodiment of the present invention;

[0023] Figure 7This is a structural block diagram of a hydro-generator abnormal state detection device according to an embodiment of the present invention;

[0024] Figure 8 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0025] 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.

[0026] According to an embodiment of the present invention, an embodiment of a method for detecting abnormal states of a hydro-generator set is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0027] This embodiment provides a method for detecting abnormal states of a hydro-generator unit, which can be used in the aforementioned terminals, such as mobile phones and tablets. Figure 1 This is a flowchart of an abnormal state detection method for a hydro-generator unit according to an embodiment of the present invention, as follows: Figure 1 As shown, the process includes the following steps:

[0028] Step S101: Obtain the current characteristic data of the hydro-generator unit, including unit operating parameters and stability monitoring parameters.

[0029] In one optional embodiment, the unit operating parameters include one or more of the following: working head, servo travel, active power, reactive power, etc.

[0030] Operating head refers to the water level difference between the inlet and outlet of a hydro-generator unit, which is the head that the turbine can utilize. The greater the operating head, the greater the output power of the turbine.

[0031] The servo drive stroke refers to the distance the servo drive piston moves, which is the opening degree of the turbine generator's guide vanes. The greater the servo drive stroke, the greater the opening degree of the turbine guide vanes, the greater the flow rate through the turbine, and thus the greater the output power.

[0032] Active power refers to the actual power output of a hydro-generator unit, that is, the power on the turbine shaft. The magnitude of active power depends on factors such as the turbine's rotational speed, head, and flow rate.

[0033] Reactive power refers to the reactive power output of a hydroelectric turbine unit, specifically the reactive power on the turbine shaft. The magnitude of reactive power depends on factors such as the turbine's rotational speed, head, and flow rate.

[0034] Stability monitoring parameters are vibration and displacement indices of a hydro-generator unit during operation, used to monitor and evaluate the unit's operating status. Specifically, they include one or more of the following parameters during the operation of the hydro-generator unit: upper guide X-axis swing, upper guide Y-axis swing, lower guide X-axis swing, lower guide Y-axis swing, water guide X-axis swing, water guide Y-axis swing, peak-to-peak value of horizontal vibration in the upper frame X-axis, peak-to-peak value of horizontal vibration in the upper frame Y-axis, peak-to-peak value of vertical vibration in the upper frame Z-axis, peak-to-peak value of horizontal vibration in the lower frame X-axis, peak-to-peak value of horizontal vibration in the lower frame Y-axis, peak-to-peak value of vertical vibration in the lower frame Z-axis, peak-to-peak value of horizontal vibration in the top cover X-axis, peak-to-peak value of horizontal vibration in the top cover Y-axis, peak-to-peak value of vertical vibration in the top cover Z-axis, and peak-to-peak value of axial displacement.

[0035] Excessive sway, vibration, or axial displacement may indicate a malfunction or imbalance in the generator set, requiring timely inspection and maintenance. Real-time monitoring and analysis of these parameters allows for early detection of problems and the implementation of appropriate measures to prevent equipment damage and downtime. Specific monitoring and evaluation standards are typically provided by the turbine generator set manufacturer or relevant technical specifications.

[0036] The stability monitoring parameters of a hydro-generator unit vary depending on its operating conditions. For example, the operating head and servo travel affect the output power of the generator unit, thus influencing its vibration and displacement parameters. For instance, an increase in operating head leads to an increase in the generator unit's output power, resulting in increased vibration and displacement parameters. Active and reactive power also affect the operating status of the hydro-generator unit, thereby impacting its vibration and displacement parameters. For example, an increase in active power leads to an increase in the generator unit's output power, resulting in increased vibration and displacement parameters.

[0037] Therefore, in this embodiment of the invention, when acquiring the characteristic parameters of the hydro-generator unit, the unit's operating condition parameters and stability monitoring parameters are acquired. By combining the unit's operating condition parameters and stability monitoring parameters to analyze whether the hydro-generator unit is abnormal, the operating parameters of the hydro-generator unit can be fully considered, resulting in more accurate anomaly detection results and reducing misjudgments.

[0038] Step S102: Input the feature data into the pre-trained abnormal state detection model to obtain the mapped features of the feature data.

[0039] In one optional embodiment, the acquired feature data includes multiple parameters, which can be combined to form two-dimensional data. The two-dimensional data is input into an anomaly detection model, and the anomaly detection model maps the two-dimensional data onto a hypersphere with a single class center to obtain the mapped features.

[0040] Step S103: Calculate the distance from the mapped feature to the single-class center, which is determined based on the positive sample data collected when the hydro-generator unit is in normal operation. If the distance is less than a preset threshold, proceed to step S104; if the distance is greater than or equal to the preset threshold, proceed to step S105.

[0041] In one optional embodiment, the distance from the mapped feature to the single class center can be Euclidean distance, cosine distance, etc.

[0042] Step S104: Determine that the hydro-generator unit is in normal condition.

[0043] Step S105: Determine that the hydro-generator unit is in an abnormal state.

[0044] In one optional embodiment, the single-class center is obtained by clustering or other similarity measurement methods on the positive sample data collected when the hydro-generator unit is in normal operation. If the mapping feature of the hydro-generator unit is closer to the single-class center, it means that the hydro-generator unit is closer to normal operation. Therefore, the distance from the mapping feature to the single-class center can be combined to determine whether the hydro-generator unit is abnormal.

[0045] In an alternative embodiment, the preset threshold can be determined based on relevant experience.

[0046] The present invention provides a method for detecting abnormal states of hydro-generator units. This method acquires the unit's operating parameters and stability monitoring parameters, analyzes these parameters using an abnormal state detection model, and obtains mapping features of the characteristic data. Based on these mapping features and the distance to a single-class center, it determines whether the hydro-generator unit is in a normal state. Since the operating state of the hydro-generator unit varies under different operating conditions, the abnormal state detection model fully learns the relationship between different operating conditions and monitoring parameters. When detecting abnormal states of the hydro-generator unit, it considers the impact of different operating conditions on the unit's operating state, thus enabling accurate detection of abnormal states, preventing destructive accidents, and guiding unit operation and maintenance, thereby ensuring the safe and economical operation of the unit.

[0047] In an optional embodiment, in the abnormal state detection model used in step S102 above, such as Figure 2As shown, it includes an encoder and a mapping network. The encoder is used to map the feature data to obtain initial multidimensional features and sends the initial multidimensional features to the mapping network. The mapping network is used to perform secondary mapping on the initial multidimensional features to obtain mapped features.

[0048] In an alternative embodiment, such as Figure 3 As shown, the steps for training an abnormal state detection model include:

[0049] Step S301: Obtain training feature data of the hydro-generator unit at different time points.

[0050] In an optional embodiment, the training feature data obtained when training the abnormal state detection model is of the same type as the feature data obtained in step S101 above.

[0051] Step S302: Construct multiple sets of sample data based on the training feature data of the hydro-generator unit at different time points.

[0052] In one optional embodiment, after obtaining training feature data at different time points, the feature data is divided into multiple groups of sample data based on the time points, with each group of sample data corresponding to the same number of time points.

[0053] Step S303: Use multiple sets of sample data to establish positive sample data and negative sample data.

[0054] In one optional embodiment, if the hydro-generator unit operates according to the parameters in the positive sample data, it indicates that the hydro-generator unit is in a normal operating state; if the hydro-generator unit operates according to the parameters in the negative sample data, it indicates that the hydro-generator unit is in an abnormal operating state.

[0055] In an optional embodiment, positive and negative sample data can be determined in any way. For example, when the hydro-generator unit is in normal operation, step S301 can be executed to obtain training feature data of the hydro-generator unit at different time points. After forming multiple sets of sample data based on the training feature data, these multiple sets of sample data are used as positive sample data. The values ​​of one or more parameters in each set of sample data are replaced with outliers to obtain negative sample data.

[0056] Step S304: Train the initial encoder using positive sample data to obtain the encoder.

[0057] In one alternative embodiment, an autoencoder can be built and self-supervised training performed to obtain the encoder.

[0058] Step S305: Train the initial mapping network using the encoder, positive sample data, and negative sample data to obtain the mapping network.

[0059] In an optional embodiment, step S302 specifically includes the following steps:

[0060] Step a1: Construct the first two-dimensional feature matrix based on the training feature data at different time points.

[0061] For example, after reading feature data from multiple time points under normal conditions from the hydropower station computer monitoring system, a first two-dimensional feature matrix is ​​constructed with time points as the row dimension and features as the column dimension.

[0062] For example, if information is collected from L nodes, and 20 feature data are collected at each time node, then the number of row dimensions is L, the number of column dimensions is 20, and the constructed first two-dimensional feature matrix is: The i-th feature sequence is denoted as row vector a. i (1≤i≤20), the length of the vector is L, then the matrix

[0063] Step a2: Normalize the first two-dimensional feature matrix to obtain the second two-dimensional feature matrix.

[0064]

[0065] a ij For the data of the i-th feature at the j-th time node, a imax a represents the maximum value of the i-th type of feature. imin ε represents the minimum value of the i-th type of feature, where ε is a positive value. For example, ε is usually chosen to be no greater than 10. -5 The positive value.

[0066] Step a3: In the time dimension, the second two-dimensional feature matrix is ​​divided by overlapping sampling to obtain multiple sets of sample data.

[0067] In an optional embodiment, if the length of each sample is M, the sampling step length is S, and the overlap length is MS, then the number of samples obtained is:

[0068]

[0069] The function floor() represents rounding down, and the dataset of the sample data is denoted as . Among the samples

[0070] For example, such as Figure 4 As shown, for a two-dimensional data matrix A containing 5 time nodes, assuming the sample length is 3 and the sampling step length is 1, the overlap length is 2, and a total of 3 samples can be obtained.

[0071] In an optional embodiment, after obtaining multiple sets of sample data through steps a1-a3, the unit operating parameters in each sample data are kept unchanged. The original data of the stability parameters in the sample data are randomly replaced with out-of-range values ​​to obtain out-of-range sample data. The out-of-range sample data is then normalized to obtain negative sample data. When generating negative sample data, the fewer out-of-range values ​​of the stability parameters are replaced, the greater the training difficulty of the anomaly detection neural network, and the better its detection effect.

[0072] In an alternative embodiment, such as Figure 5 As shown, step S304 specifically includes the following steps:

[0073] Step b1, keep the sample data x i The unit operating parameters remain unchanged, and the sample data x is distributed according to a preset probability. i The stability parameter in the sample is randomly set to 0 to obtain the enhanced sample data x. i ′ .

[0074] In one optional embodiment, the larger the preset probability p, the more difficult the self-supervised training of the autoencoder becomes; while when p is too large, all the stability monitoring parameters of the sample data are close to 0, making it difficult for the model to capture the relationship between the parameters, resulting in reduced performance; therefore, the effect of model training first increases and then decreases as p increases.

[0075] Step b2, using the initial encoder E to augment the sample data x i ′ Mapped to the first multidimensional feature Z i For example, if the enhanced sample data x i ′ If the mapping is to d-dimensional features, then,

[0076] Step b3, use the decoder network D to process the first multidimensional feature Z i Reconstructed into the original sample For example,

[0077] Step b4, process the sample data x i and original sample The input is fed into the first loss function to obtain the first loss value.

[0078] In an optional embodiment, the first loss function is constructed based on the sum of the feature differences of each group of positive sample data, wherein the feature difference of a group of positive sample data refers to the feature difference of that group of positive sample data x i With the first multidimensional feature Z i difference.

[0079] Specifically, the first loss function is:

[0080]

[0081] Where N represents the number of positive sample data. Let x represent the original sample. i This represents positive sample data.

[0082] If the first loss value does not meet the first preset condition, the parameters of the initial encoder and decoder networks are optimized, and the process returns to step b2 until the first loss value meets the first preset condition, thus obtaining the encoder.

[0083] In an optional embodiment, the optimization function can be expressed as:

[0084]

[0085] If the above optimization function is used to optimize the initial encoder, the first preset condition is: the first loss value reaches the minimum value. When the first loss value is the minimum value, the current initial encoder is used as the encoder.

[0086] In an optional embodiment, the first preset condition may be that the first loss value is less than a preset value, that is, when the first loss value is less than the preset value, the current initial encoder is used as the encoder in the abnormal state detection model.

[0087] In the above embodiments, when training the encoder in the abnormal state detection model, the operating condition parameters in the sample data are kept unchanged, and the stability monitoring parameters in the sample data are randomly set to 0 to form enhanced sample data. The enhanced sample data is used to train the initial encoder. The monitoring data is randomly set to 0 with a certain probability. The encoder can automatically correct the data parameters set to 0 and learn the intrinsic relationship between the operating conditions and the monitoring parameters. Therefore, the abnormal state detection model can analyze the stability monitoring parameters in combination with the unit operating conditions to obtain the abnormal detection results. The results obtained in combination with the unit operating conditions are more accurate.

[0088] In an alternative embodiment, such as Figure 6 As shown, step S305 specifically includes the following steps:

[0089] Step c1: The encoder is used to map the positive sample data and negative sample data into the second multidimensional feature and the third multidimensional feature, respectively. Both the second multidimensional feature and the third multidimensional feature are d-dimensional features.

[0090] Step c2: Using the initial mapping network, the second and third multidimensional features are mapped to the fourth and fifth multidimensional features, respectively. Both the fourth and fifth multidimensional features are d. ′ Dimensional features.

[0091] Step c3: Calculate the second loss value of the initial mapping network using the fourth multidimensional feature and the distance between the single class centers.

[0092] In an optional embodiment, the mapping features of the positive sample data are located on a hypersphere with center c, and the smaller the distance from the center, the better. The loss function for calculating the second loss value is:

[0093]

[0094] Where N represents the number of positive sample data, G(E(x) i )) represents the fourth multidimensional feature of the i-th group of positive sample data, and c is the d ′ Learnable single-class centers in a 3D feature space. λ1 represents the adjustable balance coefficient, J represents the number of layers in the neural network G, and W j Let ||*|| be the parameters of the j-th layer of the mapping network G. F The Frobenius norm of * is represented. In an alternative embodiment, the neural network in the mapping network includes, but is not limited to, convolutional neural networks, fully connected networks, Transformer networks, recurrent neural networks, etc.

[0095] Step c4: Calculate the third loss value of the initial mapping network using the fifth multidimensional feature, the single-class center, and the preset radius of the hypersphere of the single-class center.

[0096] In an optional embodiment, the mapping network G is trained using negative sample data. The mapping features of the negative sample data should be far from the single-class center c, but need to be restricted to a hypersphere of radius r to reduce the risk of openness in the mapping feature space. The loss function for calculating the third loss value is:

[0097]

[0098] Where K represents the number of negative samples. Let c represent the fifth multidimensional feature of the i-th negative sample data, and c be the d ′ Learnable single-class centers in a 3D feature space. r represents the preset radius of the hypersphere with a single class center, and λ2 is an adjustable parameter. F Let Frobenius norm represent *.

[0099] If the sum of the second and third loss values ​​does not meet the second preset condition, the parameters of the initial mapping network are optimized, and the process returns to step c1 above until the sum of the second and third loss values ​​meets the second preset condition, thus obtaining the mapping network.

[0100] In an optional embodiment, the optimization function of the mapping network G is:

[0101]

[0102] If the above optimization function is used to optimize the initial encoder, the second preset condition is: the sum of the second loss value and the third loss value reaches the minimum value. When the sum of the second loss value and the third loss value is the minimum value, the current initial mapping network is used as the mapping network.

[0103] In one optional embodiment, the second preset condition may be that the sum of the second loss value and the third loss value is less than a preset value, that is, when the sum of the second loss value and the third loss value is less than the preset value, the current initial mapping network is used as the mapping network.

[0104] In the above embodiments, when training the mapping network in the abnormal state detection model, a method of training a single classifier using positive and negative samples is proposed to increase the mapping network. In addition to normal samples, artificially generated negative samples are also used to train the network, thereby increasing the distance between positive and negative samples in the mapping feature space and improving the model's abnormal detection capability.

[0105] In an optional embodiment, the preset threshold used to determine whether the hydro-generator unit is in a normal state is determined through the following steps:

[0106] Step d1: Input each group of data from the positive sample data into the abnormal state detection model to obtain the sixth multidimensional feature corresponding to each group of data.

[0107] Step d2, calculate the distance between each sixth multidimensional feature and the single-class center: d(G(E(x)) i ),c)=||G(E(x) i ))-c||2.

[0108] Step d3: Determine a preset threshold based on the distance corresponding to each sixth multidimensional feature, so that the proportion of distances less than the preset threshold meets the preset ratio.

[0109] This embodiment also provides a device for detecting abnormal states of a hydro-generator set. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also conceivable.

[0110] This embodiment provides a device for detecting abnormal states of a hydro-generator set, such as... Figure 7 As shown, it includes:

[0111] The data acquisition module 701 is used to acquire the current characteristic data of the hydro-generator unit, including unit operating parameters and stability monitoring parameters.

[0112] The mapping feature acquisition module 702 is used to input feature data into a pre-trained abnormal state detection model to obtain the mapping features of the feature data;

[0113] The distance calculation module 703 is used to calculate the distance from the mapped features to the single class center, which is determined based on the positive sample data collected when the hydro-generator unit is in normal operation.

[0114] If the distance is less than a preset threshold, the status detection module 703 is used to determine that the hydro-generator unit is in a normal state.

[0115] In some alternative implementations, the constant-state detection model includes an encoder and a mapping network.

[0116] The encoder is used to map the feature data to obtain initial multidimensional features, and then sends the initial multidimensional features to the mapping network; the mapping network is used to perform secondary mapping on the initial multidimensional features to obtain the mapped features.

[0117] In some optional implementations, the abnormal state detection model is trained through a model training module, which includes:

[0118] The feature data acquisition unit is used to acquire feature data of the hydro-generator unit at different time points;

[0119] The sample construction unit is used to construct multiple sets of sample data based on the characteristic data of the hydro-generator unit at different time points.

[0120] Positive and negative sample construction units are used to build positive and negative sample data using multiple sets of sample data.

[0121] The encoder training unit is used to train the initial encoder using positive sample data to obtain the encoder.

[0122] The mapping network training unit is used to train the initial mapping network using the encoder, positive sample data, and negative sample data to obtain the mapping network.

[0123] In some alternative implementations, the sample construction unit includes:

[0124] The first two-dimensional feature matrix construction sub-unit is used to construct the first two-dimensional feature matrix based on the feature data at different time points;

[0125] The normalization processing subunit is used to normalize the first two-dimensional feature matrix to obtain the second two-dimensional feature matrix;

[0126] The feature partitioning subunit is used to partition the second two-dimensional feature matrix in the time dimension using overlapping sampling to obtain multiple sets of sample data.

[0127] In some alternative implementations, the positive and negative sample construction unit includes:

[0128] The negative sample construction sub-unit is used to keep the unit operating parameters in each sample data unchanged, randomly replace the original data of the stability parameters in the sample data with out-of-standard values ​​to obtain out-of-standard sample data, and normalize the out-of-standard sample data to obtain negative sample data.

[0129] In some alternative implementations, the encoder training unit includes:

[0130] An enhanced sample construction sub-unit is used to keep the unit operating parameters in each sample data unchanged, and randomly set the stability parameters in the sample data to 0 according to a preset probability to obtain enhanced sample data.

[0131] The first data mapping subunit is used to map the augmented sample data into a first multidimensional feature using the initial encoder;

[0132] The feature reconstruction subunit is used to reconstruct the first multidimensional features into the original sample using the decoder network;

[0133] The first loss value calculation unit is used to input each sample data and its corresponding original sample into the first loss function to obtain the first loss value;

[0134] If the sum of the first loss values ​​does not meet the first preset condition, the first optimization subunit optimizes the parameters of the initial encoder and decoder networks and returns to the step of mapping the enhanced sample data to the first multidimensional feature using the initial encoder until the first loss value meets the first preset condition, thus obtaining the encoder.

[0135] In some alternative implementations, the mapping network training unit includes:

[0136] The second data mapping subunit is used to map positive sample data and negative sample data into second multi-dimensional features and third multi-dimensional features respectively using the encoder;

[0137] The third data mapping subunit is used to map the second and third multidimensional features into the fourth and fifth multidimensional features respectively using the initial mapping network;

[0138] The second loss value calculation subunit is used to calculate the second loss value of the initial mapping network using the fourth multidimensional feature and the distance between the single class center;

[0139] The third loss value calculation subunit is used to calculate the third loss value of the initial mapping network using the fifth multidimensional feature, the single class center, and the preset radius of the hypersphere of the single class center;

[0140] If the sum of the second loss value and the third loss value does not meet the second preset condition, the second optimization subunit optimizes the parameters of the initial mapping network and returns to the step of using the encoder to map the positive sample data and negative sample data into the second multi-dimensional feature and the third multi-dimensional feature respectively, until the sum of the second loss value and the third loss value meets the second preset condition, and the mapping network is obtained.

[0141] In some optional implementations, the preset threshold is calculated by a preset threshold calculation module. The preset threshold calculation module is used to input each group of data in the positive sample data into the abnormal state detection model to obtain the sixth multidimensional feature corresponding to each group of data; calculate the distance between each sixth multidimensional feature and the single class center; and determine the preset threshold based on the distance corresponding to each sixth multidimensional feature so that the proportion of distances less than the preset threshold meets the preset ratio.

[0142] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0143] In this embodiment, the abnormal state detection device for the hydro-generator set is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0144] This invention also provides a computer device having the above-described features. Figure 8 The device shown is for detecting abnormal conditions of a hydro-generator unit.

[0145] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 8As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 8 Take a processor 10 as an example.

[0146] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0147] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0148] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0149] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0150] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 8 Taking the example of a connection between China and Israel via a bus.

[0151] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0152] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0153] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0154] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for detecting abnormal state of a hydroelectric generating unit, characterized by, The method comprises: obtaining current characteristic data of a hydroelectric generating set, the characteristic data comprising unit operating condition parameters and stability monitoring parameters; inputting the characteristic data into a pre-trained abnormal state detection model to obtain mapped features of the characteristic data; calculating distances from the mapped features to single-class centers, the single-class centers being determined according to positive sample data collected when the hydroelectric generating set is in a normal operating state; if the distances are less than a preset threshold, determining that the hydroelectric generating set is in a normal state; the abnormal state detection model comprises an encoder and a mapping network, the encoder is configured to map the characteristic data to obtain initial multi-dimensional features and send the initial multi-dimensional features to the mapping network; the mapping network is configured to perform secondary mapping on the initial multi-dimensional features to obtain the mapped features; the step of training the abnormal state detection model comprises: obtaining training characteristic data of the hydroelectric generating set at different time nodes; constructing multiple sets of sample data according to the training characteristic data of the hydroelectric generating set at different time nodes; establishing positive sample data and negative sample data using the multiple sets of sample data; training an initial encoder using the positive sample data to obtain the encoder; training an initial mapping network using the encoder, the positive sample data and the negative sample data to obtain the mapping network; the step of training the initial encoder using the positive sample data to obtain the encoder comprises: keeping the unit operating condition parameters in each of the sample data unchanged, randomly setting the stability parameters in the sample data to 0 according to a preset probability to obtain enhanced sample data; mapping the enhanced sample data into first multi-dimensional features using the initial encoder; reconstructing the first multi-dimensional features into original samples using a decoder network; inputting each of the sample data and the corresponding original sample into a first loss function to obtain a first loss value; if the sum of the first loss values does not satisfy a first preset condition, optimizing parameters of the initial encoder and the decoder network, returning to the step of mapping the enhanced sample data into first multi-dimensional features using the initial encoder, until the first loss values satisfy the first preset condition, and obtaining the encoder; the step of training the initial mapping network using the encoder, the positive sample data and the negative sample data to obtain the mapping network comprises: mapping the positive sample data and the negative sample data into second multi-dimensional features and third multi-dimensional features respectively using the encoder; mapping the second multi-dimensional features and the third multi-dimensional features into fourth multi-dimensional features and fifth multi-dimensional features respectively using the initial mapping network; calculating a second loss value of the initial mapping network using distances from the fourth multi-dimensional features to the single-class centers; calculating a third loss value of the initial mapping network using the fifth multi-dimensional features, the single-class centers and a preset radius of a hypersphere of the single-class centers; If the sum of the second loss value and the third loss value does not satisfy a second preset condition, the parameters of the initial mapping network are optimized, and the step of mapping the positive sample data and the negative sample data into second multi-dimensional features and third multi-dimensional features by using the encoder is returned until the sum of the second loss value and the third loss value satisfies the second preset condition, and the mapping network is obtained.

2. The method of claim 1, wherein, The training feature data of the hydroelectric generating set at different time nodes is used to construct a plurality of sample data, including: A first two-dimensional feature matrix is constructed according to the training feature data of the hydroelectric generating set at different time nodes; The first two-dimensional feature matrix is normalized to obtain a second two-dimensional feature matrix; The second two-dimensional feature matrix is divided in an overlapping sampling manner in the time dimension to obtain a plurality of sample data.

3. The method according to claim 1 or 2, characterized in that, The step of establishing negative sample data by using the plurality of sample data includes: The original data of the stability parameter in the sample data is randomly replaced with an over-standard value while keeping the unit operating condition parameters in each sample data unchanged to obtain over-standard sample data; The over-standard sample data is normalized to obtain the negative sample data.

4. The method of claim 1, wherein, The sample data includes a plurality of data, and the step of determining the preset threshold includes: Each group of data in the positive sample data is input into the abnormal state detection model to obtain a sixth multi-dimensional feature corresponding to each group of data; The distance between each sixth multi-dimensional feature and the single-class center is calculated; The preset threshold is determined according to the distance corresponding to each sixth multi-dimensional feature, so that the proportion of the distance less than the preset threshold satisfies a preset proportion.

5. A device for detecting abnormal conditions of a hydro-generator unit, characterized in that, The device includes: A data acquisition module is configured to acquire current feature data of a hydroelectric generating set, wherein the feature data includes unit operating condition parameters and stability monitoring parameters; A mapping feature acquisition module is configured to input the feature data into a pre-trained abnormal state detection model to obtain mapping features of the feature data; A distance calculation module is configured to calculate the distance between the mapping features and a single-class center, wherein the single-class center is determined according to positive sample data collected when the hydroelectric generating set is in a normal operating state; A state detection module is configured to determine that the hydroelectric generating set is in a normal state if the distance is less than a preset threshold. The abnormal state detection model includes an encoder and a mapping network, The encoder is configured to map the feature data to obtain initial multi-dimensional features and send the initial multi-dimensional features to the mapping network; The mapping network is configured to perform secondary mapping on the initial multi-dimensional features to obtain the mapping features; The step of training the abnormal state detection model includes: Training feature data of a hydroelectric generating set at different time nodes is acquired; A plurality of sample data is constructed according to the training feature data of the hydroelectric generating set at different time nodes; Positive sample data and negative sample data are established by using the plurality of sample data; An initial encoder is trained by using the positive sample data to obtain the encoder; The initial mapping network is trained by using the encoder, the positive sample data and the negative sample data, so as to obtain the mapping network. The training of the initial encoder by using the positive sample data to obtain the encoder comprises: The stability parameters in the sample data are randomly set to 0 according to a preset probability while keeping the unit operating condition parameters in the sample data unchanged, so as to obtain enhanced sample data; The enhanced sample data is mapped into first multi-dimensional features by using the initial encoder; The first multi-dimensional features are reconstructed into original samples by using a decoder network; The sample data and the corresponding original samples are respectively input into a first loss function, so as to obtain a first loss value; If the sum of the first loss values does not satisfy a first preset condition, the parameters of the initial encoder and the decoder network are optimized, and the step of mapping the enhanced sample data into first multi-dimensional features by using the initial encoder is returned until the first loss value satisfies the first preset condition, so as to obtain the encoder. The initial mapping network is trained by using the encoder, the positive sample data and the negative sample data, so as to obtain the mapping network, which comprises: The positive sample data and the negative sample data are respectively mapped into second multi-dimensional features and third multi-dimensional features by using the encoder; The second multi-dimensional features and the third multi-dimensional features are respectively mapped into fourth multi-dimensional features and fifth multi-dimensional features by using the initial mapping network; A second loss value of the initial mapping network is calculated by using the distance between the fourth multi-dimensional features and the single-class center; A third loss value of the initial mapping network is calculated by using the fifth multi-dimensional features, the single-class center and a preset radius of the hypersphere of the single-class center; If the sum of the second loss value and the third loss value does not satisfy a second preset condition, the parameters of the initial mapping network are optimized, and the step of mapping the positive sample data and the negative sample data into second multi-dimensional features and third multi-dimensional features by using the encoder is returned until the sum of the second loss value and the third loss value satisfies the second preset condition, so as to obtain the mapping network.

6. A computer device, comprising: Comprise: A memory and a processor, which are connected in communication with each other, and the memory stores computer instructions, and the processor executes the computer instructions to perform the method for detecting abnormal state of a hydroelectric generating unit according to any one of claims 1 to 4.

Images